Project achievement evaluation system combining literature measurement and super network model

Abstract

The invention discloses a project achievement evaluation system combining literature measurement and a super network model. The system comprises a research basis acquisition module, a project element extraction module, a super network model construction module, a super network analysis and project evaluation module and the like. According to the system, all elements of science and technology projects are abstracted into network nodes, a hypernetwork model is used for carrying out correlation analysis on the project elements, and project achievements are comprehensively evaluated by combining methods such as literature metering, clustering analysis, balance score card evaluation and association rule mining. According to the method, the influence of project elements on project achievements is quantified, the accuracy and scientificity of project evaluation are improved, and the method has important significance in optimizing scientific and technological resource configuration and improving scientific and technological project management efficiency.

Description

Technical Field

[0001] The invention belongs to the field of project evaluation, and more specifically relates to a project achievement evaluation system combining bibliometrics with a hypernetwork model. Background Art

[0002] At present, the evaluation of scientific and technological project results mainly adopts the method of quantitative measurement by constructing an indicator system. Qiao Dehua believes that innovative results focus on the innovation and integration level of technical theories, key technologies, common technologies and core technologies, so four first-level indicators including innovation value, output of results, innovation process and benefit of results are used to evaluate scientific and technological achievements. Zhu Shiyun et al. constructed an indicator system with three dimensions, examining the quality of scientific research results from the perspective of "scientific and technological achievement evaluation", the subsequent promotion of scientific research results from the perspective of "evaluation of results transformation process", and the actual production application from the perspective of "evaluation of transformation effect". Gao Chao and Wang Yanxia proposed a five-element value collaborative evaluation model for scientific and technological achievements, dividing scientific and technological achievements into four categories: basic research, applied research, technology development and industrialization, and soft science, and conducting scientific and technological achievement evaluation from five aspects: scientific value, technical value, social value, economic value and cultural value of scientific and technological achievements. Xiang Wang et al. constructed a diversified scientific and technological achievement evaluation indicator system, including 7 first-level indicators and 23 second-level indicators including degree of independent innovation, technological advancement, technical difficulty and complexity, promotion and application prospects, economic benefits, social benefits and achievement indicators. Guo Wensheng et al. constructed a hierarchical structure of evaluation indicators and proposed for the first time a model of multi-level classification to construct a dynamic set of meta-indicators, including economic benefits, social benefits, ecological benefits, market demand, market risks, and policy risks. The market risk indicator not only takes into account whether there is market demand, but also the number of competitive achievements, the scale ratio of competitive achievements, the industry competition structure, the substitutability of achievements, and downstream supporting channels.

[0003] At this stage, high-level scientific and technological self-reliance has put forward higher mission requirements for the implementation of scientific and technological projects. The use of traditional bibliometric quantitative methods for project achievement evaluation cannot meet the new requirements for carrying out project achievement evaluation. Some scholars have constructed a similarity calculation model based on the BERT architecture, and explored the correlation evaluation of scientific and technological project-achievement documents through the degree of text semantic matching. Some scholars also believe that the performance evaluation of scientific research projects needs to integrate intelligent technology in all links, collect massive data from multiple sources, multiple dimensions, multiple granularities, and multiple types in real time, and use natural language processing, machine learning and other technologies to mine and establish the relationship between factual data and performance. These studies provide us with useful thinking for studying project achievement evaluation methods from the perspective of network structure. The overall idea of ​​the present invention to evaluate the achievements of major scientific and technological projects is to comprehensively consider traditional bibliometric methods and network structures.

[0004] However, a single-level network cannot meet all the needs of problem research. A more comprehensive complex network structure is urgently needed to characterize the multivariate relationships between different elements of the project. Nagurney and Dong called the network that is higher and beyond the existing network a super network. It has heterogeneity, multi-attribute, multi-layer, multi-level and multi-dimensionality, which meets the diversified needs of project element abstraction. At present, the super network research based on system science analyzes the properties of the network's own structure, such as centrality, agglomeration and robustness, and has been applied to knowledge discovery, knowledge management, organizational cooperation and public opinion evolution.

