Construction of Intelligent Search Engine for Big Data Multimedia Resource Subjects Based on Partial Least Squares Structural Equation

In statistical analysis, even observed differences can be constantly plagued by the problem of measurement error. In normal regression analysis, the results of measurement errors of independent variables will lead to large deviations in the parameter estimates of conventional regression models, resulting in unreliable design models. While routine analysis allows for multiple diagnoses of latent variables, it also controls for error measures, but fails to identify relationships between items. Equivalence modeling, on the other hand, enables scientists to address the problem of error measurement in analysis while identifying model relationships between latent variables. Therefore, the simultaneous equation system of the structural equation model includes the following two types of equations:

(1) Measurement equation (1) y=Λyη+ε y = {\Lambda _y}\eta + \varepsilon (2) x=Λxξ+δ x = {\Lambda _x}\xi + \delta

(2) Structural equation (3) η=Bη+Γξ+ζ \eta = B\eta + \Gamma \xi + \zeta

A structural equation is a system of equations that express the relationship between latent variables and latent variables. Equation Description:

x ⇀ is a vector of exgenous indicators

From these two sets of equations, plus some model settings, through an iterative solution process, each parameter in the structural equation model can be calculated.

In a wide range of social, psychological, and other research, scientists often encounter research-related changes that cannot be accurately and directly measured. These changes are called latent variables, such as intellectual property, educational support, family health, consumer satisfaction, consumer confidence, and more. These latent variables cannot be directly and accurately measured, but can be estimated by some indirect estimation. For example, some visual variables (eg, those that can be identified directly and concisely, rather than latent variables) can be used to measure latent variables. For example, when studying the problem of student achievement [7], researchers can use student achievement on topics such as Chinese, mathematics, genre, etc., and foreign languages (see Different Languages) are an indicator (latent variable) of student achievement. Process identification processes cannot always resolve latent variables well, whereas an equivalent model can control for both latent variables and their measurements. In other words, an equation model is a mathematical model that identifies some relationship related to the underlying discrepancy. In clinical trials, scientists have used many previous evaluation studies (often including regression measures), and although measurement error of change is allowed, in regression models it must be assumed that independent changes are wrongly measured without error, otherwise Regression analysis cannot be performed on the entire model.

In the structural equation model, for each latent variable, there are three calculation models for partial least squares, which can be called as: (1) Mode A partial least squares method, (2) Mode B partial least squares method, (3) Mode A and Mode B mixed partial least squares method, this computational model can be called Mode C partial least squares. In the mode C partial least squares method, in the iterative calculation process, it uses a part of the weight relationship of mode A and mode B respectively. According to the different parts of the selected weight relationship, the researchers divide the Mode C partial least squares method into two categories: One is called mode C₂₁ partial least squares method; the other is called mode C₁₂ partial least squares method.

According to the partial least squares method of the three modes mentioned above, the mathematical specifications are written respectively.

(1) Mode A partial least squares method

The mathematical calculation formula (weight relationship) specification of Mode A is as follows: (4) x1h=ω1hξ2+δ1h {x_{1h}} = {\omega _{1h}}{\xi _2} + {\delta _{1h}} (5) x2k=ω2kξ2+δ2k {x_{2k}} = {\omega _{2k}}{\xi _2} + {\delta _{2k}}

(2) Mode B partial least squares method

The mathematical calculation formula specification of Mode B is as follows: (6) ξ2=∑h(ω1hx1h)+δ1 {\xi _2} = \sum\limits_h {\left({{\omega _{1h}}{x_{1h}}} \right) + {\delta _1}} (7) ξ1=∑k(ω2kx2k)+δ2 {\xi _1} = \sum\limits_k {\left({{\omega _{2k}}{x_{2k}}} \right) + {\delta _2}}

(3) Mode C flat least squares method

Mode C partial least squares are further divided into the following two categories:

