Technology service resource data processing method, system and computer-readable storage medium based on optimization algorithm

Abstract

The present invention provides a text data processing method based on an intelligent optimization algorithm. This data processing method can perform data extraction, data preprocessing, feature selection, and feature analysis on the text data of scientific and technological resources, and finally obtain a feature vector and a thesaurus data with the highest accuracy and fewest feature quantities, effectively improving the efficiency and performance of running a classifier. It also relates to a corresponding system, including: a data extraction module that extracts various service data from a scientific and technological service resource database and adds classification labels according to the industry; a data preprocessing module that cleans the original data, processes to obtain a feature vector and a feature thesaurus, and divides the training and test data sets according to cross-validation; a feature selection module that uses a binary banyan tree growth optimization algorithm for feature selection to obtain a solution set with the optimal classification accuracy and feature quantity; a feature analysis module that obtains the optimal feature words from the feature words of the initial text feature vector and displays them in a feature heat map and a word cloud. The finally processed data realizes the conversion of the text data of scientific and technological service resources into an optimal feature vector that is conducive to the processing of the classifier.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data, and particularly relates to the field of science and technology service resource data processing. Specifically, it relates to a method, a system and a computer-readable storage medium for processing science and technology service resource data based on an optimization algorithm. Background Art

[0002] As an integral part of modern service industries, the science and technology service industry has become an essential link in driving economic growth and scientific and technological progress. Science and technology resources are the collection of various production factors for science and technology services and technological innovation, and their quantity and quality are important factors affecting and restricting scientific and technological activities. Different from traditional commodity service resources, science and technology services involve both supply and demand sides, including various small and medium-sized enterprises, governments, major universities and research institutions that need to carry out science and technology services such as development and research services, inspection and testing, etc. In the processing of science and technology service resource data, a large number of science and technology service resources emerge. These resources include resource types such as talent pools, technological achievements, intellectual property rights, instrument facilities, policies, etc. Their categories are diverse and complex, and the effective processing of these resource data is the basis for providing science and technology services.

[0003] Science and technology service resources are composed of semi-structured or unstructured text data. The automated integration of such texts involves multiple technologies such as natural language processing, data mining, machine learning, and information retrieval. The processing of science and technology resources requires the use of feature extraction (Feature Selection, FS) and text mining algorithms to mine potential features from a large amount of science and technology texts and classify them according to text features and types. In engineering applications, text, as an important data type, contains a large amount of feature information, and the aggregation of science and technology service resources requires a text feature selection algorithm. Different from traditional feature selection algorithms, intelligent optimization algorithms do not require domain knowledge and do not make any assumptions about the search space. Feature selection can be considered as the process of selecting the best subset of candidate features to enable a machine learning model to achieve superior performance. Therefore, intelligent optimization algorithms can be widely used to solve optimization problems and have been applied to solve feature selection in the field of text mining.

[0004] In recent years, to break through the limitations of classical optimization algorithms that require domain knowledge and spatial assumptions, a variety of intelligent optimization algorithms inspired by natural phenomena have emerged, such as genetic algorithms, particle swarm optimization, differential evolution algorithms, etc. For a specific dataset, its FS problem can be regarded as a process of selecting the best candidate subset to enable a machine learning model to achieve superior performance. Therefore, intelligent optimization algorithms can be applied to solve FS in the field of text mining, effectively excluding the influence of irrelevant and redundant features, and having better accuracy than traditional feature selection methods. In the actual text feature selection problem of scientific and technological service resources, due to complex samples, large amounts of data, and a highly sparse feature matrix, common optimization algorithms have insufficient selection efficiency in this problem and cannot efficiently process similarities or class memberships between texts. Therefore, it is necessary to improve the text document feature processing method for scientific and technological service resources.