[0005] In short, the research on project outcome evaluation methods at home and abroad has achieved relatively rich results, which has laid a good theoretical foundation for this invention. However, the existing literature rarely mentions project evaluation from the perspective of network structure, and there is also a lack of research on project evaluation methods that comprehensively considers quantitative evaluation of outcomes and network structure. Summary of the invention

[0006] This paper takes major scientific and technological projects as the research object, carries out qualitative sorting and hypernetwork modeling research on project achievements and their related elements, verifies the model through Matlab simulation analysis, and then constructs a project achievement evaluation method combining bibliometrics and hypernetwork analysis, specifically quantifies the impact of each project element on project achievements, and finally provides theoretical and empirical support for improving the evaluation method of major scientific and technological project achievements in my country and optimizing the allocation of scientific and technological resources.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the system comprises:

[0008] Research basic collection module: automatically collects research project data information, saving time and effort for manual collection;

[0009] Project element extraction module: quickly and accurately extract keywords or phrases from a large amount of text data as project elements;

[0010] Build a supernet model module: abstract project elements into network nodes, use vector space model to perform word segmentation and term weight calculation, and use Jaccard index to analyze the relevance of terms;

[0011] Hypernetwork analysis: Through community division, centrality analysis, and network path analysis, the hypernetwork analysis method takes into account the interactive relationship between project elements, thereby obtaining a more comprehensive and in-depth evaluation result;

[0012] Project evaluation module: Apply the hyperedge similarity algorithm for cluster analysis and the Balanced Scorecard evaluation factor measurement method, and combine text analysis and statistical analysis to combine qualitative and quantitative evaluation; through the association rule mining algorithm FP-tree frequency set algorithm, explore the hidden relationship between project elements to improve the accuracy and scientificity of project evaluation.

[0013] In one embodiment, the research basic acquisition module includes:

[0014] Identify target websites and information: Obtain relevant text data for the research project from academic paper databases, news websites, patent websites, and government gazettes.

[0015] Based on Python and other programming languages, use BeautifulSoup and Scrapy crawler frameworks to write crawler scripts, including the URL of the target web page, crawling strategies, and how to parse and extract information from the crawled web page data;

[0016] Data cleaning: The information captured from the web usually contains a lot of irrelevant information and noise, so the data is cleaned and stored in the database.

[0017] In one embodiment, the project element extraction module includes:

[0018] Text preprocessing: remove stop words and special characters, perform case conversion and stem extraction;

[0019] Text segmentation: Segment text data according to word boundaries and semantics, dividing the entire text into independent words;

[0020] Feature extraction: Use the TF-IDF method to extract features from the text data after word segmentation;

[0021] Model training and feature extraction: Use the extracted features to train the machine learning model.

[0022] Select deep learning support vector machine (SVM) for model training; the trained model will be used to identify and extract keywords or phrases in the text as project elements.

[0023] In one embodiment, the supernet model building module includes:

[0024] Element nodeization: transform project elements into nodes in the network;

[0025] Establish node connections: Establish connections between nodes and formalize these relationships by constructing a matrix; in the matrix, each row and column represents a node. If there is a connection between two nodes, the value in the corresponding cell is 1, otherwise it is 0. The mathematical representation is as follows:

[0026] Let N = {n1, n2, ..., nn} be the node set. If there is a connection between ni and nj, then a[i][j] = 1 in the matrix A, otherwise a[i][j] = 0.

[0027] Constructing a hypernetwork model: Using the adjacency matrix A, a hypernetwork model is formally constructed;

[0028] Vector space model and weight calculation: In the vector space model, each node is represented by a vector, each dimension of the vector corresponds to an attribute, and a TF-IDF value is assigned to each dimension; if n terms are {t1, t2, ..., tn}, the TF-IDF weight vector of text d is V(d) = (w(t1), w(t2), ..., w(tn)), where:

[0029] w(ti)=tf(ti,d)*log(N / df(ti)),

[0030] df(ti) represents the number of texts containing term ti, N represents the total number of texts, and tf(ti,d) represents the number of occurrences of term ti in text d;

[0031] Correlation analysis: When calculating the correlation between terms (nodes), cosine similarity is used to calculate the similarity of two vectors (i.e., two nodes). The mathematical formula is: Cos(A,B)=(AB) / (||A||*||B||), where A and B are the vectors corresponding to the two nodes, * represents the dot product of the vectors, and ||.|| represents the modulus of the vector.

