Application of machine learning algorithm in electronic book database management system

Intelligent algorithms have excellent performance in managing large databases and mining effective knowledge, and many researchers have introduced these algorithms into research to improve the performance of book management systems. This article introduced the application of machine learning algorithms in electronic book database management systems. A linear classifier (linear regression algorithm is one of machine learning algorithms) was used to analyze the empirical loss value. This paper studied the application of machine learning algorithm in the database management of library special collection resources, aiming to optimize the management of library special collection resources database through machine learning algorithm, and provided students with better experience of using special collection resources. This paper verified the effectiveness of the machine learning algorithm from three aspects: the retrieval time before and after the special resource database, the number of crashes in a month and the number of library staff. The average time of the five sets using the machine learning algorithm was about 150% faster than the average search time of the five sets of special resource databases. The number of crashes after using machine learning algorithm was 2, 1, 1, 2, 1, which was much lower than that of special resource database. After using the machine learning algorithm, the five groups of workers were 7, 5, 9, 6, and 4, respectively, far fewer than the five groups of special collectors. Finally, by comparing the linear regression model, Support vector machine model, and random forest model, it was found that the accuracy of the linear regression model reached 98.2%, an increase of 8.2% compared to the random forest model, the precision rate reached 96.7%, and the recall rate reached 98.2%. 97.8%, F1 value reached 97.5%.It can be seen from the experimental data that the machine learning algorithm plays a good role in the database management of the library special collection resources. Through linear regression algorithm, it can be identified by features, so this paper establishes an attribute linear classifier for each attribute. This paper found that machine learning algorithm can reduce the crash problem of special resource database. This paper found that machine learning algorithms improve the efficiency of a database dedicated to collecting resources, thereby saving human resources. Through linear regression algorithm, it can be identified by features, so this paper establishes an attribute linear classifier for each attribute. This paper found that machine learning algorithm can reduce the crash problem of special resource database. This paper found that machine learning algorithms improve the efficiency of a database dedicated to collecting resources, thereby saving human resources.


Introduction
The 1980s marked the beginning of the automation reform in library management.During this period, libraries began to introduce automated management systems to more effectively handle and manage their increasing collection resources.However, most library automation management systems have been used for more than 15 years and have accumulated a large amount of historical data.These historical data include resource search records, reader borrowing records, book purchase records, etc., and are valuable assets for libraries.However, the potential of these data has not yet been fully exploited because they are limited by the capabilities of existing management systems.In particular, these data can usually only be stored and queried basicly, unable to achieve advanced analysis and mining, and cannot fully realize their potential in library management and providing better services.As the center of modern document information, the library has rich unique collection resources.These resources include unique documents, valuable archives and field-specific research materials.In order to better manage and display these special resources, many libraries have actively established special collection resource libraries.These libraries not only help better organize and preserve featured resources, but also provide more efficient ways to access and utilize them.Through the library's special collection resource library, readers can more easily obtain the information they need, which improves the level of information resource protection.The research and construction of special collection resources can be traced back to the United States in the early twentieth century.After entering the twenty-first century, the library community has paid more and more attention to the construction and development of these resources.This trend reflects that libraries are not only centers for the storage and dissemination of knowledge, but also guardians and promoters of cultural heritage.This article aims to explore how to better manage historical data, optimize automated management systems, and how to further develop and improve the availability and accessibility of the library's unique collection resources so that the library can better meet the needs of readers and researchers.Provide stronger support for the dissemination of knowledge and the preservation of cultural heritage.

