Optimization of Personalized Service System for Student Work in Smart Campus Colleges and Universities with the Aid of Artificial Intelligence

Examining the data source, it can be concluded that the smart campus platform’s multi-source employment data supply mainly originates from network resources and students’ employment history data. For students with different disciplinary backgrounds, different sources of employment-related data resources are selected to provide accurate employment recommendations. The prediction and analysis of employment market changes based on the collected data is reflected in the proactive employment services, and the degree of acceptance and recognition of the proactive employment services by students will be the direction of further targeted management, mining and analysis of employment-related data.

It is difficult for traditional university students’ employment recommendation service to meet the precision service requirements in a wide range, while relying on the existing data of the smart campus platform to portray the students’ characteristics, and pushing relevant information such as employment guide to students through the service platform or new media, which is more timely and precise than the traditional university employment service.

Combined with the precision type service, analyze the demand for personalized employment recommendation work service under the smart campus platform, integrate the employment data under the smart campus platform, put forward the idea of matching the students’ cultivation program with the keywords of the positions, and construct the personalized service recommendation model of employment recommendation work based on the mining of the information of the students’ cultivation program.

Under the personalized employment recommendation service on the smart campus platform, the service can be pushed to meet the personalized employment service needs of students by analyzing the attributes of different students. The precise employment service based on the virtual space of the smart campus platform can strengthen the intelligent, precise, and personalized service capability of the employment data service under the smart campus platform.

B/S three-tier architecture is now the standard structure of software systems, suitable for large-scale high-level system mode of operation, the core of the system is divided into three parts of the so-called three-tier architecture system, refers to the client and the data server between the put a “middle layer”, also called the component layer [21]. B/S architecture layering is to realize “low coupling, high cohesion”, the use of “divide and conquer” idea, the division of the problem to solve each one, easy to control, easy to extend, easy to allocate resources. The B/S application places the responsibility of ensuring legitimacy, business rules, data access, and other related tasks on the middle layer for processing. Usually, in principle, the client belonging to the representation layer does not directly interact with the database, but establishes a connection with the middle layer through COM/DCOM communication, and then interacts with the database through the middle layer.

LSTM neural network is mainly used to process sequence data with time dependence, such as speech, text, etc [22-23]. Compared with the traditional RNN, LSTM network has better memory ability when processing long sequences, and can effectively avoid problems such as gradient vanishing and gradient explosion. The key of LSTM networks is that it introduces a gating mechanism, including an input gate, forgetting gate, and output gate, to control the flow of information.

Among them, the input gate controls the importance of the input data at the current moment, the forgetting gate controls the importance of the memorized state at the previous moment, and the output gate controls the importance of the output result. In addition, the LSTM has a cell state that holds internal state information for the network. The mathematical expressions for the input gate, forgetting gate, output gate, and cell state of the LSTM neural network are: 1)

Oblivion gate: f_t = σ(W_f·[h_t–1,x_t]+b_f]).

Dynamic Time Warping (DTW) is a widely used algorithm in pattern recognition and time series analysis [24]. It is a technique used to measure the similarity between two sequences, even if they have different lengths or nonlinear temporal dependencies. The DTW algorithm can effectively measure the similarity between two sequences. The mathematical formulation of the DTW algorithm can be expressed as a matrix D, where each element in the matrix denotes the distance between the corresponding elements of two sequences. This is specifically shown in equation (1): (1) D=[ d1,1d1,2……d1,nd2,1d2,2……d2,n…………dm,1dm,2……dm,n ]

where m is the length of the template sequence and n is the length of the test sequence.

The goal of DTW is to find the optimal distortion path that minimizes the sum of distances between alignment elements. The optimal twisting path is usually found by dynamic programming, where the optimal twisting distance can be computed according to Equation (2): (2) g(c,k)=g(ik,jk)=g(i,j)=min{ g(i−1,j)+d(i,j)g(i−1,j)+2d(i,j)g(i,j−1)+d(i,j)

Where g denotes the matrix used to store the twisted distance between the two templates, g(i, j) denotes that both 2 templates are matched one by one from the starting component, and have been matched to the i component in M and the j component in T . Matching up to this step is the distance between the 2 templates. And both add d(i, j) or 2d(i, j) to the result of the previous match and take the minimum.

The final DTW distance is the sum of distances along the optimal twisted path. Mathematically, the DTW distances of the two sequences X and Y can be shown in Eq. (3): (3) d(i,j)=dist(xi,yi)+min(d(i−1,j−1),d(i−1,j),d(i,j−1))

where dist(x_i, y_i) denotes the distance between elements x_i and y_i, and d(i, j) denotes the DTW distance between the first i elements of X and the first j elements of Y .

Matrix D also provides us with the optimal twisted path, i.e., the path from D(0,0) to D(m,n), which minimizes the sum of the distances between the aligned elements.

In practical recommendation scenarios, most users’ interest preferences tend to change over time. Therefore, the user’s data can be viewed as a kind of time series, and by modeling the time series, personalized recommendation services can be provided to users more accurately. In this paper, the DTW algorithm is integrated into the LSTM model, aiming to provide a more accurate and reliable recommendation model for the personalized service of college students’ employment recommendation work. The LSTM model based on bidirectional DTW algorithm is shown in Figure 1.