[0005] Currently, for the data processing method of scientific and technological service resources, Chinese Patent Application No. "CN109447266B" discloses "An intelligent sorting method for agricultural scientific and technological services based on big data"; Chinese Patent Application No. "CN201710181572.8" discloses "A text classification feature selection method"; Chinese Patent Application No. "CN201810315024.4" discloses "A text feature selection method optimized by particle swarm optimization"; Chinese Patent Application No. "CN201410795989.X" discloses "An improved text classification feature selection method"; Chinese Patent Application No. "CN202010313160.7" discloses "A scientific and technological resource data service system"; Chinese Patent Application No. "CN201710889732.4" discloses "An integrated service system for enterprise scientific and technological resources". In the disclosed inventions or literatures, no instances of data processing methods for scientific and technological service resources using the optimization algorithm based on the present invention are mentioned. Summary of the Invention

[0006] The present invention provides a text data processing method based on an intelligent optimization algorithm and applies it to the data processing problem of scientific and technological service resources. This data processing method can perform data extraction, data preprocessing, feature selection, and feature analysis on the text data of scientific and technological resources, and finally obtain a feature vector and vocabulary data with the highest accuracy and the fewest features, improving the efficiency and performance of running a classifier. To this end, the technical solutions adopted are as follows:

[0007] The data processing method of scientific and technological service resources based on an optimization algorithm is mainly characterized in that the method includes the following steps:

[0008] (1) Extract data from various resource libraries in the scientific and technological service database to be used as the original dataset for subsequent feature extraction and analysis, and add classification labels to the corresponding data according to the industries in the scientific and technological service resource library;

[0009] (2) Clean, perform Chinese word segmentation, remove stop words, and perform feature vectorization on the extracted dataset to obtain the initial text feature vector and the feature word library for subsequent feature extraction and analysis, and divide the training dataset and the test dataset according to the cross-validation method;

[0010] (3) Use a feature selection method based on the Binary Banyan Tree Growth Optimization (BBTGO) algorithm to optimize the text feature vector to obtain the classification accuracy of the machine learning algorithm and the optimal solution set of the feature quantity. Among them, the optimal solution set is used to obtain the text features of science and technology services;

[0011] (4) Use the optimal solution of feature selection obtained in the above steps to obtain the optimal feature words from the feature words of the initial text feature vector and display them in the feature heat map and word cloud.

[0012] Preferably, in step (1), the metadata of each resource library in the science and technology service database includes talent teams, technical products, intellectual property rights, instrumentation, and policy support, and the metadata of each resource library is classified according to each industry.

[0013] Preferably, in step (2), according to the original dataset provided in step (1), clean the data with missing resource description information or non-compliant data, apply the TF-IDF method for word vectorization processing and screen out vectors with too low word segmentation frequencies, and apply the cross-validation method to divide the training dataset and the test dataset.

[0014] Preferably, step (2) specifically includes the following steps:

[0015] (2.1) Clean the constructed original dataset, and delete the data with missing resource description information or non-compliant data from the dataset;

[0016] (2.2) Perform automatic Chinese word segmentation and stop word removal on each piece of data to construct a feature set (t1, t2,..., t n ), where the feature set consists of different feature words after word segmentation. Here, n represents the total number of feature words, and calculate the Term Frequency-Inverse Document Frequency (TF-IDF) weight according to the following formula:

[0017]

[0018] Among them, t i represents the i-th feature word, tf i represents the number of times the feature word appears in this resource, df i represents the number of resources in the dataset that contain t i , and idf i represents the inverse dfi Weight; N(t i , d) represents the number of times the feature term t i appears in the resource text d, and N is the total number of resource data;

[0019] (2.3) The original feature vector T of the entire dataset and the corresponding labels are obtained through the above steps. According to the number of each type of label, the 10-fold cross-validation method is applied to divide the original feature vector into training and test datasets.

[0020] Preferably, in step (3), the feature vector processed by the above step (2) is used, and a feature selection method based on the Binary Banyan Tree Growth Optimization (BBTGO) algorithm is applied to search the feature vector space, and the optimization target weights of the classification error and the number of selected features are set to achieve the optimal error and feature quantity simultaneously.

