Analysis on College Students’ Extracurricular Learning Interests Based on Their Book Borrowing Behaviours

Extracurricular reading can broaden the horizons of college students. Exploring the patterns of students’ extracurricular learning and analyzing the relationships between extracurricular learning interests and book borrowing behaviours and the intervention mechanism of book borrowing can provide scientific guidance and basis for the comprehensive development of college students. Most of the existing learning interest analysis models have ignored other feedbacks generated in learner’s behaviours, and there are very few studies on learning interests based on book borrowing behaviours. Therefore, this paper conducts an analysis of college students’ extracurricular learning interests based on book borrowing behaviours. Based on the data of college students’ book borrowing behaviours obtained from the service terminal of the library management system, an extracurricular learning interest model for learners was constructed, and the extracurricular learning interest degrees of college students were calculated based on the borrowing time and the number of borrowing times. A multi-layer interest model based on knowledge graph was also constructed to further improve the accuracy of the analysis results of learners’ extracurricular learning interests.


Introduction
Books are the "intellectual food" for college students [1][2][3][4][5][6]. Unlike in senior high school, college students have more extracurricular study time, and reading is their most important extracurricular activity [7][8][9][10][11]. Extracurricular reading can broaden college students' horizons, cultivate their minds and relax their bodies. Besides, colleges and universities are currently aiming to cultivate interdisciplinary talents through the integrated development of multiple majors, so college students are more motivated to integrate their majors and social needs with their own learning interests [12][13][14][15]. Therefore, it is necessary to explore the patterns of students' interdisciplinary extracurricular learning and analyze the relationships between college students' extracurricular learning interests and their book borrowing behaviours and the intervention mechanism of book iJET -Vol. 17, No. 05, 2022 133 borrowing, so as to provide scientific guidance and basis for colleges and universities in the comprehensive development of college students and the cultivation of high-level talents.
With the rapid development of science and technology, interdisciplinary talents have become the target of talent cultivation and technological innovation, and thus college students' interdisciplinary learning interests have become more and more important. Zhang et al. [16] analyzed the book borrowing data of 14,600 students in Northwest Normal University using the complex network method and constructed an interdisciplinary learning interest network with disciplines as the nodes. It discussed the topological properties of the weighted network. and the importance of nodes, and divided students' learning interest communities using the hierarchical clustering algorithm. Liang et al. [17] proposed a method based on reading behaviours to capture users' interests in online learning systems, which requires pre-defined topic ontology as the reference for constructing user's interest model. It used a behaviour matrix and a weight matrix to calculate the user's interests in each leaf topic in the topic ontology. The user's interest in the topic ontology can be extended by the support factor. Takano and Li [18] proposed a prototype of an adaptive learning book system based on users' learning behaviours. The learner's learning history and behaviours are stored in the learning archive database through the monitoring function of smart phones, and their learning interests are extracted in the form of a weighted term set, so that the learning content can be adaptively provided according to users' learning interests in a series of learning opportunities. Ohsawa et al. [19] applied a chance discovery method to ID-free RFID tag data obtained in the experimental library, and proposed that the patterns obtained from the RFID tag data be regarded as the log of the learning behaviours of the learner group. The preferences of learners converge to the interests latently shared by the group in the extension of each customer's behaviours when they pick items looking at the scenario maps.
The theoretical methods for studying students' learning interests with the use of big data have provided some new ideas for the analysis of college students' extracurricular learning interests based on book borrowing behaviours in this paper. Most of the existing learning interest analysis models only sequentially model students' online learning behaviours, while ignoring other feedbacks generated in such behaviours, resulting in the lack of consideration of learners' personalized learning in the learning interest analysis models, and what is more, there are very few studies on the analysis of learning interests based on book borrowing behaviours. To this end, this paper conducts an analysis of college students' extracurricular learning interests based on their book borrowing behaviours. Section 2 presents the process of how the learner interest degree model is established. It collects and extracts the book borrowing behaviours data of college students from the service terminal of the library management system, constructs an extracurricular learning interest model for learners, and calculates the extracurricular learning interest degrees of college students based on their borrowing time and number of borrowing times. Section 3 establishes a multi-layer interest model based on knowledge graph to improve the accuracy of the analysis results of learners' extracurricular learning interests. The experimental results prove the feasibility of the proposed model for analysis of the extracurricular learning interests of college students.

