#### Abstract

In order to meet the differentiated needs of students and improve the satisfaction of college mental health textbook recommendation, a college mental health textbook recommendation scheme based on improved deep learning algorithm is proposed. Based on the analysis of the principle of deep learning data recommendation system, the data interest is calculated according to the browsing records of students on college mental health textbooks. Combine the deep learning algorithm and collaborative filtering algorithm to collect the demand data of college mental health textbooks, then use the Naive Bayesian classification method to divide the college mental health textbooks into interested and uninterested parts, and recommend the interested college mental health textbooks to the students in need. Experiments show that the longest recommendation time of the college mental health textbook recommendation scheme based on the improved deep learning algorithm proposed in this paper is 4.5 min, the highest recommended recall rate is 95.18%, the average accuracy is 97.2%, the highest content richness is 0.8, the system stability coefficient is 1.06, and the overall average praise rate is 97.79%. It has a good recommendation effect.

#### 1. Introduction

With the rapid development of science and technology, the research results produced by various industries and disciplines have increased exponentially. The generation of a large number of academic achievements not only provides scholars with rich academic resources but also brings difficulties and challenges to the retrieval of resources. Taking mental health data as an example, in the process of using by students or teachers, it is necessary to query and quote the corresponding mental health data and conduct further research and optimization on the basis of the previous research, so as to ensure the value and feasibility of the research results. However, in the actual retrieval process, it is difficult for users to accurately obtain the most valuable mental health materials in a short time [1]. In order to improve the retrieval speed of information related to the mental health of users, scholars not only recommend the methods of searching information in the network, but also solve the problems of mental health of users [2, 3]. At this stage, the recommendation methods of college mental health teaching materials at home and abroad can be divided into content-based filtering [4], collaborative filtering [5, 6], and rule-based recommendation methods [7] according to the input data and composition structure. However, the above traditional resource recommendation methods are mainly aimed at static mental health data stored in solid-state database, which will lead to poor recommendation effect and slow recommendation speed [8, 9].

In order to solve the problems existing in the traditional college mental health teaching material recommendation methods, although the resource recommendation method based on deep learning improves the recommendation performance, it only considers the user’s rating data of resources and inhibits the recommendation effect [10, 11]. Therefore, based on the improvement of deep learning algorithm [12], this paper designs and studies the recommendation method of college mental health teaching materials. The demand data of college mental health teaching materials are collected, the Naive Bayesian classification method is used to classify the demand data accurately and quickly, and the college mental health teaching materials of interest are recommended to the students in need according to the category. In order to ensure the recommendation performance, improve the recommendation effect and meet the resource needs of users.

#### 2. Data Recommendation System Based on Deep Learning

The existing technology includes content-based recommendation method, collaborative filtering recommendation method, rule-based recommendation method, utility-based recommendation method [13], and knowledge-based recommendation method [14]. Although the above methods can realize data recommendation to varying degrees, there are still some technical deficiencies in the single use process. The disadvantages of the recommended method are shown in Table 1.

Therefore, when studying the recommendation method, this paper adopts a stacked denoising autoencoder (sdae) algorithm [15, 16], which integrates the limit learning machine into the noise reduction self-encoder stack with the help of the learning idea of layer by layer progressive and self-coding. Finally, a deep learning model of stack noise reduction self-coder based on extreme learning machine (ELM) calculation is formed [17, 18]. The algorithm also constitutes the mathematical modeling form of extreme learning machine (ELM). The stack noise reduction self-coder further calculates the collected teaching data affecting college mental health. In the learning process, further extract the sample features, image features, or text data features of the data to be learned [19], so as to further learn the received abstract sample data, and apply the learning results to the nearest neighbor algorithm for further prediction and scoring [20, 21]. Through this algorithm, we can actively search and find the information with high user interest from the massive database information and recommend the information to the users. It can effectively solve the problem of users’ personalized expression of information in the form of personalized information retrieval service.

Based on the above description, this paper constructs an application architecture including input layer, model layer, output layer, and application layer. The recommendation system is shown in Figure 1. In the input layer, various data information in the database are retrieved, such as university mental health teaching materials and psychology books. Then, the hybrid sdae recommendation model algorithm is used for calculation. After the calculation results are output, they are displayed in the form of recommendation list. By controlling these data in time, users can grasp the storage of college mental health teaching materials in time. The architecture is shown in Figure 1.

The recommendation model studied in this paper can fully consider the nonlinear characteristic changes of data, so the model has better stability in data output. However, the constructed data model also has a series of problems such as data sparsity [22] and cold start. The method proposed in this paper can better learn the different distribution characteristics of information data in the most perfect way.