[0032] In one embodiment, the hypernetwork analysis comprises:

[0033] Community division: Community division is to divide project elements (represented as nodes in the network) into multiple communities according to their interaction relationships (represented as edges in the network); during the evaluation process, a community is regarded as a unit and the evaluation is analyzed based on the community as a unit;

[0034] Centrality analysis: The interaction between project elements is obviously different. Centrality analysis mainly includes degree centrality, closeness centrality, and betweenness centrality.

[0035] Degree centrality: For a node in a network, its degree is the number of nodes directly connected to it; for a directed network, the degree can be divided into in-degree and out-degree; degree centrality is the normalization of the degree, expressed as Ci = D(i) / (N-1), where D(i) represents the degree of node i, and N represents the total number of nodes in the network;

[0036] Closeness centrality: The closeness centrality of a node measures the inverse of the average shortest path from it to all other nodes; the formula is CC(i) = 1 / (sum(d(i,j))), where d(i,j) represents the shortest path from node i to node j, and sum represents the sum of all j.

[0037] Betweenness centrality: The betweenness centrality of a node is defined as the proportion of all shortest paths that pass through the node; it indicates the importance of the node in the network. The formula is CB(i) = sum(sij(i) / sij), where sij(i) represents the number of paths that pass through node i in the shortest path from node j to node k, sij represents the number of all shortest paths from node j to node k, and sum represents the sum of all j≠k≠i;

[0038] Network path analysis: Network path analysis is to study the path of information transmission in the network, so as to explore the main factors affecting the project, including shortest path analysis, full path analysis, and path dependency analysis.

[0039] In one embodiment, the project assessment module includes:

[0040] Cluster analysis: Use the hyperedge similarity algorithm to cluster project elements; used for cluster analysis of project elements, clustering similar links (project elements with the same attributes) together;

[0041] Balanced Scorecard evaluation: Use the Balanced Scorecard to conduct a comprehensive evaluation of the project. This method uses objective and scientific metrics to measure and manage the performance of the organization. The evaluation is divided into four dimensions: financial perspective, customer perspective, internal process perspective, and learning and growth perspective, which help determine which factors will affect the project's financial performance, customer satisfaction, internal process efficiency, and organizational learning and growth;

[0042] Combine text analysis and statistical analysis: Analyze text data through machine learning to extract valuable information; use hypothesis testing methods to process and analyze data to help determine the relationship between various factors and project results;

[0043] Association rule mining: Use the FP-tree frequency set algorithm to mine project elements and discover certain rules in project elements;

[0044] The FP-tree frequent item set algorithm is a tree structure, one of whose nodes represents an item set. Frequent item sets can be found through the FP-tree, while avoiding the candidate set generation process in the Apriori algorithm and improving the mining efficiency.

[0045] In one embodiment, the association rule mining includes:

[0046] 1) Construct FP-tree:

[0047] Calculate the frequency of item sets: first scan the database (that is, the set of project elements) to find all frequent 1-item sets; sort these items in descending order of support;

[0048] Construct FP tree: Next, for each transaction in the database, for example, for each record of project element data, first sort each record according to the support of its project element, and then insert the records one by one into the FP tree according to the sorting results; during the insertion process, if the prefix of a record already exists in the FP tree, a new path is added, and the counters of all nodes on the path are increased by one. If it does not exist, a new node is inserted at the corresponding place.