[0021] Preferably, the step (3) specifically includes the following steps:

[0022] (3.1) Set the parameters of the initialized Binary Banyan Tree Growth Optimization (BBTGO) algorithm and the classifier parameters, specifically: set the number of banyan tree branches m, population size popSize, maximum number of iterations Gmax, growth stage threshold T growth , and the regeneration stage threshold T recove ;

[0023] (3.2) Set the optimization function of the classification error rate and the number of features based on the test set of the classification model according to the following formula:

[0024]

[0025] where ζ is the proportionality factor, represents the error rate of the test set after the classification model performs 10-fold cross-validation, |F| represents the number of features selected by the feature selection method, and n represents the length of the feature vector;

[0026] (3.3) The Binary Banyan Tree Growth Optimization (BBTGO) algorithm adopts a binary coding framework. Each branch and leaf individual represents a set of feature subsets. The bit positions of the individual are encoded by 0 and 1, representing the selected or unselected state of the feature item at that position. When initializing, the individual can be randomly initialized to 0 or 1. The i-th individual can be represented as follows:

[0027] x i = [x i,0 , x i,2 ,..., x i,d ,..., x i,n

[0028] ​(3.4) Apply the fitness function in step (3.2) to calculate the fitness value of each branch and leaf individual;

[0029] (3.5) Initialize the aerial roots, i.e., the individual historical optimal position P root , and evenly divide the population individuals into m subgroups according to the fitness sorting. The best fitness individual of each subpopulation is used as the optimal position of the branch subgroup, denoted as P trunk ;

[0030] (3.6) Calculate the best position of the individuals in all current subgroups and denote it as P best ;

[0031] (3.7) According to the iteration situation, execute different operators to update the individuals of the banyan population. Specifically:

[0032] When the number of times the individual i has not been updated, count i is lower than T growth , select and execute the aerial root operator or the multi-trunk operator with a probability of pr; when the number of times the individual i has not been updated, count i is greater than less than T growth , execute the adjustment operator to generate a new population;

[0033] When the individual probability is less than pr, execute the rooting operator. The update of the j-th position of this individual is as follows:

[0034]

[0035] When the individual probability is greater than pr, execute the multi-trunk operator. Its bit update is expressed as follows:

[0036]

[0037] Where represents the optimal position of the trunk group where the i-th individual is located;

[0038] When count i is higher than T growth and less than T recover , execute the adjustment operator. Randomly select a trunk subgroup c r , and replace a part of the bits of the i individual with the corresponding bits of the optimal position of this subgroup at a selection ratio of r. Its expression is as follows:

[0039]

[0040] When count i is higher than T recover , execute the rebirth operator to randomly initialize the bits of this individual again. Its expression is as follows:

[0041]

[0042] (3.8) Calculate the individual fitness values of the new population according to the optimization function in step (3.2);

[0043] (3.9) Re-group the individuals into the trunk subgroups according to the spatial distance, calculate the distances between the individual and the optimal positions of each trunk subgroup, and divide it into the trunk subgroup with the closest distance. The distance calculation is expressed as follows:

[0044]

[0045] (3.10) Compare the fitness values of the updated individual i and If the fitness value of the new individual is smaller, then replace the position of with the position of the new individual, and at the same time set count i to 0; otherwise, increment its value by 1. Similarly, update the positions and fitness values of population P trunk and P best ;

[0046] (3.11) Determine whether the termination condition is reached. If the maximum number of iterations is not reached, return to step (3) to continue execution; otherwise, output the optimal individual P best in the current dataset and its corresponding classifier classification accuracy.

[0047] Preferably, in step (4), retrieve the obtained optimal features, specifically: obtain the preferred features and corresponding feature words from the original feature vector and the thesaurus, display them visually in the heat map according to the feature values, and analyze according to the distribution of effective feature values.

[0048] Preferably, step (4) specifically includes the following steps:

[0049] (4.1) According to the bit positions of the optimal individual P best , extract the corresponding feature words and feature weights from the original feature thesaurus and the feature vector respectively, and form an optimized feature vector;

[0050] (4.2) Display the optimized feature vectors one by one in the heat map according to the industry labels, compare the feature distributions within each industry label and between different industries, and display the feature words with large differences between various categories in the heat map in the word cloud;

[0051] (4.3) Record the list of feature words with significant industry influence for reference in subsequent data processing, and finally obtain the feature vector and thesaurus data with the highest accuracy and the fewest feature quantities.

