Online course mixed recommendation method based on network language formal background

Abstract

The invention discloses an online course mixed recommendation method based on a network language form background. The method comprises the following steps: collecting and processing user behaviors and social data: initializing the collected data, constructing a network adjacent matrix, forming the network language form background, and calculating network language concept knowledge; mining association rules to form an association rule base; using user collaborative filtering recommendation to obtain a preliminary recommendation sequence; performing rule verification and optimizing the recommendation sequence; and recommending online courses by adopting Top-N according to the recommendation intensity and the optimized course list. According to the method, historical behaviors and social network data of the users are comprehensively considered, and compared with a traditional recommendation system, the method can more accurately capture user relations and interests, and the reliability and accuracy of recommendation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining and recommendation systems, and in particular to an online course hybrid recommendation method based on a network language form background. Background Art

[0002] At present, the course resources in the field of online education are extensive, and the information overload problem faced by users is becoming increasingly serious. How to accurately find content that suits personal needs in massive courses has become a research hotspot. Traditional recommendation systems mainly include the following: content-based recommendation systems proposed by Pazzani et al., which analyze the project features of the user's past history and recommend new projects with similar features; model-based matrix decomposition technology recommendation systems proposed by Koren et al., and model-based recommendation systems also include clustering, neural networks, etc. The above recommendation methods usually rely on single behavioral data of users, such as browsing records and course ratings. This method is often insufficient when facing the dynamic changes in user needs and the diversity of course content. Since users' social relationships can significantly affect their learning preferences and choices, the integration of social networks into recommendation systems has emerged. Li Jianghao et al. proposed a topic-aware network embedding method based on online social networks to provide intelligent recommendation services; Ma Gangfeng et al. proposed a social preference network, and used multiple different graph neural networks to represent and recommend.

[0003] Formal concept analysis organizes data into a hierarchical structure, making the relationship and structure of the data clearer. This visual hierarchical structure helps understand the relationship between users and items, and provides easy-to-interpret recommendation results. More importantly, formal concept analysis can reveal implicit rules and patterns in the data, thereby optimizing the recommendation algorithm. Li Xin used formal concept analysis to mine implicit associations in book borrowing records and implement book recommendations based on association rules; Liu Shengri implemented a personalized recommendation system on a video website based on concept lattice association rules. With the continuous development of formal concept analysis research, the intersection with the field of recommendation systems has also been widely studied. Zhao Xuejian et al. proposed a recommendation rating prediction method based on the simplification of user interest concept lattices, which effectively dealt with the problem of data sparsity; Chen Haowen et al. combined the idea of ​​collaborative filtering with formal concept analysis and proposed a Top-N recommendation algorithm based on concept neighborhood.

[0004] As one of the traditional recommendation methods, collaborative filtering is still the most common recommendation method today. It is a technology that recommends by analyzing the similarity between users and items. However, collaborative filtering often faces the following two problems. First, collaborative filtering is like an automatically running "black box". In user-based collaborative filtering, the recommendation system will automatically find users similar to the target user, and then take a weighted average of the scores of similar users to predict the target user's score for the item, but people know nothing about the scoring rules and recommendation mechanism; another problem faced by collaborative filtering is "cold start". Since a large amount of interactive information between the target user and the system is needed as the basis for recommendation, it will cause the problem of "less is less, more is more".

[0005] Therefore, a new recommendation system is needed that can comprehensively consider users' social relationships and historical behavior data, especially language value data, and make effective improvements to the "black box" and "cold start" problems, so as to provide more accurate and personalized online course recommendations with a reliable mechanism. Summary of the invention

[0006] The present invention aims to solve the above-mentioned technical problems existing in the prior art and provides an online course hybrid recommendation method based on the network language form background.

[0007] The technical solution of the present invention is: a hybrid online course recommendation method based on the network language form background, which is carried out according to the following steps:

[0008] A. User behavior data collection and preprocessing

[0009] A1. Set the user's language term set to S = {s t |t=-τ,...,-1,0,1,...,τ}, the set of online courses is L n = {l 1 ,l 2 ,l 3 ,...,l n}, the user set is U = {x1,x2,...,x m}, n is the total number of online courses, and m is the total number of users;

[0010] A2. Collect user x m ∈U uses the language value s t Describe the online course n ∈L n The concept of language The language concept set is Form a user evaluation form for the course language;

[0011] A3. Collect and quantify user x i , x j∈U social relationship is Sim ij ∈[0,1], construct the network adjacency matrix SM=(Sim ij ) n×n ;

[0012] A4. Based on user set U and language concept set The user's course language evaluation form and the network adjacency matrix SM construct the network language form background, and use the following definition to obtain network language concept knowledge;

[0013] definition and The induced operator on : If any user x i ∈U and x j ∈U is related and but It is a network language concept knowledge;

[0014] Define user x i , x j If there is a relationship, then there is a path x between two users in the social network. i ,x k ,...,x l ,x j Sim ik ,...,Sim lj ≥λ, where λ is the given minimum relationship threshold;

[0015] B. Online Course Association Rules Mining

[0016] B1. Based on the knowledge of network language concepts, determine the concept tuple (C s ,C f ), C f It is C s The super concept of C s It is C f sub-concepts of;