Related works
Library special collection resources are the center of scholars' research.Posner believed that the success and challenges of the sharing of special collections of libraries provide important lessons for librarians considering the future of special collections of academic libraries.Library special collection resource sharing experts can provide special collection resource loan service, because neither the Internet nor any library can master all the information people need [1].Garnett believed that although the library's characteristic collection resources are becoming more and more digital, the Australian University Library continued to collect tangible original works, which has contributed to the national and global collective heritage [2].Jenkins believed that Wilson Library has increased the number of collections available to readers online by creating digital characteristic collection pages [3].Liemohn believed that the citation volume and other indicators of characteristic collection papers are slightly higher than those of non-characteristic collection papers.The researchers examined some paper characteristics to assess whether they can explain the high citation and download value of characteristic collection papers [4].Library special collection resources are the focus of library resources, which need to be well preserved and well promoted by this generation.Machine learning algorithms often play a good role in the database management of library special collections.
The management of library special collection resources database is very important, which often determines the efficiency of readers' borrowing special collection resources.Machine learning algorithm plays a huge role in the management of library special collection resources database.Chen believed that machine learning (ML) is regarded as a promising tool that can be used to design and discover widely used new materials [5].Collingwood believed that the history and development of special collection resources have great use in teaching, learning and research, and it has great value in enhancing the role and diversity of access in the academic library environment [6].In order to reflect the value of the library in terms of diversity in characteristic collections, Proctor stated that it is necessary to evaluate the coverage of diverse content related to characteristic collections, serving users from different backgrounds and identities [7].Sauceda's research on special collection resources provided academic librarians and archivists with an additional motivation to attract students' attention to special collection resources.In addition, he also added digital collections [8].Special collection resources are of great significance to libraries, and machine learning algorithms would also play a huge role in the library special collection resources management database.Therefore, based on the above literature, the computational cost of the linear regression model is relatively low, especially for large data sets.It can fit and predict a large amount of data in a short time.It is reasonable to use this model in this article.Compared with Other models have more advantages.
The library's special collection of resource information materials is the focus of the library, so the professionalism and timeliness of the library's special collection of resource information materials are the foundation of the library.Library should focus on collection, so it is a very important process to strengthen the management of special collection resources database based on machine learning algorithm.In the process of strengthening, the collection efficiency of library data can be improved.Machine learning algorithm can optimize the database management of library's special collection resources, and it can also make the information content of the library expand continuously.Through experiments, this paper proved that machine learning algorithm really has a better role in the database management of library special collections.

Overview of machine learning algorithms
Machine learning is one of the methods to solve the problem of data mining.Machine learning is widely used in many fields, as well as in the database management of library special collections [9].Machine learning has low computing cost, short development cycle, strong data processing ability and high prediction performance, and is widely used in material detection, material analysis and material design [10].With the rapid development of machine learning technology, as a regression problem that helps people find rules from massive data to achieve prediction results, machine learning has attracted more and more attention [11].The advantages of machine learning include flexibility and scalability, which makes it applicable to many tasks, such as risk stratification, diagnosis and classification [12].Machine learning is an interdisciplinary subject of cybernetics (control science) and computer science.Recently, it has aroused great interest from professionals and the public, who briefly outlined the historical development of machine learning [13].Book search methods also include the following: Common method: it can use bibliography, abstracts, indexes and other search tools to search for literature.The key to use this method is to be familiar with the nature, characteristics and search process of various search tools, and search from different angles.The common law can be divided into forward and backward inspection.Traceability method: the method of constantly tracking and searching the references attached to the existing literature, in the absence of search tools or incomplete search tools, this method can obtain highly targeted data, high accuracy rate and poor recall rate.Segment method: it is a combination of retrospective method and common law, and it uses the two methods alternately in stages and segments until the required information is found.

Classification of machine learning algorithms
Machine learning is a technology that can automatically build models to deal with complex relationships.The classification of machine learning can be divided into supervised learning, semi-supervised learning and unsupervised learning according to the needs of data processing.Supervised learning is to learn data with classification labels.Machine learning trains the labeled data and uses the trained model to predict new data results.So the ultimate goal of supervised learning is to train the generalization ability of machine learning.Unsupervised learning is to process data without classification labels, and can find out the potential structural characteristics of samples without training in advance.Therefore, unsupervised learning is to distinguish data according to the principle of relativity.
Semi-supervised learning is to train the model with a large number of unlabeled data and a small number of labeled data, so as to learn all samples or unlabeled samples.