1) Mode C₂₁, Partial Least Squares

Mode C₂₁, the partial least squares method is in the iterative calculation, the formula (5) of mode A and formula (7) of mode B are respectively taken as the iterative calculation formula. Therefore, its mathematical formula specification is as follows: (8) ξ2=∑h(ω1hx1h)+δ1 {\xi _2} = \sum\limits_h {\left({{\omega _{1h}}{x_{1h}}} \right) + {\delta _1}} (9) x2k=ω2kξ1+δ2k {x_{2k}} = {\omega _{2k}}{\xi _1} + {\delta _{2k}}

2) Mode C₂₁ partial least squares method

Mode C₁₂ partial least squares is an iterative calculation, the formula (4) of mode A and formula (7) of mode B are respectively taken as the iterative calculation formula. Therefore, its mathematical formula specification is as follows: (10) x1h=ω1hkξ2+δ1h {x_{1h}} = {\omega _{1hk}}{\xi _2} + {\delta _{1h}} (11) ξ1=∑k(ω2kx2k)+δ2 {\xi _1} = \sum\limits_k {\left({{\omega _{2k}}{x_{2k}}} \right) + {\delta _2}}

In the process of partial least squares iterative calculation of structural equations, which mode to choose for iterative calculation is determined according to the specific conditions of the established structural equation model. Which calculation mode is chosen depends on the specific structural equation model. From the mathematical specification of the above three modes, we can know that, the calculation of the iterative process in the partial least squares method in structural equations is carried out, it is mainly based on the basic principles of the structural equation mentioned above: That is, the relationship or related information between each block structure is transmitted through the relationship between latent variables (formula (7)). Therefore, when selecting the partial least squares calculation mode of the structural equation, the relationship between the cross-block structures is actually selected. Take the structural equation model of two structures and two latent variables as an example. After careful analysis and research, we can get the following conclusion: In the structural equation model of two structures and two latent variables, if the choice of the calculation mode is based on the observed variables, it is related to whether the observed variables are exogenous or endogenous. The following is an example of an arrow diagram [8].

Based on the background of big data, the refinement of search engine types can meet the needs of the development of the times, and the progress of intelligence is an important part of the development of search engines in the new era. In the author's opinion, the key to realizing the intelligent search engine framework is to improve the framework of the big data intelligent search engine, and to improve the performance of the intelligent search engine by regularly updating the function. Therefore, especially the intelligent search engine under the background of big data analysis, can be developed from the following aspects.

The key to the realization of intelligent search engines is to strengthen the effective integration, integration and optimization design of core technologies, and to provide complete technical support for intelligent search engines. In practical research, it is found that the application dimension of intelligent search engine in big data analysis is relatively narrow, it can be effectively imported through multi-level technology, can better improve the level of intelligence, provide good environmental conditions for the intelligent construction of search engines. As shown in Figure 1, it is an intelligent search engine system based on big data analysis. It can be seen from this that in the intelligent search engine system, through the Internet, big data platform and data source import, it provides search engines with functions such as “data search”, “data acquisition” and “data classification”, the intelligent search and its system functions are ensured. In the process of building an intelligent search engine, the core technology lies in the effective implementation of data storage, modification, etc., to achieve data indexing and classification navigation. Therefore, in the process of building an intelligent search engine, the effective introduction of big data analysis provides data and platform carriers for the realization of intelligent functions, and meets the needs of intelligent development [9].

Figure 1

Search engine optimization is an important part of improving search engine performance and provides good support for achieving a clear definition of services. According to the research, we need to pay attention to the scientific use of these services for the import function of intelligent search engines. In the analysis of massive information, we should constantly improve the function system of search engines to know. Features of Smart Services Research. First of all, when using an intelligent search engine in big data analysis, you should pay attention to the function of the search engine, and improve the performance of the search engine by regularly improving the performance. The second is to complete the introduction of specific services based on big data analysis and performance to improve the operational efficiency of intelligent search engines. Finally, the generation of search engines should pay attention to the development of differentiated content, and ensure the efficiency of big data intelligent research by writing good and complete data.