[0017] Calculate all association rules using partial order relations between all concepts in

[0018] B2. Determine the support of association rules: Confidence of association rules: By setting the minimum support and minimum confidence of association rules, the association rules with higher reliability are screened out to form an association rule base;

[0019] C. Generate and optimize recommendation sequences

[0020] C1. Use user neighborhood collaborative filtering to form a preliminary recommendation sequence, that is, use the network adjacency matrix SM to screen several users who are closest to the target user, and form a preliminary recommendation sequence with the courses watched by the closest users but not watched by the target user. The recommendation strength is the social network weight of the adjacent users and the target user;

[0021] C2. Verify the rules and optimize the recommendation sequence, that is, use the historical course viewing records of the target user to perform association rule matching to optimize the preliminary recommendation sequence: for online courses in the rule consequences, if they have positive semantics, then their recommendation strength is increased in the preliminary recommendation sequence; if they have negative semantics, then their recommendation strength is weakened in the preliminary recommendation sequence; if they have neutral semantics, then their recommendation strength remains unchanged in the preliminary recommendation sequence; the final optimized recommendation sequence makes Top-N recommendations to the target user according to the recommendation strength.

[0022] The hybrid recommendation method for online courses based on the network language form background proposed in the present invention is a hybrid recommendation model that conducts formal analysis on social relationships between users and user behaviors, mines semantic association rules that can consider user social relationships, and optimizes and adjusts the user neighborhood recommendation sequence of social networks with rules. Compared with traditional machine learning recommendation methods, first of all, user information is considered more comprehensively, which can not only consider user social relationships, but also analyze user historical behavior preferences; secondly, the transparency and reliability of the recommendation mechanism is another advantage of this recommendation system. The calculation of semantic rules and the user's recommendation process have a causal relationship, and users can understand how to obtain the recommendation results. In general, the present invention can alleviate the "black box" and "cold start" problems of traditional recommendation systems by means of cascade recommendation and rule verification, and ultimately achieve accurate and reliable personalized recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a structural block diagram of an embodiment of the present invention.

[0024] Figure 2 A social network diagram of all users in the embodiment of the present invention.

[0025] Figure 3 Embodiment SL of the present invention 0.5 (U,L St ,SM,I) Add the structural diagram of the concept partial order relationship.

[0026] Figure 4 This embodiment of the present invention determines the nearest user topology graph based on the social network graph of all users. DETAILED DESCRIPTION

[0027] The online course hybrid recommendation method based on network language form background of the present invention is as follows Figure 1 As shown, follow the steps below:

[0028] A. User behavior data collection and preprocessing

[0029] A1. Set the user's language term set to S = {s t |t=-τ,...,-1,0,1,...,τ}, the online course (feature) set is L n = {l 1 ,l 2 ,l 3 ,...,l n}, the user (sample) set is U = {x1, x2, ..., x m}, n is the total number of online courses, and m is the total number of users;

[0030] For example, the user set U = {x1, x2, x3, x4, x5, x6} represents six users, and the online course set L n = {l 1 ,l 2 ,l 3 ,l 4} indicates four online courses;

[0031] A2. Collect user x from the transaction database m ∈U uses the language value s t Describe the online course n ∈L n The concept of language The language concept set is Form a user evaluation form for the course language;

[0032] If τ=1, the language term set S={s -1 = bad, s0 = average, s1 = good} is used to describe the user's language value preference information for the four courses, forming a user's course language evaluation table, see Table 1;

[0033] Table 1 User evaluation of course language

[0034]

[0035] A3. Collect and quantify user x from the transaction database i , x j ∈U social relationship is Sim ij ∈[0,1], construct the network adjacency matrix SM=(Sim ij ) n×n ;

[0036] The network adjacency matrix constructed in the embodiment of the present invention is as follows:

[0037]

[0038] All users of social networks Figure 2 ;

[0039] A4. Based on user set U and language concept set The user's course language evaluation form and the network adjacency matrix SM construct the network language form background, and use the following definition to obtain network language concept knowledge;

[0040] definition and The induced operator on : If any user x i ∈U and x j ∈U is related and but It is a network language concept knowledge;

[0041] Define user x i , x j If there is a relationship, then there is a path x between two users in the social network. i ,x k ,...,x l ,x j Sim ik ,...,Sim lj ≥λ, where λ is the given minimum relationship threshold;

[0042] The course recommendation network language form background of the embodiment of the present invention is shown in Table 2 and Table 3. When λ=0.5, the network language concept knowledge obtained is shown in Table 4.