[0052] The technology service resource data processing system based on the optimization algorithm for implementing the above method is mainly characterized in that the system includes:

[0053] A data extraction module, which is used to extract data from various resource libraries in the science and technology service database, and add classification tags according to the industries in the resource libraries, so as to facilitate the use as the original data set for subsequent feature extraction and analysis;

[0054] A data preprocessing module, which is connected to the above-mentioned data extraction module, and is used to perform data cleaning, Chinese word segmentation, stop word removal, and feature vectorization processing on the extracted data set, so as to obtain an initial text feature vector and a feature word library, and divide the training and test data sets according to the cross-validation method, so as to facilitate subsequent feature extraction and analysis;

[0055] A feature selection module, which is connected to the above-mentioned data preprocessing module, and is used to optimize the obtained text feature vector based on the feature selection method of the binary banyan tree growth optimization (BBTGO) algorithm, so as to obtain the classification accuracy of the machine learning algorithm and the optimal solution set of the feature quantity, and thus obtain the science and technology service text features;

[0056] A feature analysis module, which is connected to the above-mentioned feature selection module, and is used to obtain the optimal feature words from the feature words of the initial text feature vector according to the obtained optimal solution of the feature selection, and display them in the feature heat map and word cloud for feature analysis.

[0057] This computer-readable storage medium is mainly characterized in that a computer program is stored thereon, and the computer program can be executed by a processor to implement each step of the above-mentioned science and technology service resource data processing method based on the optimization algorithm.

[0058] Adopting the science and technology service resource data processing method, system and its computer-readable storage medium based on the optimization algorithm of the present invention has the following beneficial effects:

[0059] (1) The present invention marks various collected science and technology service resources according to the sub-sectors, and analyzes the characteristics of various resources from the feature level, providing a strong theoretical basis for data processing.

[0060] (2) Using the new meta-heuristic method BBTGO to solve the feature selection problem of high-dimensional sparse science and technology service resources, effectively improving the search speed and quality of feature selection.

[0061] (3) Using the feature selection method based on the intelligent optimization algorithm to obtain the important feature word libraries of each industry and effectively reduce the number of features, improving the operation efficiency of the machine learning algorithm.

[0062] (4) The present invention is simple to implement and easy to expand, and this method can be conveniently extended to other text feature processing problems in the field of text data processing. Description of the Drawings

[0063] Figure 1 This is the structural diagram of the scientific and technological service resource data processing system based on the optimization algorithm of the present invention.

[0064] Figure 2 This is the flowchart of processing step 1 of the scientific and technological service resource data processing method based on the optimization algorithm of the present invention.

[0065] Figure 3 This is the flowchart of processing step 3 of the scientific and technological service resource data processing method based on the optimization algorithm of the present invention. Detailed implementation manners

[0066] To make the technical solutions and advantages of the present invention more clearly explained, the following part will be described in detail in combination with the corresponding drawings and examples. It should be noted that the specific examples described here are only used to explain the present invention and are not used to limit the present invention.

[0067] The system block diagram of the present invention is as Figure 1 shown. The following will be described in combination with the drawings and technical implementation solutions to clarify the application of the feature processing method based on the intelligent optimization algorithm in the scientific and technological service resource data processing:

[0068] A scientific and technological service resource data processing method based on the optimization algorithm specifically includes the following steps:

[0069] Step 1: S1 data extraction module extracts data from the actual scientific and technological service resource database, classifies the data according to the industries in the resource database, and uses it as the original data.

[0070] Construct a data set S for the resource texts in the databases of talent teams, technical products, intellectual property rights, instruments and meters, policy support, etc. respectively. The i-th resource (data D i ) in each data set includes the Chinese name Title i and the Chinese description information Des i of the resource, that is, D i ={Title i , Des i}. Collect data of different industry categories and assign labels Cls i . The composition example of the original data is shown in Table 1:

[0071] Table 1 Composition of original data

[0072] Serial number Resource name Resource description Category Category serial number 1 <![CDATA[Title1]]> <![CDATA[Des1]]> <![CDATA[Cls1]]> <![CDATA[id1]]> 2 <![CDATA[Title2]]> <![CDATA[Des2]]> <![CDATA[Cls2]]> <![CDATA[id2]]> ... ... ... ... ... i <![CDATA[Title i > <![CDATA[Des i > <![CDATA[Cls i > <![CDATA[id i > m <![CDATA[Title m > <![CDATA[Des m > <![CDATA[Cls m > <![CDATA[id m >

[0073] Step 2: S2 data preprocessing module performs data cleaning, Chinese automatic word segmentation, stop word removal, and feature vectorization on the extracted data set to obtain the initial text feature vector and feature word library, and divides the training set and the test set. Its program flow is as Figure 2 shown.