Overview of linear regression algorithm
Linear regression algorithm is one of machine learning algorithms.Linear regression model is the simplest and most basic learning model in machine learning, but linear regression model also has many complex model linear regression algorithms.The problem to be solved by linear regression is to find the relationship between the target value and the input value for a given set of samples and the corresponding target value of samples, and predict the corresponding target value.Linear regression is very similar to linear classification.In the linear algorithm model, the form of the linear model is simple and intuitive, and very easy to understand, so it is very important in the linear regression algorithm.The regression algorithm model is a classification algorithm, which can fit the classification algorithm model.If the target variable is a continuous variable, the regression algorithm model can be fitted.Univariate linear regression uses a straight line to fit the data.When there is a value of x, the value of y can be predicted by the linear equation.Regression in linear regression can be understood as variable Y would change with the change of variable X.Multiple linear regression is to fit the data through a plane.Linear regression is a very common model, which is widely used in many fields.LSTM plays an important role in book search.The key of LSTM model is to replace the hidden node with a memory unit to save the history information, and dynamically access, update and forget the history information.
The LSTM model solves the long-term dependency problem.In the RNN model, the nonlinear activation function between hidden layers at different times can be regarded as updating the state of historical information record by means of overlay, which can only learn information of short period due to the problem of gradient disappearance.In the LSTM model, the Cell state parameter is introduced to record the history information of the sequence by linear accumulation, so that the long-period information can be learned.

Linear regression and linear correlation
Everything in the world is connected with other things or affects each other.This relationship can be divided into two types.One is functional relationship, which means that there is a completely certain relationship in variables.The other is called correlation relationship, which means that the relationship between variables cannot be expressed by functions.This functional relationship is uncertain.The analysis of correlation relationship is called correlation analysis.If one variable in the correlation is controllable and the other is random, this relationship is called regression relationship, and the analysis of regression relationship is also called regression analysis.Linear regression analysis is mainly used in two aspects.One is prediction.Linear regression can be used to simulate a prediction model by observing the data set.When a model is completed, for the new X value, when no matching Y value is given, a Y value can be predicted, and the other is correlation strength.

Principle of linear regression algorithm
Linear regression algorithm uses kernel method to improve the positioning accuracy by increasing the weight of adjacent anchor nodes.In the simulation part, the algorithm would be evaluated in two deployments and three topologies: conventional deployment and random deployment, and l-type, o-type and x-type topologies.As performance indicators, the average positioning error and cumulative distribution function are used [14].Linear regression accumulates input variables.These variables would be included in the linear regression equation to fill the missing value in the next incomplete variable [15].In cluster linear regression (CLR), the purpose is to divide the data into a given number of clusters at the same time, and find the regression coefficient for each cluster [16].Linear regression plays an important role in the process of variable accumulation and division [17].

Application of algorithm in special resource database
The linear regression algorithm is used in many modules of the special resource database, and the linear model of the linear regression algorithm is usually composed of multiple input variables.For a given dataset, the attributes of the data object would be used as input variables to reflect the properties of the class in the way of the data object.Therefore, classes can be identified by these characteristics, which means that each attribute has the ability to distinguish.In this paper, an attribute linear classifier was established for each attribute.A single linear classifier is easily limited, so in order to improve the accuracy of classification, it is necessary to integrate multiple linear classifiers.The empirical loss value can reflect the prediction accuracy of the model to a certain extent, so this paper used the empirical loss value to evaluate and screen the linear classifier.For the given dataset A = {a i , b i }, a i is the attribute value, and b i is the class tag value.The linear model of the linear classifier is defined as y 1 = m(x) , which means that the attribute value x is mapped to the predicted value y through m(x) .The functional rela- tionship of linear classifier is shown in Formula (1): In Formula (1), h and c are unknown parameters.When y 1 is closer to the true value y, it indicates that the predic- tion accuracy is higher.The key to solving h and c is to know the difference between y 1 and y.According to the empirical risk minimization strategy, the average loss of the predicted value of the empirical loss function y 1 and the actual value y on the data set can be obtained and recorded as N; the difference between y 1 and y calculated by the loss function is F; the value of function N is the empirical loss value, and recorded as G; then there is a function relationship as shown in Formula (2): The model with the least empirical risk is the optimal model, so how to solve the optimal solution of h and c is the key problem to minimize the empirical loss.How to solve the optimal h and c is shown in Formula (3): (1) In linear regression, the least square method is used to try to find the shortest straight line and minimize the sample loss on the straight line.The definition of the loss function is shown in Formula (4): The optimal closed form solution of h and c is shown in Formulas ( 5) and ( 6):