#### 3. Recommendation of College Mental Health Teaching Materials Based on Improved Deep Learning Algorithm

Based on the above theoretical analysis, this algorithm combines the collaborative filtering recommendation algorithm [23, 24] with the hybrid sdae recommendation model to mine the personalized demand information of college mental health teaching materials and then uses the Naive Bayesian classification method to realize the recommendation of college students’ mental health teaching materials.

##### 3.1. Personalized Demand Information Mining Based on Improved Deep Learning Algorithm

The information of college mental health teaching materials for students’ personalized needs is determined according to the students’ interest in the materials [25]. Therefore, the interest of the materials can be effectively judged through the browsing records of students’ college mental health teaching materials.

In the process of data learning, each resource includes the resource itself and resource profile at the same time. The interest degree of data is calculated according to the characteristics of the content. Based on collaborative filtering recommendation algorithm and hybrid sdae recommendation model, the interest of college mental health teaching materials is calculated [26]. Assuming that the total number of times students browse a certain material is, the length of a single browsing is, the total number of bytes is. Before solving the students’ interest in the mental health teaching materials of the university, it is necessary to solve the browsing interest in the data content and profile [27], and then, the calculation formula of interest in the data content is where represents the number of views, represents the duration of the browsing, and represents the total number of content bytes.

The calculation formula of interest is where represents the number of views, represents the duration of the browse, and represents the total number of introduction bytes.

Data browsing interest is the sum of data content interest and data profile interest. The calculation formula is

The student’s interest in the information resource is: where represents the total number of college mental health teaching materials. When , it means interest; otherwise, it means no interest.

Through the obtained interest, the university mental health teaching material information of students’ personalized needs is constructed, and the system classifies the materials according to the information.

The similarity based on conditional probability is asymmetric. In order to avoid the limitation of data classification and further modify the classification similarity calculation, an optimized collaborative filtering information mining method based on improved conditional probability is designed. The interest between different categories is adjusted through parameter , and the personalized demand information is obtained according to the category item similarity matrix : where is the category with high similarity, l is the number of recommendation sets, is the classification item matrix, and is the recommendation set of output target users. So far, personalized information demand mining has been completed.

##### 3.2. Recommendation of College Mental Health Teaching Materials Based on Naive Bayes

Taking video teaching resources as an example, video resources are a new type of teaching resources formed by the gradual development of multimedia technology. At present, it is widely used in school teaching. First of all, because it can integrate a large number of teaching contents into video resources in a short time, its convenience and quickness are deeply loved by the majority of psychological educators. For example, by using video teaching resources in college students’ mental health education and teaching, students can learn and understand more mental health education knowledge through the character cases and events in video teaching. Secondly, video teaching resources are reusable. A typical teaching case can be reused, which greatly alleviates the work intensity of teachers and enables teachers to have more energy to provide individual counseling for students’ personal psychological problems. Teachers should carry out video teaching in order to improve the teaching effect. For example, at present, college students commit suicide frequently due to online loans and small loans, mainly because college students do not develop a good psychological quality and choose not to solve but to escape in the face of strong pressure, which leads to the frequent occurrence of suicide. Not only that, the number of college students committing suicide due to love is also rising year by year. In the final analysis, college students do not have good mental health quality. Therefore, teachers should show, make, and integrate relevant videos for students according to the psychological problems of college students and relevant cases that affect college students to establish good mental quality, so as to show the harm of college students’ bad psychology to students and the examples around them that can make students feel intuitively, solving students’ psychological obstacles.

Therefore, the essence of automatic recommendation of college mental health teaching materials using personalized needs is to classify the materials according to the records of students’ seen materials. Therefore, the seen materials are set as a sample set for automatic classification of materials [28, 29]. Naive Bayes is a classification method with simple structure, accurate classification, fast operation, and stable performance [30]. Let the sample set be , which contains categories, then and represent the attribute values contained in a single sample, then , the feature vector is , and the posterior probability formula is where is a constant, representing all categories. According to the concept of maximum likelihood hypothesis, equation (6) is simplified as

Using the personalized demand data information mined, the viewed data are divided into and categories, which are interested and uninterested [31, 32], and the training set is constructed according to the extracted data resource feature vector. If the new university mental health teaching material is not classified and described, its feature vector is constructed. Calculate the category probability of data through equation (7). The formula for calculating the probability of belonging to category is

The formula for calculating the probability of belonging to category is

If the total amount of class data resources is and the total amount of class data resources is , the a priori probability [33, 34] is divided into

Calculate the probability of class and class , respectively, according to formula (10) and formula (11), namely, assuming that the total eigenvector of data is , the total number of occurrences of in categories is , and the total number of occurrences of in categories is . The calculation formulas of category attribution probability [35, 36] of data are as follows:

According to the calculated probability of interest of class and class , if , it means that students are interested in the material; otherwise, it is not interested. According to the judgment results, the system automatically recommends college mental health teaching materials with personalized needs to students.