Extracurricular learning interest model and calculation of interest degrees
The extracurricular learning interest model for college students proposed in this paper needs to be established based on the borrowing behaviours of learners in the library management system after learner and book sets are formed. Figure 1 shows the construction process of the learners' interest degree model. Considering the analysis of the extracurricular learning interests needs to cover different types of college students, it is necessary to collect complete data on their book borrowing behaviours. In this paper, such data were obtained from the service terminal of the library management system, and then the information on the interests of learners was extracted. Based on the data of the extracurricular learning interests of learners obtained from the analysis, the extracurricular learning interest model for learners was further established.

Modelling
The book borrowing behaviours data of learners include information on book browsing, inquiry, borrowing and rating. Learner's book ratings reflect their preferences for the books they have finished reading. If different learners have very similar preference for a certain book, it indicates that their extracurricular learning interests are similar. Assuming that learner v's rating for book i is represented by Ev,i, the following matrix of A×B can be constructed: In the matrix, there are A learners and B books, so the elements in the matrix are the learners' ratings for the books they have read, ranging from Level I to V. If a learner has not read a certain book, the learner's rating for the book will be 0. A learner's interest in a book can be represented by the rating given by the learner for the book. Suppose that the rating of learner x for book i is represented by Ex,i, that the rating of learner y for book i by Ey,i, and that the set of books that have been rated by both learner x and learner y is represented by EQx,y. Eq.(2) shows how to calculate the similarity between the ratings given by learner x and learner y: , xy x y If learner x and learner y have similar extracurricular learning interests, they will have similar preferences for book types. One book can have multiple class attributes. Assuming that the class attributes of the book are represented by Yj, Eq. (3) shows the expression of the class attribute set of the book: Each book corresponds to a two-dimensional matrix of class attributes. When the value of the matrix element Xi,j is 1, it means that book i falls within the book class attribute j. When the value of the matrix element Xi,j is 0, it means that book i does not fall within the book class attribute j. The extracurricular learning interest degree of a learner shows how much this learner prefers a certain type of extracurricular books. The more times a learner borrows a certain type of books, the more interested this learner is in this type of books. The interest degree of learner x in book class attribute j can be calculated by Eq. (4): , , Suppose that the total number of times learner x has borrowed books of class attribute j is denoted as Mx,j and that the total number of times this learner has borrowed books of all class attributes is denoted as Mx.
Suppose that the preference of learner x for books of class attribute j is represented by EQx,j, that the preference of learner y for books of class attribute j by EQy,j, and that the total number of book class attributes by m. Learner x's preference set INTx,j={EQx,1, Suppose that the balance between the similarity of the learners' ratings of books SIMPJ(x,y), and the similarity of the learners' preferences for book class attributes SIMINT(x,y) is represented by δ, whose value ranges between [0,1]. Based on SIMPJ(x,y) and SIMINT(x,y), the similarity SIM(x,y) between learners can be calculated. Substitute SIMPJ(x,y) and SIMINT(x,y) into the weighted calculation to obtain SIM(x,y):

Calculation of interest degree based on borrowing time and number of borrowing times
Through analysis of the scaling laws of learners' book borrowing behaviours, it can be found that the patterns of learners' book borrowing behaviours are related to borrowing time. Suppose that a learner's extracurricular learning interest degree corresponding to the borrowing time is ETo, and that the actual borrowing time O, calculate the extracurricular learning interest degree of every two intervals, and there are:

where, when Omax<Oi≤2Omax, O=10×(Omax-Omax)/(O-Omax).
According to the definition of extracurricular learning interests, the more times a learner borrows books of a certain class attribute, the more interested this learner is in the knowledge presented in books of such class attribute, and the higher the learner's extracurricular learning interest degree is. The preferences for a certain type of books are divided by the number of times Y such books have been borrowed. With the maximum number of times Y_max the books have been borrowed by learners majoring in a iJET -Vol. 17 Through averaging of the interest degrees related to borrowing time and number of borrowing times, the comprehensive extracurricular learning interest degree is ELI=[ETo+Y]/2. Through the above operations, a rating matrix SM of learner-bookextracurricular learning interest degree can be formed based on the book borrowing behaviours data of learners, as shown in Eq. (10), where the number of rows and that of columns of SM correspond to the number of learners n and the number of books m, the student set is r={r1, r2,...,rm}, the book set isτ={τ1, τ2,..., τn}, and the extracurricular learning interest degree set of learners is SR={SR11,SR2,1,...,SRn,m}, and there is: 3