DBMS model
Modern database management systems (DBMS) expose a number of configurable knobs that control runtime behavior.The right choice of DBMS configuration is crucial to improve system performance and reduce costs.However, due to the complexity of a DBMS, tuning a DBMS usually requires considerable effort on the part of experienced database administrators (DBAs).Recent studies on autotuning methods using Machine Learning (ML) have shown that ML-based auto-tuning methods achieve better performance compared to professional DBAs.
DBMS model mainly includes the establishment and maintenance of background database and the development of front-end application.For the former, it is necessary to establish a database with strong consistency and integrity and good data security.For the latter, the application program is required to be functional and easy to use.Divided by function, it contains the following four modules: Const for managing constants in the project; DbFunc is used to manage the declarations, variables, and constants associated with database operations in the project; GeneralFunc is used to manage general custom functions in a project.Variable is used to manage global variables in a project.
The DBMS trains the ML-generated model on one workload and then uses the model to tune another workload's DBMS to analyze the quality of the ML-generated configuration on the other workload.The model is first trained for each algorithm using the TPC-C workload executed by OLTP-Bench.The TPC-C trained model is then used to iteratively tune the DBMS on the TicketTracker workload.The output of the TPC-C workload based on OLTP-Bench records performance data, including response time, throughput, resource utilization, etc., and uses this data to train machine learning models, which are then used to automatically adjust another job DBMS configuration under load and evaluate the performance quality of MLgenerated configurations, enabling tight integration of performance data with automated configuration and performance evaluation.The TPC-C workload based on OLTP-Bench was chosen precisely because it is a widely recognized standardized workload that represents a typical online transaction processing application scenario with complexity and diversity and can be used for database performance testing.Using TPC-C based on TPC-C's model can be used for performance evaluation, automated configuration optimization and performance prediction, and helps improve the performance and efficiency of database management systems (DBMS), so this experiment uses this model.
The performance of DBMS platform is also related to its workload, that is, the on-graph search is real-time, online automatic tuning; another feature is that it believes that the current software version update is relatively fast, and the version update will modify its available parameters.The reuse of information from the previous knowledge base makes the tuning problem more complicated, so the method proposed by it does not need to use the previous knowledge base.