[0049] The final result is the FP tree;

[0050] 2) Mining frequent itemsets from FP-tree:

[0051] Starting from the bottom of the FP tree, for each frequent item, find the path in the FP tree that contains the frequent item, form a conditional pattern base, and construct a conditional FP tree;

[0052] For each conditional FP tree, repeat the above steps until the tree becomes very small or cannot be divided any further, and find all frequent item sets;

[0053] Calculation of support of itemsets. For an itemset X, its support is defined as: Where D is the total number of transactions (in this scenario, it refers to the total number of projects).

[0054] Beneficial effects of the present invention:

[0055] The system can automatically collect and clean up relevant data information of research projects, greatly saving the time and energy of manual data collection and processing.

[0056] The project element extraction module can quickly and accurately extract keywords or phrases as project elements from large amounts of text data, which provides an accurate data basis for subsequent evaluation and analysis.

[0057] The system uses a hypernetwork model to abstract project elements and their relationships, facilitating in-depth analysis of project elements and comprehensive evaluation of project outcomes.

[0058] Through the hypernetwork analysis method, the system can consider the interactive relationship between project elements and better conduct a comprehensive and in-depth evaluation of project results.

[0059] By using the project evaluation module, this system can explore the hidden relationships between project elements through association rule mining algorithms, thereby improving the accuracy and scientificity of the evaluation.

[0060] This system combines qualitative and quantitative evaluation, making the evaluation results more scientific and objective, and providing decision makers with a more comprehensive and accurate basis for decision making. The application of this system is conducive to the optimal allocation of scientific and technological resources and promotes the smooth progress of scientific and technological projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a system block diagram of the present invention;

[0062] Figure 2 It is a flowchart of extracting project elements of the present invention;

[0063] Figure 3 A flow chart of constructing a supernet model for the present invention;

[0064] Figure 4 This is a flow chart of the hypernetwork analysis of the present invention;

[0065] Figure 5 This is a flow chart for evaluating the project of the present invention;

[0066] Figure 6 This is a flow chart of the FP-tree frequent item set mining algorithm of the present invention. DETAILED DESCRIPTION

[0067] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0068] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0069] like Figure 1 As shown, the project achievement evaluation system combining bibliometrics with a hypernetwork model includes:

[0070] Research basic collection module: When implementing research, you can use web crawling technology such as crawler tools to automatically collect the required research project data information, thereby saving time and energy for manual collection.

[0071] Determine target websites and information: clarify the needs of the research project, determine the type and scope of data and information to be collected, and the target websites, etc. For example, obtain relevant text data for the research project from academic paper databases, news websites, patent websites, government gazettes, etc.

[0072] Write and run crawlers: Based on programming languages ​​such as Python, use crawler frameworks such as BeautifulSoup and Scrapy to write crawler scripts. The crawler script mainly includes the URL of the target web page, the crawling strategy, and how to parse and extract the required information from the crawled web page data.

[0073] Specifically, the crawler first sends a request to the target website to obtain web page data, and then parses the obtained web page data to extract necessary information based on established tags, attributes, etc.

[0074] Data cleaning: Information captured from web pages usually contains a lot of irrelevant information and noise, so data cleaning is required. Common data cleaning operations include removing HTML tags, special symbols, blank characters, text standardization (such as case conversion, stem extraction, etc.), etc.

[0075] Data storage: Store the cleaned data in a suitable place, such as a database or CSV, Excel, etc. These data will be used for subsequent data analysis and project evaluation.

[0076] like Figure 2 As shown in the figure, the project element extraction module: using the TF-IDF method and machine learning models such as deep learning or support vector machines for text mining is very innovative. They can quickly and accurately extract keywords or phrases from large amounts of text data as project elements.

[0077] Text preprocessing: Process the text data involved, such as removing stop words (and, the, is, etc.), special characters, performing case conversion and stem extraction, etc., so that the text data can be segmented and feature extracted.

[0078] Text segmentation: Segment text data according to word boundaries and semantics, dividing the entire text into independent words (tokens), which is the basis for text mining.

[0079] Feature extraction: Use the TF-IDF method to extract features from the text data after word segmentation. TF-IDF is the abbreviation of TermFrequency-Inverse Document Frequency, which is used to evaluate the importance of a word to a document set or one of the documents in a corpus.