Figure 1.

LSTM model based on a two-way DTW algorithm

Suppose user_a = [u₁,u₂,……,u_m] is the set of locations visited by the user, and time_a = [a₁,a₂,……,a_m] is the set of times the user has visited the locations. In the case of sparse data, the neural network model cannot accurately predict the locations that users may visit in the future. Aiming at the data sparsity and cold-start problem in the user dataset, based on the idea of collaborative filtering, this chapter calculates the similarity between the target user and other users through the bidirectional DTW algorithm, so as to supplement the target user’s dataset. First, the similarity between the target user and other users is calculated according to the bidirectional DTW algorithm, and the formula for calculating the similarity through the bidirectional DTW algorithm is shown in Equation (4): (4) similarity=d(i,j)d′(i,j)

Where d(i, j), d′(i, j) represent the minimum and maximum distortion distance between the target user and other users, respectively.

Secondly, in order to avoid the problem of too much noise and slow data processing speed caused by the introduction of too much noise in the original data set, the model randomly selects N users to calculate their similarity with the target user, and selects the common points in the user data set with the largest similarity and the target user dataset to join the target data set, where Cp = [cp₁,……,cp_n] is the set of these common points, and Ct = [ct₁,……,ct_n] is the set of the time that the target user has been to these common points. Finally, based on the perspective of “exploration”, “utilization” of the data and the “spatiotemporal” properties of the data, the model randomly adds noise to the common points in the dataset, and the specific formula is as follows: (5) cpi=cpi+zp (6) cti=cti+zt

where i ∈ [1,n], z_p, and z_t are noise added in the temporal and spatial directions, respectively, to avoid model overfitting problems.

The transformation of employment services assisted by artificial intelligence is conducive to better serving student-oriented college students with intelligent services and enhancing students’ stickiness to the platform. According to the push results of the data service can be further based on the existing students’ service satisfaction, browsing frequency, etc., to establish a deeper level of personalized service. In addition, it is also able to combine the personalized needs of students in terms of data push time period, information type, recommendation method and frequency to deepen the more intimate employment data services of the smart campus platform.

Therefore, this system pre-adopts B/S three-tier architecture to design the system. The student employment work recommendation service system is shown in Figure 2, which mainly consists of a three-layer structure, namely, interface performance layer, business logic layer and data access layer.

Figure 2.

Student employment recommendation work personalized service system

Representation layer: It is used for the interactive operation between the user and the platform, presenting the platform data to the user in a friendly and simple style that is easy for the user to understand, compatible with the access of all kinds of browsers, with robust interface usability, and ultimately supporting the interaction between the user and the employment platform. The system interface data transmission adopts asynchronous transmission, realizes local refresh of the interface, and adopts the mode of preserving the user’s front-end temporary data, preserving the user’s back-end data, and object preserving global variables for the transmission and calculation of the data between pages and pages.

Middle layer: that is, the business logic layer of the system, which is mainly composed of two parts, namely, the class files created by the developers themselves according to the business requirements and the components supported by various frameworks. In this system, the first can be specifically embodied in the method of obtaining a collection of JobInfo classes, and the second can be embodied in the namespace of the referencing system. In order to save the memory expansion problem caused by the creation of model classes, the backend of the system will use technology to instantly create a data source for the front-end server control calls.

Data Layer: In this system, the data layer is mainly embodied in two pieces: the former, which can be categorized into three data management classes according to the roles, namely, student management, management and administrator operation classes; and the latter, a public data access help class. This layer also implicitly contains its own components, which refer to the namespace according to the function’s realization. The component provides all the methods to access and manipulate backend data, which includes four provider objects: Connection, DataReadet, Command, DataAdapter, and a user object DataSet.

According to the previous analysis, the functional modules of this system can be categorized into five parts: the recommendation module, user communication module, user management module, recruitment management module, and other public modules. The overall functional modules of this system are shown in Figure 3.

Figure 3.

System function module diagram

Figure 4 shows the three-tier system structure of B/S. The classic browser-server mode three-tier architecture is used due to the system’s business realization requirements, and the following figure illustrates the three-tier system structure. Based on this architecture model, the scalability of the system and the versatility of different platforms can be better achieved.

Figure 4.

B/S trilaminal structure diagram

According to the requirements of the system business, a three-tier system architecture diagram is designed for the system, where all data processing operations are performed on the server. The intermediate business logic layer is achieved jointly through business base classes and class files, and finally reached by submitting the form to the browser. The client is only used to display data, and its display interface file is a small part of the interface.

The specific working mechanism of this system is that the user sends a page request to the server through the client’s browser, if the interface has been accessed by multiple users, the system will be cached static page to the user, to give the user an intuitive visual experience. Otherwise, it will send commands to the system server through data sets and simple object access protocols, and the system back-end analyzes and processes the user interaction requirements by calling the business base classes and public components of the business logic layer, and finally calls the five objects, scripts or stored procedures to complete the specified operations on the data tables in the database, and then finally returns the required data in the form of data tables or objects to the interface to show to the user.