[0074] (2.1) Clean the original dataset constructed. If there is missing resource description information or data that does not meet the specifications, delete it from the dataset.

[0075] (2.2) Perform Chinese automatic word segmentation and stop word removal on each piece of data to construct a feature set (t1, t2,..., t n ). The feature set consists of different feature words after word segmentation, and n represents the total number of feature words. Then calculate the Term Frequency-Inverse Document Frequency (TF-IDF) weights, which are calculated as follows:

[0076]

[0077] Among them, t i represents the i-th feature word, tf i represents the number of times the feature word appears in this resource, df i represents the number of resources in the dataset that contain t i , idf i represents the inverse df i weight; N(t i , d) represents the number of times the feature word t i appears in the resource text d, and N is the total number of resource data.

[0078] (2.3) Obtain the original feature vector T of the entire dataset and the corresponding labels from the above steps. Apply the 10-fold cross-validation method according to the number of each type of label to divide the original feature vector into training and test datasets.

[0079] Step 3: S3 Feature Selection Module. Use the feature selection method based on the Binary Banyan Tree Growth Optimization (BBTGO) algorithm, including the rooting operator, multi-trunk operator, and adjustment operator to optimize the feature vector subset in Step 2.

[0080] (3.1) Initialize the parameters of the Binary Banyan Tree Growth Optimization BBTGO algorithm and the classifier parameters. Set parameters such as the number of banyan tree trunks m, population size popSize, maximum number of iterations Gmax, growth stage threshold T growth , and regeneration stage threshold T recover , etc.

[0081] (3.2) Set the optimization function based on the classification error rate and the number of features of the classification model test set, which can be expressed as follows:

[0082]

[0083] Among them, ζ is the proportionality factor, Denote the error rate of the test set after the classification model performs 10-fold cross-validation, |F| denote the number of features selected by the feature selection method, and n denote the length of the feature vector.

[0084] (3.3) BBTGO adopts a binary coding framework, and each branch and leaf individual represents a set of feature subsets. Among them, the bit positions of the individual are encoded by 0 and 1, indicating whether the feature item at that position is selected or not. When initializing, the individual can be randomly initialized to 0 or 1, and the i-th individual can be represented as follows:

[0085] x i = [x i,1 , x i,2 ,..., x i,d ,..., x i,n (3)

[0086] (3.4) Apply the fitness function in step (3.2) to calculate the fitness value of each branch and leaf individual;

[0087] (3.5) Initialize the position of the aerial roots (the historical best of the individual) P root , and evenly divide the population individuals into m subgroups according to the fitness ranking. The individual with the best fitness in each subpopulation is used as the optimal position of the branch subgroup, denoted as P trunk .

[0088] (3.6) Calculate the best position of the individuals in all current subgroups and denote it as P best .

[0089] (3.7) According to the iteration situation, execute different operators to update the individuals of the banyan tree population.

[0090] In the feature selection module, when the number of times the individual i has not been updated count i is lower than T growth , select to execute the aerial root operator or the multi-trunk operator with a probability of pr; when it is greater than less than, execute the adjustment operator to generate a new population.

[0091] When the individual probability is less than pr, execute the rooting operator, and the update of the j-th position of this individual is as follows:

[0092]

[0093] When the individual probability is greater than pr, execute the multi-trunk operator, and its bit position update is expressed as follows:

[0094]

[0095] where denotes the optimal position of the trunk group where the i-th individual is located.