Overview of library special collection resources
The document service provided by the special collection resources of the library depends on the collection quality, which is an important indicator of document value.
After the documents are selected as characteristic documents, they become special collections.The characteristic collection system is gradually developing to replace the outdated system, which would increase the consistency of metadata creation and provide users with better network access [18].The special collection should include the mechanism of managing courses, the consideration of student support and the tips of other librarians who provide such courses [19].The evaluation of characteristic collections of university libraries has gone through a transformation from a project-based model to a continuous model aimed at improving the responsiveness of library characteristic collections to campus needs [20].The special collections of the library are formed through the accumulation of time, so the construction of the library collection needs to be focused on.The contents of the characteristic collections of each school are different, but the general division is shown in Fig. 1.
In the content of library collection resources, the two words "characteristics" are highlighted.There are many colleges and universities.If the university collection needs to have unique value, it is necessary to avoid the repeated construction of library collection resources.If a school does not have the support of characteristic collection, it is difficult to stand on it.Therefore, the characteristic collection of university library is very important and needs to be constantly constructed and improved.The characteristic collection can not be completed by a single department of the library.It needs the cooperation and communication of everyone.People in all departments need to implement the work and do a good job in the management of the special collection of documents, so as to ensure the quality of the characteristic collection of the library.Different colleges and universities have different investment in software and hardware in the collection, which also leads to different collection styles.According to the requirements, schools need to carry out reasonable construction and layout of the collection resources, so as to achieve the effect of improving the quality of running schools.
It is necessary to realize the sharing of special collection resources, because the special collection resources of each school are scarce.The sharing of special collection resources makes up for the lack of special collection resources of each university, which is an effective way to improve the service value of university libraries.
The integration of characteristic subject resources by libraries is not the ultimate goal of the construction of special collection resources.The use of various information technologies to transform subject resources into digital formats and transmit them to students would further enhance the value of special collection resources.The construction of characteristic collection system is shown in Fig. 2.
The characteristic collection system of university library can convert the information input by the module into the information that can be recognized by the system, and then get the corresponding results.In order to better meet the needs of characteristic subject information, university libraries would build a complete characteristic collection document system, because the online information retrieval platform would only provide digital information.Therefore, if paper document resources alone cannot meet the needs of students, it is necessary to build a database to facilitate students to retrieve library resources, which would play a significant role in the integration and optimization of characteristic collections.The construction of characteristic database needs long-term accumulation and improvement.There are many problems in the content of resources, such as the funds for building the database, and the allocation of personnel.This requires the cooperation of professionals to do everything from the retrieval, maintenance, and updating of the database, so that the majority of college students can make full use of the digital resources of the characteristic collection.The influence of the digitalization of characteristic collections would also gradually expand, and the vigorous development of characteristic digital resources would greatly help the use of special collections.The integration of characteristic collection resources of university libraries is also the focus of characteristic collection construction.The integration of characteristic collection resources of university libraries is shown in Fig. 3.

Characteristic collection
Regional history and culture The resource integration of the library should be based on the principle of overall planning and comprehensive planning.It needs to be reasonably integrated by combining its own advantages, so as to give full play to the unique advantages of the library.This can not only improve the efficiency of the use of special collection resources, but also avoid the duplication of resources in many schools, so that college students can have a better experience.There are not only many kinds of special collection resources, but also the form of resources has surpassed many traditional documents, such as voice and image.Although the library manages the paper documents well, it manages the special collection resources in various forms generally.Therefore, better management of special collection resources would enhance the core competitiveness of the library.The characteristic collection is an important part of the library document construction and provides an important document guarantee for the teaching and scientific research work.The characteristics of the resource structure of characteristic collections should be based on the needs of users, and the resources of characteristic collections should be excavated and collected through the guidance of key disciplines.It is not enough just to buy books and periodicals and databases, but also to build databases according to the needs of readers, so as to bring better services to readers.The characteristic collection is the foundation of the library's core competence.To build core competence, libraries need to collect documents of different disciplines systematically and comprehensively, so as to form documents with different characteristics, and such libraries would also be more valuable.Today's library is not only responsible for the number of collections, but also for the number of valuable resources it can provide.Therefore, building characteristic collections can improve the information service of the library and make the utilization rate of special collection resources higher.

Difficulties in the database of special library resources
As an effective way to spread knowledge, the library has some shortcomings, but the special collection resource database can make up for these shortcomings.In fact, there are some problems that need to be solved in the library's special collection resource database.By analyzing the current situation of the existing special collection resource database, it can be clearly understood that the database application has a high degree of professionalism in the management of special collection resources, which leads to the difficulty in improving and perfecting the system during the construction of the special collection resource database.This would lead to many hidden dangers in the use of the special resource database.Therefore, corresponding management methods should be established and professional technical level should be improved, and relevant technical and management personnel should be trained, so that they can better build and manage the data of special resources.In addition, the publicity of the special collection resource database is not in place and the utilization rate is low.Many special collection resource databases are in the early stage of use, which leads to insufficient publicity.Many people do not know that there are special collection resource databases.The utilization rate of the special resource database is very low, and few people know how to use the special resource database.In addition, the special resources have not achieved the effect of resource sharing.In this case, the special collection resources cannot be effectively searched by the people in need, which makes the special collection resources limited.The technology of database is not mature enough, which is easy to cause database paralysis.This would greatly reduce readers' experience of using the special collection resource database, and would also reduce the number of people using the special collection resource database.