According to the “ScholarSpace” website developed by the Network and Mobile Data Management Laboratory (WAM-DM) of Renmin University of China, combined with the development characteristics of the computer field, big data analysis and as a technical process can be divided into two. It is divided into four stages: data Aggregation and Collection of Data, Processing and Aggregation of Data, Analysis of Data and Interpretation of Data.

(1) Data integration and collection.

Data integration and collection is the basic work of the big data processing process, at present, the commonly used data collection methods include sensor collection, radio frequency identification technology, data retrieval and classification, etc.

(2) Data processing and integration.

After the data is mixed and compiled, the data needs to be merged. The purpose is to merge independent and uninterrupted data and share it completely. It not only solves the problem of data distribution and heterogeneity, but also recognizes the characteristics of high efficiency and easy management and management of data. Data processing and integration are generally divided into two approaches: extraction and filtering [10].

(3) Data analysis.

Data analysis refers to the processing of filtered data by means of data mining, intelligent algorithms, statistical analysis, etc., based on mathematics, through machine processing of natural language, analyzing the logical relationship between the data, and mining the potential laws of the data.. data, and summarize to form guiding conclusions. This is the core part of the whole big data processing pipeline. There are many ways and methods of data analysis, including professional analysis software, such as QlikTech, etc.; there are also artificial modeling methods, such as mathematical modeling; and neural network methods for knowledge self-improvement.

(4) Data interpretation

Translating data is an important part of the big data production process, and it is a process of using data to cultivate skills. The retrieval process is usually divided into two stages. First of all, in professional research, the big data search engine will convert hard data into design data for filtering according to the intended purpose to ensure the availability and reliability of the data. Currently, one practice is to create filters when processing data, data and products are not programmed according to the laws of aggregation or integration. Research has been carried out to ensure the success of the research; the second is searching on cloud storage and cloud computing platforms. Use data distribution, batch processing and other technologies to analyze and process the data, enter the platform, dynamically modify the retrieval strategy according to the user's needs and expectations for the content, identify and publish the preliminary evaluation results; Data that meets recovery needs is released directly. Otherwise, the content needs to be updated according to the user's recovery needs until the results are received.

In this part of the verification, the corresponding order of magnitude is measured by selecting the number of databases in the big data management platform of Siping Fire Bureau, 5G only corresponds to basic database, 100G corresponds to basic database + office database + public service library, 25T corresponds to system platform + data management and control platform, 50P corresponds to data management and control platform + system platform + business platform, the comparison of the search results is shown in Table 1. As can be seen from Table 1, with the continuous increase of the data order of magnitude, the advantages of the big data intelligent search engine based on the partial least squares structural equation algorithm are more obvious, there are significant improvements in retrieval time, precision and recall [12].

The “Rush” of the intelligent search engine based on the fractional square structural equation algorithm, the big data intelligent search engine and the big data search engine are still retrieved respectively, and the acquisition rate is shown in Table 2.

By comparing the experimental data, it can be seen that the least squares model equation algorithm is used in the big data search engine, and the recovery time is short and the volume is large in the case of real integrity[13]. Data search engines are more in line with the needs of users, unlike ordinary research, which mixes data that users do not need, which greatly improves the overall efficiency of retrieval. From another point of view, the big data search engine based on the minimum partial equivalent algorithm also improves user satisfaction, and the specific functions are as follows. (1) In terms of personalized service, the big data search engine based on partial least squares structural equation algorithm adopts the coding method, focusing on the information characteristics of retrieval conditions, through big data processing technology, the filtering of information is realized, and the personalized service is closer to the needs of users. (2) In terms of quality and quantity, the focus of big data search engines based on partial least squares structural equation algorithm is the accuracy of search results, as the order of magnitude is larger, the retrieval advantage is more obvious. (3) In terms of architecture, the big data search engine based on partial least squares structural equation algorithm drives the search mode through “Internet + big data + algorithm”, with a distributed architecture, the search efficiency is significantly improved. (4) In terms of carrier, the big data search engine based on partial least squares structural equation algorithm, through the organic combination of DE algorithm and online and offline behavior data, it provides users with more comprehensive search results [14].