[0096] When count i is higher than T growth and less than T recover execute the adjustment operator, randomly select a trunk subgroup c r , and replace some bit positions of individual i with the corresponding bits at the optimal position of this subgroup at the selection ratio of r, which is expressed as follows:

[0097]

[0098] When count i is higher than T recover execute the regeneration operator, randomly initialize the bit positions of this individual again, which is expressed as follows:

[0099]

[0100] (3.8) Calculate the individual fitness values of the new population according to Equation (2);

[0101] (3.9) Re-group individuals into trunk subgroups according to the spatial distance, calculate the distance between this individual and the optimal positions of each trunk subgroup, and divide it into the trunk subgroup with the closest distance. The distance calculation is expressed as follows:

[0102]

[0103] (3.10) Compare the fitness values of the updated individual i and , if the fitness value of the new individual is smaller, then replace the position of with the position of the new individual, and at the same time set count i to 0, otherwise increment its value by 1. Similarly, update the positions and fitness values of population P trunk and P best .

[0104] (3.11) Judge whether the termination condition is reached. If the method has not reached the maximum number of iterations, return to step 3 to continue execution, otherwise output the optimal individual P best in the current dataset and its corresponding classifier classification accuracy rate.

[0105] Step 4: S4 feature analysis module, use the optimal individual obtained above to extract feature words from the feature set, and output to the heat map to compare the feature distributions under different industry labels.

[0106] (4.1) According to the bit positions of the optimal individual P best , extract the corresponding feature words and feature weights from the original feature word library and feature vectors respectively, and form an optimized feature vector;

[0107] (4.2) Display the optimized feature vectors one by one in the heat map according to the industry labels, compare the feature distributions within each industry label and between different industries, and display the feature words with large inter-class differences in the heat map in the word cloud.

[0108] (4.3) Record the list of feature words significantly affected by the industry for subsequent reference in data processing. Finally, obtain the feature vectors and thesaurus data with the highest accuracy and the fewest feature quantities, effectively improving the efficiency and performance of the running classifier.

[0109] The technology service resource data processing system based on the optimization algorithm for implementing the above method, wherein the system includes:

[0110] A data extraction module, used to extract data from various resource libraries in the technology service database, and add classification labels according to the industries in the resource libraries, so as to facilitate the original data set for subsequent feature extraction and analysis;

[0111] A data preprocessing module, connected to the data extraction module, used to perform data cleaning, Chinese word segmentation, stop word removal, and feature vectorization on the extracted data set to obtain the initial text feature vectors and the feature thesaurus, and divide the training and test data sets according to the cross-validation method to facilitate subsequent feature extraction and analysis;

[0112] A feature selection module, connected to the data preprocessing module, used to optimize the obtained text feature vectors based on the feature selection method of the binary banyan tree growth optimization (BBTGO) algorithm to obtain the optimal solution set of the classification accuracy and the number of features of the machine learning algorithm, so as to obtain the technology service text features;

[0113] A feature analysis module, connected to the feature selection module, used to obtain the optimal feature words from the feature words of the initial text feature vectors according to the optimal solution of the obtained feature selection, and display them in the feature heat map and the word cloud for feature analysis.

[0114] The computer-readable storage medium, characterized in that a computer program is stored thereon, and the computer program can be executed by a processor to implement each step of the above-mentioned technology service resource data processing method based on the optimization algorithm.

[0115] Adopting the technology service resource data processing method, system and its computer-readable storage medium of the present invention has the following beneficial effects:

[0116] (1) The present invention marks various technology service resources collected according to the sub-sectors, analyzes the feature situations of various resources from the feature level, and provides a strong theoretical basis for data processing.

[0117] (2) Use the new meta-heuristic method BBTGO to solve the problem of feature selection for high-dimensional sparse technology service resources, effectively improving the search speed and quality of feature selection.

[0118] (3) Use the feature selection method based on intelligent optimization algorithms to obtain the important feature libraries for various industries and effectively reduce the number of features, improving the operation efficiency of machine learning algorithms.

[0119] (4) The implementation of the present invention is simple and easy to expand, and this method can be conveniently extended to other text feature processing problems in the field of text data processing.