Machine learning algorithm library database experiment
The data in this article comes from microsoftaccess database, which can query a total of seven book categories.
According to the query of relevant information, microsoftaccess database query is divided into seven categories according to the content of books, including novels, children's books, non-fiction, professional books, reference books, manuals, diaries.Experimental environment: The experimental processor uses Intel Core i7-6800k CPU, NvidiaTITAN Xp (12 GB) graphics card, and 16 GB memory.
Through investigation, this paper made statistics on the satisfaction of 100 students with the special resource database without using machine learning algorithm.The total score of satisfaction was 100 points.The statistical results are shown in Fig. 4.
It can be seen from Fig. 4 that students were not very satisfied with the special collection resource database without machine learning algorithm, and the score was concentrated between 20 and 60 points, so there were still some deficiencies in the modern library special collection resource database.This paper selected three aspects to reflect the actual effect of machine learning algorithm applied in the special resources database: the retrieval time of the special resources database before and after using the machine learning algorithm, the number of times the special resources database crashed within a month, and the number of library staff (when the cache crashes due to some reason (such as downtime, cache service down, or not responding), it will result in a large number of requests to the back-end database.This can cause the database to crash and the entire system to crash).In this paper, ten libraries were selected for the experiment, five of which were special collection resource databases without machine learning algorithm, and the other five were special collection resource databases with machine learning algorithm (each library has approximately 3000 to 5000 types of books).The retrieval time of special collection resource database before and after using machine learning algorithm is shown in Fig. 5. Figure 5A shows before using machine learning algorithm, and Fig. 5B shows after using machine learning algorithm.
Machine learning algorithm can change data formats to speed up data loading and reduce memory usage, such as binary formats.It can also take advantage of streaming data or use progressive loading of all the data.This may require algorithms that use optimization techniques to iteratively learn, rather than algorithms that require all data in memory to perform matrix operations, such as some implementations of linear and logistic regression.It can be seen from Fig. 5A and B that the document retrieval time of the special collection resource database without machine learning algorithm in the first group of comparison was 0.75 s, and the document retrieval time of the special collection resource database with machine learning algorithm was 0.2 s.In the second group of comparison, the document retrieval time of the special collection resource database without machine learning algorithm was 0.65 s, and the document retrieval time of the special collection resource database with machine learning algorithm was 0.3 s.In the third group of comparison, the document retrieval time of the special collection resource database without machine learning algorithm was 0.7 s, and the document retrieval time of the special collection resource database with machine learning algorithm was 0.35 s.In the fourth group of comparison, the document retrieval time of the special collection resource database without machine learning algorithm was 0.8 s, and the document retrieval time of the special collection resource database with machine learning algorithm was 0.25 s.In the fifth group of comparison, the document retrieval time of the special collection resource database without machine learning algorithm was 0.65 s, and the document retrieval time of the special collection resource database with machine learning algorithm was 0.3 s.The average time of the five groups of special resource database retrieval time without machine learning algorithm is 0.71 s, and the average time of the five groups with machine learning algorithm is 0.28 s, which is about 150% faster than that without machine learning algorithm.The experimental data showed that the machine learning algorithm is beneficial to shorten the retrieval time of the special resources database.
The number of crashes of the special resource database within one month before and after using the machine learning algorithm is shown in Fig. 6. Figure 6A shows before using machine learning algorithm, and Fig. 6B shows after using machine learning algorithm.
From the data in Fig. 6A and B, it was clear that the number of crashes of the first group of special resource databases without machine learning algorithm was 5, and the number of crashes of the special resource databases with machine learning algorithm was 2. In the second group, the number of crashes of the special resource database without machine learning algorithm was 6, and the number of crashes of the special resource database with machine learning algorithm was 1.In the third group, the number of crashes of the special resource database without machine learning algorithm was 7, and the number of crashes of the special resource database with machine learning algorithm was 1.In the fourth group, the number of crashes of the special resource database without machine learning algorithm was 5, and the number of crashes of the special resource database with machine learning algorithm was 2. In the fifth group, the number of crashes of the special resource database without machine learning algorithm was 8, and the number of crashes of the special resource database with machine learning algorithm was 1. From the experimental data, it can be concluded that the machine learning algorithm can reduce the crash problem for the special resource database.