[0080] The mathematical formula is: TF-IDF = TF*IDF, where:

[0081] TF (term frequency): The number of times a given word appears in a document. It is usually normalized (usually the word frequency divided by the total number of words in the article) to prevent it from being biased towards long documents. (The formula is: TF(t) = (the number of times the word appears in the article) / (the total number of words in the article))

[0082] IDF (inverse document frequency, IDF): A measure of the general importance of a word, mainly a penalty for common words. The IDF of a particular word can be obtained by dividing the total number of documents by the number of documents containing the word, and then taking the logarithm of the quotient. (The formula is: IDF(t) = log_e(total number of documents / number of documents containing the word + 1)).

[0083] Model training and feature extraction: Use the extracted features to train the machine learning model, such as support vector machine (SVM). The trained model will be used to identify and extract keywords or phrases in the text as project elements.

[0084] like Figure 3 As shown in the figure, a supernet model module is constructed: the project elements are abstracted into network nodes, the vector space model is used for word segmentation and term weight calculation, and the relevance analysis of terms is performed through the Jaccard index.

[0085] Element nodeization: It mainly transforms project elements into nodes in the network. This process is mainly a symbolic process, without a specific mathematical formula. For example, you can make "researcher A" node 1, "research topic B" node 2, etc.

[0086] Establish node connections: Establish connections between nodes and formalize these relationships by constructing a matrix. In the matrix, each row and column represents a node. If there is a connection between two nodes, the value in the corresponding cell is 1, otherwise it is 0. This matrix is ​​usually called an adjacency matrix, and the mathematical representation is as follows:

[0087] Let N = {n1, n2, ..., nn} be the node set. If there is a connection between ni and nj, then a[i][j] = 1 in the matrix A, otherwise a[i][j] = 0.

[0088] Constructing a hypernetwork model: Using the adjacency matrix A, we can formally construct a hypernetwork model. This step is more about data expression and visualization.

[0089] Vector space model and weight calculation: In the vector space model, each node is represented by a vector, each dimension of the vector corresponds to an attribute, and each dimension is assigned a TF-IDF value. If n terms are {t1, t2, ..., tn}, the TF-IDF weight vector of text d is V(d) = (w(t1), w(t2), ..., w(tn)), where:

[0090] w(ti)=tf(ti,d)*log(N / df(ti)),

[0091] df(ti) represents the number of texts containing term ti, N represents the total number of texts, and tf(ti,d) represents the number of times term ti appears in text d.

[0092] Correlation analysis: When calculating the correlation between terms (nodes), the Jaccard index can be used to calculate the similarity of the set, and the mathematical definition is as above. The cosine similarity can also be used to calculate the similarity of two vectors (i.e. two nodes), and its mathematical formula is: Cos(A,B) = (AB) / (||A||*||B||), where A and B are the vectors corresponding to the two nodes, * represents the vector dot product, and ||.|| represents the modulus of the vector.

[0093] like Figure 4 As shown, hypernetwork analysis: through community division, centrality analysis, and network path analysis hypernetwork analysis methods take into account the interactive relationship between project elements to obtain more comprehensive and in-depth evaluation results.

[0094] Community division: Community division is to divide project elements (represented as nodes in the network) into multiple communities according to their interaction relationships (represented as edges in the network). In the evaluation process, a community can be regarded as a unit, and the evaluation is analyzed in units of communities. Specific community divisions include Louvain community division, Girvan-Newman community division, and Modularity community division.

[0095] Centrality analysis: The interaction between project elements is obviously different. Some elements may be core elements with more interactions, while some elements may be more peripheral. Centrality analysis is to quantify this difference through certain algorithms and mathematical methods to find the real core elements. Centrality analysis mainly includes degree centrality, closeness centrality, betweenness centrality, etc.

[0096] Specifically:

[0097] Degree centrality: For a node in a network, its degree is the number of nodes directly connected to it. For a directed network, the degree can be divided into in-degree and out-degree. Degree centrality is the normalization of the degree, expressed as Ci = D(i) / (N-1), where D(i) represents the degree of node i, and N represents the total number of nodes in the network.