#### 4. Experimental Analysis

##### 4.1. Setting Up Experimental Environment

The experiment uses floyhub as the training and recommendation effect test platform. The test environment includes a server and multiple computer devices. The configuration of the hardware and software environment of the experimental environment is shown in Table 2.

In addition, due to the application of knowledge map and deep learning algorithm in the recommended method of college mental health teaching materials, it is necessary to embed the corresponding running program plug-ins on the basis of the experimental environment to ensure the cooperative operation of the two technologies.

##### 4.2. Preparation of Mental Health Data Samples

The resource data samples used in the experiment can be provided by multiple university libraries. In addition, web crawlers can be used to obtain the resource sample data of academic papers, academic conferences, and other types in multiple academic and teaching networks. The prepared resource data samples include Chinese and English. After analysis and unification, the experimental data samples are obtained, as shown in Table 3.

In addition, 100458 comment records and behavior records were imported into the experimental environment. Upload all the prepared resource data samples to the experimental environment, and select the methods of literature [6] and literature [7] as the comparison method for the following tests.

##### 4.3. Establishment of Evaluation Indicators

In order to ensure the accuracy of recommendation ranking of recommendation algorithm, it is necessary to measure the correlation between recommendation elements and judge the ranking quality of recommendation algorithm according to the position of recommendation results, so as to obtain the index to measure the recall rate of recommendation algorithm. The calculation formula is as follows. where is the positive example of all test sets and is the comprehensive number of test sets of the first mental health teaching materials in the database. When the number of mental health teaching materials browsed by several browsing users in the test set is 10, 11, and 12, respectively, there are 7, 8, and 9 books in the recommendation list arranged by the recommendation algorithm in the test set. At this time, the value of the index measuring the recall rate of the recommendation algorithm is . The closer the index is to 1, the more accurate the recommendation algorithm is.

##### 4.4. Analysis of Experimental Results

Let the training data set be the negative sample in the experiment, and the positive sample is the browsing mental health teaching material record of the browsing user. According to the browsing volume of individual users, randomly select from the nonbrowsed mental health teaching material information. After 15 rounds of iteration, calculate the index size of the sampling data set according to the positive and negative samples with a ratio of 1 : 1, as shown in Figure 2.

By analyzing the data in Figure 2, it can be seen that the recommended recall rate of the recommendation algorithm based on optimized clustering is 0.42 at the lowest when the number of negative samples is 1 and then increases with the increase of the number of negative samples, with the highest recommended recall rate of 0.62. The highest recall rate of the method in literature [6] is 0.44, and the highest recall rate of the method in literature [7] is 0.44, which is lower than that of this method, which proves that the recommendation effect of the recommendation algorithm based on optimized clustering proposed in this paper is more accurate.

The number of target recommended contents is set to 500. The methods of literature [6], literature [7], and this paper are used to recommend to the target group, respectively. The completion time of 500 recommended contents by the three methods is counted to verify the complexity of different algorithms, as shown in Figure 3.

By analyzing the data in Figure 3, it can be seen that the maximum recommendation time of the recommendation algorithm based on optimized clustering is 4.5 min, the maximum recommendation time of the method in literature [6] is 5.8 min, and the maximum recommendation time of the method in literature [7] is 5.5 min. It can be seen that the recommendation algorithm based on optimized clustering proposed in this paper has the least recommendation time, lower complexity, and higher recommendation efficiency. The data set is used to compare the recall and precision of data information. The experimental results are shown in Figure 4.

**(a) Data automatic recommendation recall**

**(b) Data automatic recommendation precision**

As can be seen from Figure 4, compared with the two comparison methods, the recall rate and precision rate of this method are the highest. The results show that the average recall rate of this method is 95.18%, which is 17.52% and 11.11% higher than that of literature [6] and literature [7], respectively. The average precision rate is 97.2%, which is 25.6% and 8.51% higher than that of literature [6] and literature [7], respectively. It shows that this method has a strong ability of personalized automatic recommendation of college mental health teaching materials.