[0120] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A method for processing scientific and technological service resource data based on an optimization algorithm, characterized in that, The method described above includes the following steps: (1) Extract data from various resource libraries in the science and technology service database to be used as the original dataset for subsequent feature extraction and analysis, and add classification labels to the corresponding data according to the industries in the science and technology service resource library; (2) Perform data cleaning, Chinese word segmentation, stop word removal, and feature vectorization on the extracted dataset to obtain the initial text feature vector and feature word library for subsequent feature extraction and analysis, and divide the training dataset and test dataset according to the cross-validation method; (3) Use a feature selection method based on the Binary Banyan Tree Growth Optimization (BBTGO) algorithm to optimize the text feature vector to obtain the classification accuracy of the machine learning algorithm and the optimal solution set of feature quantities. Among them, the optimal solution set is used to obtain the science and technology service text features; the Binary Banyan Tree Growth Optimization (BBTGO) algorithm adopts a binary coding framework, and each branch and leaf individual represents a set of feature subsets. The bit positions of the individual are encoded by 0 and 1, indicating whether the feature item at that position is selected or not; and the optimization of the feature population is realized through the rooting operator, multi-trunk operator, adjustment operator, and rebirth operator; (4) Use the optimal solution of feature selection obtained in the above steps to obtain the optimal feature words from the feature words of the initial text feature vector and display them in the feature heat map and word cloud.

2. The method for processing science and technology service resource data based on an optimization algorithm according to claim 1, wherein In the step (1), the metadata of each resource library in the science and technology service database includes talent teams, technical products, intellectual property rights, instrumentation, and policy support, and the metadata of each resource library is classified according to each industry.

3. The method for processing scientific and technological service resource data based on an optimization algorithm according to claim 2, characterized in that, In the step (2), according to the original dataset provided in the step (1), clean the data with missing resource description information or non-compliant data, apply the TF-IDF method for word vectorization processing and screen out the vectors with too low word segmentation frequencies, and apply the cross-validation method to divide the training dataset and test dataset.

4. The method for processing scientific and technological service resource data based on an optimization algorithm according to claim 3, wherein The step (2) specifically includes the following steps: (2.1) Perform data cleaning on the constructed original dataset, and delete the data with missing resource description information or non-compliant data from the dataset; (2.2) Perform Chinese automatic word segmentation and stop word removal on each piece of data to construct a feature set (t1, t2,..., t n ), where the feature set consists of different feature words after word segmentation, where n represents the total number of feature words, and calculate the term frequency-inverse document frequency (TF-IDF) weight according to the following formula: where t i represents the i-th feature word, tf i represents the number of times the feature word appears in this resource, df i represents the number of resources in the dataset that contain t i , idf i represents the inverse df i weight; N(t i , d) represents the number of times the feature word t i appears in the resource text d, and N is the total number of resource data; (2.3) Obtain the original feature vector T of the entire dataset and the corresponding labels from the above steps, and divide the original feature vector into training and test datasets according to the 10-fold cross-validation method according to the number of each type of label.

5. The method for processing scientific and technological service resource data based on an optimization algorithm according to claim 4, wherein In the step (3), use the feature vector processed in the above step (2), apply a feature selection method based on the Binary Banyan Tree Growth Optimization (BBTGO) algorithm to search the space of the feature vector, and set the optimization target weights of the classification error and the number of selected features to achieve the optimal error and feature quantity at the same time.