The number of library staff before and after using machine learning algorithm is shown in Fig. 7. Figure 7A shows before using machine learning algorithm, and Fig. 7B shows after using machine learning algorithm.
It can be seen from Fig. 7A and B that in the first group, there were 23 staff of the special collection resource database library who did not use the machine learning algorithm, and 7 staff of the special collection resource database library who used the machine learning algorithm.In the second group, the library staff of the special collection resource database who did not use the machine learning algorithm was 18, and the library staff of the special collection resource database who used the machine learning algorithm was 5.In the third group, the number of library staff of the special collection resource database who did not use the machine learning algorithm was 25, and the number of library staff of the special collection resource database who used the machine learning algorithm was 9.In the fourth group, the number of library staff of the special collection resource database who did not use the machine learning algorithm was 27, and the number of library staff of the special collection resource database who used the machine learning algorithm was 6.
In the fifth group, the library staff of the special collection resource database who did not use the machine learning algorithm was 19, and the library staff of the special collection resource database who used the machine learning algorithm was 4. From the experimental data, it can be seen that the number of library staff decreased after using machine learning algorithm.This indicates that machine learning algorithms have improved the efficiency of specialized collection resource databases, making books and borrowers more convenient, thereby saving human resources.This results were suitable in case of huge library, not applicable to small libraries.From the comparison of the above three experiments, it can be clearly found that machine learning algorithm did have a very good effect in the management of special resources database.
In order to better evaluate the performance of the linear regression model, compare it with the support vector machine model and the random forest model, as shown in Fig. 8.The abscissa represents the three models, and the ordinate represents the proportion of the four model performances.From the accuracy point of view, the linear regression model reached 98.2%, which is 8.2% higher than the random forest model.The Support vector machine model is least suitable for this kind of large e-book database management, with an accuracy of only 84.4%, which is 8.2% higher than the linear regression model.The regression model dropped by 13.8%; for precision, the random forest model reached 91.4%, which was 6.1% higher than the Support vector machine model.The linear regression model was the best, reaching 96.7%; from the recall rate, The linear regression model reached 97.8%, which is the best.The Support vector machine model is only 79.1%, which is the least effective.For the F1 value, the bigger the better, it can be seen that the linear regression model reached 97.5%, and the model is the best..In summary, the effect of linear regression model is the most suitable for the experiment, and the performance is the most suitable.

Conclusions
Machine learning algorithm would become an important environmental factor in the management process of special collection resources in university libraries in the future.University libraries should grasp the characteristics of the times and further integrate the construction of special collection resources into the digital construction of machine With the increasing complexity of the special collection resources database, university libraries should make full use of their own advantages to carry out digital construction of the special collection resources, so as to improve users' satisfaction with the information service of the special collection resources database.This paper studied the application of machine learning algorithm in library special collection resource database through experiments.By comparing the retrieval time of the special collection resource database before and after the use of machine learning algorithm, the number of crashes in a month and the number of library staff changes, the experiment proved that the machine learning algorithm has a good effect on the special collection resource database.However, there are still some deficiencies in the article.Due to the limitation of the conditions and the length of the article, this paper only carried out several experiments on the improvement points of the special resource data.In the future, it is necessary to conduct more research on the database management of special collection resources in order to improve the construction of special collection resources and meet user' demands.

Fig. 3
Fig. 3 Integration of characteristic collection resources in university libraries

Fig. 4 Fig. 5
Fig. 4 Student satisfaction with special resource database

Fig. 6 B
Fig. 6 Number of crashes before and after using machine learning algorithm

Fig. 7 Fig. 8
Fig. 7 Library staff allocation before and after machine learning algorithm