For the college mental health teaching materials recommended by the system, students are usually more likely to accept the top contents in the recommendation list. Therefore, the top 16 recommendation materials are used for analysis and evaluation, and the average ranking degree is used for the comparative experiment of recommendation accuracy. The experimental results are shown in Figure 5.

It can be seen from Figure 5 that with the increase of the number of materials in the recommendation list, the average ranking degree of the three recommendation systems gradually increases, resulting in the continuous decline of the accuracy of data recommendation. Although the average ranking degree of the method in this paper shows a slight downward trend, compared with the methods in literature [6] and literature [7], the automatic recommendation accuracy of college mental health teaching materials by this method is the highest, which shows that this method can effectively recommend college mental health teaching materials with personalized needs for students.

At present, the society is in a diversified era, and students should receive information containing diversified elements. The teaching contents recommended by the system cannot be single but need to be diverse and novel, which not only meets the needs of students for knowledge but also conforms to the development of the times. A comparative experiment is carried out with the richness of the content of college mental health teaching materials automatically recommended by the system. The results are shown in Figure 6.

It can be seen from Figure 6 that with the increase of the number of recommended list materials, the richness of the recommended contents of the three methods shows an upward trend. When the number of recommended materials is small, the difference in the richness of the data content is not obvious. With the gradual increase of the data content, the difference in the richness is more and more obvious. The results show that the highest content richness recommended by the method of literature [6] is 0.65, and the highest content richness recommended by the method of literature [7] is 0.68, while the content richness of the university mental health teaching materials recommended by this method is 0.8, which has the richest content, which is more in line with the current needs of students for diversified knowledge.

A good system should not only have good applicability but also have strong operability. A comparative experiment on the operation stability of the system itself is carried out for the three methods. The experimental results are shown in Figure 7.

It can be seen from Figure 7 that the operation stability of the three methods changes differently with the increase of college mental health teaching materials. When the number of materials in this method is about 400, the stability coefficient is the highest, which is 1.06. Before outputting 450 materials, it has maintained a relatively stable state with small fluctuation. The maximum stability coefficients of the two comparison methods are only 0.73 and 0.65, respectively, and the fluctuation is relatively large during operation. This shows that the recommendation method of college mental health teaching materials in this paper has the best operation stability.

All the subjects were divided into three groups on average, and the experience operations of the three methods were carried out, respectively. The main statistics were the students’ evaluation on the applicability of the system, recommendation satisfaction, operation convenience, and content integrity. The evaluation results are shown in Figure 8.

It can be seen from Figure 8 that the method in this paper has obtained the best evaluation in all aspects. The overall average praise rate is 97.79%, the praise rate of literature [6] method is 93.32%, and the praise rate of literature [7] method is 93.75%, which is 4.47% and 4.04% higher than that of literature [6] method and literature [7] method, respectively. Experiments show that this method has good effect in practical application, high student satisfaction, and higher practical application value.

#### 5. Conclusion

With the continuous progress of the concept of education and teaching, the traditional teaching method based on teachers’ antiaspect explanation of knowledge has not been adapted to the development of education and teaching. Especially in the classroom of students’ mental health education, it not only reduces students’ interest in learning but also makes students feel boring with a single explanation, which is difficult to produce spiritual perception and cannot obtain the expected teaching effect. If different types of teaching materials are used in teaching to guide and inspire students, it will play a positive role in improving teaching effect and students’ interest in listening to classes. This paper proposes a recommendation method of college mental health teaching materials based on the improved deep learning algorithm. Through comparison and verification, it is proved that the longest recommendation time of the proposed method is 4.5 min, the average recall rate is 95.18%, the average precision rate is 97.2%, the highest content richness is 0.8, the system stability coefficient is 1.06, and the overall average praise rate is 97.79%. It has the practical application performance of plus sign.

In the field of deep learning technology, the construction of recommendation system is based on information filtering and information retrieval technology to provide functions suitable for different personalized information needs for different users. This paper realizes the improvement of deep learning by constructing a collaborative filtering recommendation algorithm mixed with sdae and time average model. The sdae recommendation model algorithm adopts the idea of layer-by-layer self-coding. This paper makes an in-depth study on this problem, which lays a deep theoretical foundation for the further research and application based on recommendation learning system.

#### Data Availability

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

#### Conflicts of Interest

The authors declared that they have no conflicts of interest regarding this work.

#### Acknowledgments

This work supported by the Key Topics of Educational Science Planning in Heilongjiang Province (GJB1421105): Research and exploration on university data governance system from the perspective of big data, Research project of 2021 school level higher education teaching reform of Northeast Petroleum University, and Research on the construction mode of intelligent learning situation analysis platform under the background of education cloud.