[0043] Table 2 Course Recommendation Network Language Form Background

[0044]

[0045] Table 3 Course Recommendation Network Language Form Background

[0046]

[0047] Table 4 Network language concept knowledge SL 0.5

[0048]

[0049] B. Online Course Association Rules Mining

[0050] B1. Based on the network language background, determine the concept dyad (C s ,Cf ), C f It is C s The super concept of C s It is C f sub-concepts of;

[0051] Calculate all association rules using partial order relations between all concepts in

[0052] The embodiment of the present invention converts the network language concept knowledge SL 0.5 The conceptual partial order diagram obtained by adding the partial order relation is shown in Figure 3 ,Depend on Figure 3 The association rules calculated for the concept dyads shown are shown in Table 5;

[0053] Table 5 Association rules corresponding to concept tuples

[0054]

[0055] B2. Determine the support of association rules: Confidence of association rules: And by setting the minimum support and minimum confidence of association rules, the association rules with higher reliability can be screened out;

[0056] In the embodiment of the present invention, the minimum support is given as 0.17 and the minimum confidence is given as 0.33, and all association rules that meet the conditions are calculated to form an association rule base, as shown in Table 6.

[0057] Table 6 Association rule base

[0058]

[0059]

[0060] C. Generate and optimize recommendation sequences

[0061] C1. Use user neighborhood collaborative filtering to form a preliminary recommendation sequence, that is, use the network adjacency matrix SM to screen several users who are closest to the target user, and form a preliminary recommendation sequence with the courses watched by the closest users but not watched by the target user. The recommendation strength is the social network weight between the adjacent users and the target user;

[0062] according to Figure 2 The user social network shown in Figure 1 and the given minimum relationship threshold λ = 0.5 determine that the nearest users of user x5 are {x1, x2, x3, x6}, see Figure 4 The calculation process is as follows: The nearest user right Recommended for Nearest User right Recommended for The calculation of the nearest users x1 and x6 is similar. The initial recommendation sequence and the recommendation strength of the online course are obtained as {l 3 :0.8,l 2 :0.7}.

[0063] C2. Verify the rules and optimize the recommendation sequence, that is, use the historical course viewing records of the target user to perform association rule matching to optimize the preliminary recommendation sequence: for online courses in the rule consequences, if they have positive semantics, then their recommendation strength is increased in the preliminary recommendation sequence; if they have negative semantics, then their recommendation strength is weakened in the preliminary recommendation sequence; if they have neutral semantics, then their recommendation strength remains unchanged in the preliminary recommendation sequence. Finally, the optimized recommendation sequence makes Top-N recommendations to the target user according to the recommendation strength.

[0064] According to the history of x5 The matched association rule is and According to rule R1, user x5 can form l 2 The evaluation may be "average", 3 The evaluation may be "good"; for the recommended sequence {l 3 :0.8,l 2 :0.7} 2 and l 3 The recommendation strength optimization of course l 3 The recommendation strength increases, and the course l 2 The recommendation strength remains unchanged; the optimized recommendation sequence is {l 3 :1.0,l 2 :0.7}, the optimal online course x5 can be recommended as course l according to the recommendation strength 3 .

Claims

1. A hybrid online course recommendation method based on the network language background, characterized by Follow these steps: A. User behavior data collection and preprocessing A1. Set the user's language term set to S = {s t |t=-τ,...,-1,0,1,...,τ}, the set of online courses is L n = {l 1 ,l 2 ,l 3 ,...,l n }, the user set is U = {x1,x2,...,x m }, n is the total number of online courses, and m is the total number of users; A2. Collect user x m ∈U uses the language value s t Describe the online course n ∈L n The concept of language The language concept set is Form a user evaluation form for the course language; A3. Collect and quantify user x i , x j ∈U social relationship is Sim ij ∈[0,1], construct the network adjacency matrix SM=(Sim ij ) n×n ; A4. Based on user set U and language concept set The user's course language evaluation form and the network adjacency matrix SM construct the network language form background, and use the following definition to obtain network language concept knowledge; definition and The induced operator on : If any user x i ∈U and x j ∈U is related and but It is a network language concept knowledge; Define user x i , x j If there is a relationship, then there is a path x between two users in the social network. i ,x k ,...,x l ,x j Sim ik ,...,Sim lj ≥λ, where λ is the given minimum relationship threshold; B. Online Course Association Rules Mining B1. Based on the knowledge of network language concepts, determine the concept tuple (C s ,C f ), C f It is C s The super concept of C s It is C f sub-concepts of; Calculate all association rules using partial order relations between all concepts in B2. Determine the support of association rules: Confidence of association rules: By setting the minimum support and minimum confidence of association rules, the association rules with higher reliability are screened out to form an association rule base; C. Generate and optimize recommendation sequences C1. Use user neighborhood collaborative filtering to form a preliminary recommendation sequence, that is, use the network adjacency matrix SM to screen several users who are closest to the target user, and form a preliminary recommendation sequence with the courses watched by the closest users but not watched by the target user. The recommendation strength is the social network weight of the adjacent users and the target user; C2. Verify the rules and optimize the recommendation sequence, that is, use the historical course viewing records of the target user to perform association rule matching to optimize the preliminary recommendation sequence: for online courses in the rule consequences, if they have positive semantics, then their recommendation strength is increased in the preliminary recommendation sequence; if they have negative semantics, then their recommendation strength is weakened in the preliminary recommendation sequence; if they have neutral semantics, then their recommendation strength remains unchanged in the preliminary recommendation sequence; the final optimized recommendation sequence makes Top-N recommendations to the target user according to the recommendation strength.

Images