Construction of the multi-layer interest model based on knowledge graph
The structure of the multi-layer interest model based on knowledge graph is shown in Figure 2. The users' records of books that the users have finished reading are linked to their deep-level extracurricular learning preferences on the book knowledge graph, such as book types, content and authors, etc. At the same time, the attributes of learners are incorporated to find out the relationships between the users' reading behaviours and attributes.
The analysis algorithm for the extracurricular learning interests of learners with the knowledge graph as auxiliary information can not only improve the accuracy of the analysis results of learners' extracurricular learning interests, but also interpret the analysis results to some extent. The book-based knowledge graph has integrated the multisource heterogeneous data of book borrowing behaviours to accurately characterize the extracurricular learning interests of learners and in this way, the extracurricular learning interests of learners can be extracted at a deep level. Based on the above analysis, the model was constructed as follows. Figure 3 shows the book set and learner attribute generation model, and Figure 4 presents the mining model for deep-level extracurricular learning interests.

Fig. 4. Mining model for deep-level extracurricular learning interests
Suppose the book borrowing behaviour record of learner v is βv, Since the relevant entities of the constructed knowledge graph match the books, the vector representation uiℝ p corresponding to the book ui(i=1, 2,...,Mi) can be obtained through knowledge representation learning. In order to construct the multi-layer learner interest model based on knowledge graph fusion for the extracurricular learning interests of learner v, first directly average the features of the books that the learner has read, and then assign uniform weights to all the books that have been read according to the following formula: (11) Assuming that the weight factor in the adaptive weighting is represented by ξi, based on the candidate books ujR p , assign a different weight to each ui in the book set βv that learner v has finished reading, and perform the weighted average operation to further obtain the representation of the shallow-level interests of learner v: The correlation between the given books ui and uj can be fitted in the form of an inner product based on the function G', and then converted into the weight factor in the above formula through the softmax function: The book knowledge graph contains massive information on the knowledge of the books. Analyzing the correlations in knowledge between different books can help extract the deep-level extracurricular learning interests of learners. Given the book knowledge graph H, taking the books βv that the learner v has finished reading as the head entity, the triplet set Rv of other related book entities and those that have been read can be obtained by links along the book knowledge paths. Suppose the tensor representation of the book knowledge relation is siR p×p , and that the vector representation of the head entity matched by the books is fgiR p . Based on the inner product function G'', the weight factor δi of the candidate book uj and the triple (fgi,si,foi) to the head entity fgi under the relationship si can be calculated: After the calculation of the relevant weight factors of all triples in Rv is completed, link the learner to the book entities and calculate the weighted sum, which can be used to characterize the learner's deep-level extracurricular learning interests:

Experimental results and analysis
In order to build a model with better analysis performance, the relevant parameters of the model were checked, including the embedded dimension and the weight of the regularization term in the knowledge graph, with the results shown in Figures 5 and 6. It can be seen that the accuracy, recall rate and F1 value gradually increase with the increase of the embedded dimension. However, if the embedded dimension is too large, the model training will be over-fitted, and the three performance indicators will all decline. The weight of the regularization term in the knowledge graph should be neither too large nor too small; otherwise it will not be able to provide regularized constraints or make the objective function more focused on the optimization of the knowledge graph. In this paper, the embedded dimension of the model was set at 16, and the weight of the regularization term in the knowledge graph was set at 0.01, to obtain better model analysis performance.
iJET -Vol. 17, No. 05, 2022     Table 1 shows the results of the ablation experiment for the model.

1) 2)
3) Fig. 7. Performance of the model with different numbers of book class attributes This paper conducted an analysis on the extracurricular learning interests of college students based on book borrowing behaviours. Based on the book borrowing behaviours data of college students extracted from the service terminal of the library management system, an extracurricular learning interest model for learners was constructed, and based on the borrowing time and the number of borrowing times, the extracurricular learning interest degree of college students were calculated. Then, a multi-layer interest model based on knowledge graph was established to further improve the accuracy of the analysis results of learners' extracurricular learning interests. In order to build a model with better analysis performance, the embedded dimension of the model and the weight of the regularization term in the knowledge graph were checked. After that, the performance of the model with different numbers of book class attributes was discussed, showing that the proposed model can perform better than the other ones in terms of all three indicators, and that the analysis performance of the model has been effectively improved.