[0098] Closeness centrality: The closeness centrality of a node measures the inverse of the average shortest path from it to all other nodes. The formula is CC(i) = 1 / (sum(d(i,j))), where d(i,j) represents the shortest path from node i to node j, and sum represents the sum of all j.

[0099] Betweenness centrality: The betweenness centrality of a node is defined as the proportion of all shortest paths that pass through the node. It indicates the importance of the node in the network. The formula is CB(i) = sum(sij(i) / sij), where sij(i) represents the number of paths that pass through node i in the shortest path from node j to node k, sij represents the number of all shortest paths from node j to node k, and sum represents the sum of all j≠k≠i.

[0100] Network path analysis: Network path analysis mainly studies the path of information transmission in the network, so as to find out the main factors affecting the project, which is also one of the important contents of network analysis. It includes shortest path analysis, full path analysis, path dependency analysis, etc.

[0101] like Figure 5 As shown in the figure, the project evaluation module: applies the hyperedge similarity algorithm for cluster analysis and the BalancedScorecard evaluation factor measurement method, and combines text analysis and statistical analysis to combine qualitative and quantitative evaluation; through the association rule mining algorithm FP-tree frequency set algorithm, explores the hidden relationship between project elements to improve the accuracy and scientificity of project evaluation.

[0102] Cluster analysis: Use the hyperedge similarity algorithm to cluster project elements. This algorithm is a method for similarity calculation based on hyperedge attributes. The calculation results can be used for cluster analysis of project elements, clustering similar links (project elements with the same attributes) together. The specific similarity calculation formula varies depending on the attributes and architecture of the hyperedge.

[0103] Balanced Scorecard evaluation: Use the Balanced Scorecard to conduct a comprehensive evaluation of the project. This method uses objective and scientific metrics to measure and manage the performance of the organization. The evaluation is mainly divided into four dimensions: financial perspective, customer perspective, internal process perspective, and learning and growth perspective, which help determine which factors may affect the project's financial performance, customer satisfaction, internal process efficiency, and organizational learning and growth.

[0104] Combine text analysis and statistical analysis: Analyze text data through machine learning to extract valuable information; use statistical methods to process and analyze data, such as hypothesis testing, correlation analysis, regression analysis, etc., which can help determine the relationship between various factors and project results.

[0105] Association rule mining: Use the association rule mining algorithm FP-tree frequency set algorithm to mine project elements, so that certain rules existing in project elements can be discovered, making the relationship between project elements clearer. FP-tree frequency set algorithm is an improved version of Apriori algorithm, which can more efficiently discover frequent item sets.

[0106] The FP-tree frequent set algorithm is a tree structure, one of whose nodes represents an item set. The FP-tree can be used to find frequent item sets, while avoiding the candidate set generation process in the Apriori algorithm, greatly improving the mining efficiency. For project evaluation, the FP-tree algorithm can quickly find highly correlated elements, based on which a deeper analysis can be conducted.

[0107] like Figure 6 As shown in FIG. 1 , the FP-tree frequent item set mining algorithm (Frequent Pattern-tree) can be mainly divided into two steps: constructing an FP-tree (Frequent Pattern-Tree) and mining frequent item sets from the FP-tree.

[0108] 1) Construct FP-tree:

[0109] Calculate the frequency of item sets: First scan the database (that is, the set of item elements) to find all frequent 1-item sets. Sort these items in descending order of support. This process is usually completed automatically by the program during the data preprocessing stage.

[0110] Constructing FP tree: Next, for each transaction in the database (for example, for each record of project element data), first sort each record according to the support of its project element, and then insert the records one by one into the FP tree according to the sorting results. During the insertion process, if the prefix of a record already exists in the FP tree, a new path is added, and the counters of all nodes on the path are increased by one. If it does not exist, a new node is inserted at the corresponding place. The final result is the FP tree.