6. The method for processing scientific and technological service resource data based on an optimization algorithm according to claim 5, wherein The step (3) specifically includes the following steps: (3.1) Set the parameters of the initialized binary banyan tree growth optimization BBTGO algorithm and the classifier parameters, specifically: set the number of banyan tree branches m, population size popSize, maximum number of iterations Gmax, growth stage threshold T growth , regeneration stage threshold T recove ; (3.2) Set the optimization function based on the classification error rate and the number of features of the classification model test set according to the following formula: where ζ is a scaling factor, represents the error rate of the test set after the classification model performs 10-fold cross-validation, |F| represents the number of features selected by the feature selection method, and n represents the length of the feature vector; (3.3) At the initialization, the individuals can be randomly initialized to 0 or 1, and the i-th individual can be represented as follows: x i = [x i,1 , x i,2 ,..., x i,d ,..., x i,n ​ (3.4) Apply the fitness function in the step (3.2) to calculate the fitness value of each branch and leaf individual; (3.5) Initialize rooting, that is, the individual historical optimal position P root , and evenly divide the population individuals into m subgroups according to the fitness sorting. The individual with the best fitness in each subpopulation is used as the optimal position of the branch subgroup, denoted as P trunk ; (3.6) Calculate the best position of individuals in all current subgroups and denote it as P best ; (3.7) According to the iteration situation, different operators are executed to update the individuals of the banyan tree population. Specifically: When individual i does not update the count i Lower than T growth When the probability of pr is chosen to execute the rooting operator or the multi-trunk operator; when individual i does not update the number of count i Greater than or less than T growth When , the adjustment operator is executed to generate a new population; When the individual probability is less than pr, the rooting operator is executed, and the update of the j-th position of this individual is as follows: When the individual probability is greater than pr, the multi-trunk operator is executed, and its bit update is expressed as follows: Among them represents the optimal position of the trunk group where the i-th individual is located; When count i is higher than T growth and less than T recover then execute the adjustment operator, randomly select a trunk subgroup c r , and replace some bit positions of individual i with the corresponding bits of the optimal position of this subgroup at the selection ratio of r, which is expressed as follows: When count i is higher than T recover execute the regeneration operator to randomly initialize the bits of this individual again, which is expressed as follows: (3.8) Calculate the individual fitness values of the new population according to the optimization function in step (3.2); (3.9) Re-group the individuals into the trunk subgroups according to the spatial distance, calculate the distance between this individual and the optimal positions of each trunk subgroup, and divide it into the trunk subgroup with the closest distance. Its distance calculation is expressed as follows: (3.10) Compare the updated individual i and 's fitness value. If the fitness value of the new individual is smaller, then replace the position of with the position of the new individual, and at the same time set count i to 0; otherwise, increment its value by 1. Similarly, update the positions and fitness values of population P trunk and P best ; (3.11) Determine whether the termination condition is reached. If the maximum number of iterations is not reached, return to step (3) and continue to execute; otherwise, output the optimal individual P in the current dataset best and the classification accuracy of its corresponding classifier.

7. The method for processing science and technology service resource data based on an optimization algorithm according to claim 6, characterized in that, In step (4), the obtained optimal features are retrieved. Specifically: obtain the preferred features and corresponding feature words from the original feature vector and the thesaurus, display them visually in the heat map according to the feature values, and analyze according to the distribution of the effective feature values.

8. The method for processing science and technology service resource data based on an optimization algorithm according to claim 7, wherein Step (4) specifically includes the following steps: (4.1) According to the bit positions of the optimal individual P best corresponding feature words and feature weights are respectively extracted from the original feature word library and the feature vectors, and an optimized feature vector is formed; (4.2) Display the optimized feature vectors one by one in the heat map according to the industry labels, compare the feature distributions within each industry label and between different industries, and display the feature words with large differences between various types in the heat map in the word cloud; (4.3) Record the list of feature words with significant industry influence for reference in subsequent data processing, and finally obtain the feature vector and thesaurus data with the highest accuracy and fewest features.

9. A technology service resource data processing system based on an optimization algorithm for implementing the method according to any one of claims 1 to 8, characterized in that, The described system includes: A data extraction module, which is used to extract the data of various resource libraries from the science and technology service database, and add classification labels according to the industries in the resource library, so as to facilitate the use as the original data set for subsequent feature extraction and analysis; A data preprocessing module, which is connected to the data extraction module, and is used to perform data cleaning, Chinese word segmentation, stop word removal, and feature vectorization processing on the extracted data set to obtain the initial text feature vector and thesaurus, and divide the training and test data sets according to the cross-validation method, so as to facilitate subsequent feature extraction and analysis; A feature selection module, which is connected to the data preprocessing module, and is used to optimize the obtained text feature vector based on the feature selection method of the binary banyan tree growth optimization BBTGO algorithm to obtain the optimal solution set of the classification accuracy and the number of features of the machine learning algorithm, so as to obtain the science and technology service text features; 10. A computer-readable storage medium, characterized in that, A feature analysis module, which is connected to the feature selection module, and is used to obtain the optimal feature words from the feature words of the initial text feature vector for the optimal solution of feature selection obtained, and display them in the feature heat map and word cloud for feature analysis. It stores a computer program, and the computer program can be executed by a processor to implement each step of the method for processing science and technology service resource data based on the optimization algorithm described in any one of claims 1 to 8.

Images