[0111] 2) Mining frequent itemsets from FP-tree:

[0112] Starting from the bottom of the FP tree, for each frequent item, find the path (prefixpath) in the FP tree that contains the frequent item, form a conditional pattern base, and construct a conditional FP tree.

[0113] For each conditional FP tree, repeat the above steps until the tree becomes very small or cannot be divided any further, thus finding all frequent item sets.

[0114] The mathematical formula involved is mainly the calculation of the support of the itemset. For a certain itemset X, its support can be defined as: Where D is the total number of transactions (in this scenario, it refers to the total number of projects).

[0115] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0116] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A project achievement evaluation system combining bibliometrics and hypernetwork model, characterized by: The system comprises: Research basic collection module: automatically collects research project data information, saving time and effort for manual collection; Project element extraction module: quickly and accurately extract keywords or phrases from a large amount of text data as project elements; Build a supernet model module: abstract project elements into network nodes, use vector space model to perform word segmentation and term weight calculation, and use Jaccard index to analyze the relevance of terms; Hypernetwork analysis: Through community division, centrality analysis, and network path analysis, the hypernetwork analysis method takes into account the interactive relationship between project elements, thereby obtaining a more comprehensive and in-depth evaluation result; Project evaluation module: Apply the hyperedge similarity algorithm for cluster analysis and the Balanced Scorecard evaluation factor measurement method, and combine text analysis and statistical analysis to combine qualitative and quantitative evaluation; through the association rule mining algorithm FP-tree frequency set algorithm, explore the hidden relationship between project elements to improve the accuracy and scientificity of project evaluation.

2. According to claim 1, a project achievement evaluation system combining bibliometrics and hypernetwork model is characterized by: The research basic acquisition module includes: Identify target websites and information: Obtain relevant text data for the research project from academic paper databases, news websites, patent websites, and government gazettes; Based on Python programming language, use BeautifulSoup and Scrapy crawler framework to write crawler scripts, including the URL of the target web page, crawling strategy, and how to parse and extract information from the crawled web page data; Data cleaning: The information captured from the web usually contains a lot of irrelevant information and noise, so the data is cleaned and stored in the database.

3. According to claim 1, a project achievement evaluation system combining bibliometrics and hypernetwork model is characterized by: The project element extraction module includes: Text preprocessing: remove stop words and special characters, perform case conversion and stem extraction; Text segmentation: Segment text data according to word boundaries and semantics, dividing the entire text into independent words; Feature extraction: Use the TF-IDF method to extract features from the text data after word segmentation; Model training and feature extraction: Use the extracted features to train the machine learning model. Select deep learning support vector machine (SVM) for model training; the trained model will be used to identify and extract keywords or phrases in the text as project elements.

4. According to claim 1, a project achievement evaluation system combining bibliometrics and hypernetwork model is characterized by: The supernet model building module includes: Element nodeization: transform project elements into nodes in the network; Establish node connections: Establish connections between nodes and formalize these relationships by constructing a matrix; in the matrix, each row and column represents a node. If there is a connection between two nodes, the value in the corresponding cell is 1, otherwise it is 0. The mathematical representation is as follows: Let N = {n1, n2, ..., nn} be the node set. If there is a connection between ni and nj, then a[i][j] = 1 in the matrix A, otherwise a[i][j] = 0. Constructing a hypernetwork model: Using the adjacency matrix A, a hypernetwork model is formally constructed; Vector space model and weight calculation: In the vector space model, each node is represented by a vector, each dimension of the vector corresponds to an attribute, and a TF-IDF value is assigned to each dimension; if n terms are {t1, t2, ..., tn}, the TF-IDF weight vector of text d is V(d) = (w(t1), w(t2), ..., w(tn)), where: w(ti)=tf(ti,d)*log(N / df(ti)), df(ti) represents the number of texts containing term ti, N represents the total number of texts, and tf(ti,d) represents the number of occurrences of term ti in text d; Correlation analysis: When calculating the correlation between terms (nodes), cosine similarity is used to calculate the similarity of two vectors (i.e., two nodes). The mathematical formula is: Cos(A,B)=(AB) / (||A||*||B||), where A and B are the vectors corresponding to the two nodes, * represents the dot product of the vectors, and ||.|| represents the modulus of the vector.

5. According to claim 1, a project achievement evaluation system combining bibliometrics and hypernetwork model is characterized by: The hypernetwork analysis includes: Community division: Community division is to divide project elements (represented as nodes in the network) into multiple communities according to their interaction relationships (represented as edges in the network); during the evaluation process, a community is regarded as a unit and the evaluation is analyzed based on the community as a unit; Centrality analysis: The interaction between project elements is obviously different. Centrality analysis mainly includes degree centrality, closeness centrality, and betweenness centrality. Degree centrality: For a node in a network, its degree is the number of nodes directly connected to it; for a directed network, the degree can be divided into in-degree and out-degree; degree centrality is the normalization of the degree, expressed as Ci = D(i) / (N-1), where D(i) represents the degree of node i, and N represents the total number of nodes in the network; Closeness centrality: The closeness centrality of a node measures the inverse of the average shortest path from it to all other nodes; the formula is CC(i) = 1 / (sum(d(i,j))), where d(i,j) represents the shortest path from node i to node j, and sum represents the sum of all j. Betweenness centrality: The betweenness centrality of a node is defined as the proportion of all shortest paths that pass through the node; it indicates the importance of the node in the network. The formula is CB(i) = sum(sij(i) / sij), where sij(i) represents the number of paths that pass through node i in the shortest path from node j to node k, sij represents the number of all shortest paths from node j to node k, and sum represents the sum of all j≠k≠i; Network path analysis: Network path analysis is to study the path of information transmission in the network, so as to explore the main factors affecting the project, including shortest path analysis, full path analysis, and path dependency analysis.

6. According to claim 1, a project achievement evaluation system combining bibliometrics and hypernetwork model is characterized by: The project evaluation module includes: Cluster analysis: Use the hyperedge similarity algorithm to cluster project elements; used for cluster analysis of project elements, clustering similar links (project elements with the same attributes) together; Balanced Scorecard evaluation: Use the Balanced Scorecard to conduct a comprehensive evaluation of the project. This method uses objective and scientific metrics to measure and manage the performance of the organization. The evaluation is divided into four dimensions: financial perspective, customer perspective, internal process perspective, and learning and growth perspective, which help determine which factors will affect the project's financial performance, customer satisfaction, internal process efficiency, and organizational learning and growth; Combine text analysis and statistical analysis: Analyze text data through machine learning to extract valuable information; use hypothesis testing methods to process and analyze data to help determine the relationship between various factors and project results; Association rule mining: Use the FP-tree frequency set algorithm to mine project elements and discover certain rules in project elements; The FP-tree frequent item set algorithm is a tree structure, one of whose nodes represents an item set. Frequent item sets can be found through the FP-tree, while avoiding the candidate set generation process in the Apriori algorithm and improving the mining efficiency.

7. The project achievement evaluation system combining bibliometrics and hypernetwork model according to claim 6 is characterized by: The association rule mining includes: 1) Construct FP-tree: Calculate the frequency of item sets: first scan the database (that is, the set of project elements) to find all frequent 1-item sets; sort these items in descending order of support; Construct FP tree: Next, for each transaction in the database, for example, for each record of project element data, first sort each record according to the support of its project element, and then insert the records one by one into the FP tree according to the sorting result; during the insertion process, if the prefix of a record already exists in the FP tree, a new path is added, and the counters of all nodes on the path are increased by one. If it does not exist, a new node is inserted at the corresponding place. The final result is the FP tree; 2) Mining frequent itemsets from FP-tree: Starting from the bottom of the FP tree, for each frequent item, find the path in the FP tree that contains the frequent item, form a conditional pattern base, and construct a conditional FP tree; For each conditional FP tree, repeat the above steps until the tree becomes very small or cannot be divided any further, and find all frequent item sets; Calculation of support of itemsets. For an itemset X, its support is defined as: Where D is the total number of transactions (in this scenario, it refers to the total number of projects).

Images