Institute student class-assisting teacher-student double-selection recommendation method and system oriented to personalized requirements

Abstract

The invention belongs to the field of educational informatization, and discloses a student-teacher-student double-selection recommendation method and system oriented to individual requirements and used for assisting in lesson of graduate students. Calculating the demand similarity of both teachers and students, and screening part of candidate lists; establishing a scoring matrix according to the matched historical data; the singular value decomposition scoring matrix is a product of three matrixes, namely a student feature matrix, a teacher feature matrix and a diagonal matrix; optimizing a student feature matrix and a teacher feature matrix; calculating a prediction score matrix according to the optimized student feature matrix and the optimized teacher feature matrix; and for each user, obtaining a teacher or a student with the highest predicted score from the objects in part of the candidate list through the predicted score matrix, and selecting the teachers with the top M predicted scores or the students with the top M predicted scores as a recommendation list. According to the method, the teacher-student matching accuracy can be improved, similarity calculation can be carried out by utilizing attribute information of newly added students or teachers, and content-based recommendation is carried out.

Description

Technical Field

[0001] The present invention relates to the field of educational informatization, and in particular to a method and system for recommending graduate assistants to teachers and students based on personalized needs. Background Art

[0002] In the field of educational informatization, teacher-student dual-selection systems have become an important bridge connecting students and teachers, aiming to improve teaching matching and promote the rational allocation of high-quality educational resources. However, existing teacher-student dual-selection recommendation methods are mostly based on simple rule matching or traditional collaborative filtering algorithms, which face many challenges in practical applications.

[0003] First, traditional teacher-student dual-selection recommendation methods often overlook the individual needs of both teachers and students, such as the compatibility between students' learning styles and interests and the teacher's teaching methods and expertise. This neglect can lead to inaccurate matching results and make it difficult to meet the individual needs of both teachers and students.

[0004] Secondly, data sparsity and cold start issues are common challenges faced by existing recommendation algorithms. For students or teachers who have just joined the system, due to the lack of sufficient historical data, traditional recommendation algorithms have difficulty making effective recommendations, which seriously affects the experience and satisfaction of new users.

[0005] Furthermore, the evaluation data from the teacher-student dual selection process is often influenced by a variety of factors, such as the teacher-student relationship and the course difficulty, which results in significant volatility and uncertainty. Traditional recommendation algorithms often perform poorly when processing this complex data, making it difficult to produce accurate and reliable recommendations. Summary of the Invention

[0006] In view of this, the present invention provides a dual-selection recommendation method and system for graduate teaching assistants based on personalized needs. By comprehensively considering various attribute information and historical behavior data of teachers and students, more accurate and personalized matching recommendations are achieved to solve the problems existing in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for recommending graduate teaching assistants to teachers and students based on personalized needs includes the following steps:

[0009] Step 1: Calculate the similarity of needs between teachers and students and screen a partial list of candidates;

[0010] Step 2: Create a scoring matrix based on the matched historical data;

[0011] Step 3, singular value decomposition scoring matrix is the product of three matrices, wherein the three matrices are respectively a student feature matrix, a teacher feature matrix and a diagonal matrix; using the stochastic gradient descent algorithm, the element parameters in the matrix are optimized by minimizing the root mean square error (RMSE) loss function to obtain an optimized student feature matrix and an optimized teacher feature matrix; and calculating the predicted scoring matrix based on the optimized student feature matrix and the optimized teacher feature matrix;

[0012] Step 4: Generate a recommendation list: For each user, use the predicted score matrix to obtain the teachers or students with the highest predicted scores from some of the objects in the candidate list, and select the top M teachers or students with the top M predicted scores as the recommendation list.

[0013] The similarity of the needs of teachers and students is calculated using a multi-attribute similarity recommendation algorithm based on weighted content, and a partial candidate list with a need similarity greater than 0.6 is screened out based on the need similarity result.

[0014] The similarity of teacher and student needs specifically includes the following steps:

[0015] Step 1-1: Establish a hierarchical relationship of evaluation indicators, including the goal layer and the criterion layer. The criterion layer is the specific evaluation indicators or factors for achieving the goal layer. For teachers, the elements of the criterion layer include their research direction, teaching ability, and academic background. For students, the elements of the criterion layer include research direction preferences, course requirements, desired supervisor type, and graduation year.

[0016] Step 1-2: Create a judgment matrix for the criterion layer, where rows and columns represent the elements of the criterion layer. Elements in the judgment matrix represent the relative importance of two elements, determined through questionnaires. By comparing the elements of the criterion layer pairwise, evaluate their importance to the target layer.

[0017] Steps 1-3: Calculate the eigenvalues and eigenvectors of the judgment matrix; obtain the maximum eigenvalue and corresponding eigenvector; the maximum eigenvalue is tested for consistency to ensure the reliability of the weights of the evaluation indicators or factors: if the consistency ratio CR of the judgment matrix is ≤ 0.1, it means that the weights are reasonable and the subsequent similarity calculation is valid;

[0018] Step 1-4: Use the cosine similarity algorithm to evaluate the similarity between the needs of teachers and students.

[0019] A scoring matrix is established based on the matching list of graduate teaching assistants and students in previous years, where the rows represent teacher-related information, the columns represent student-related information, and each element represents the demand similarity score between teacher-related information and student-related information.

[0020] In step 3, the score matrix is decomposed using singular value decomposition, which specifically includes the following steps:

[0021] Step 3-1: The matched historical data contains m teachers and n students. The rating matrix R is represented by an m×n matrix:

[0022]

[0023] Among them, r mn represents the score given by teacher m to student n;

[0024] Step 3-2: Singular value decomposition decomposes the rating matrix into two lower-dimensional matrix multiplications:

[0025] R=P T Q;

[0026] Among them, P is an m×f matrix, which represents the teacher feature matrix; Q is an n×f matrix, which represents the student feature matrix;

[0027] Step 3-3: User u’s predicted score for matching object i is expressed as:

[0028]

[0029] f represents the dimension of the feature vector, and F represents the total number of dimensions of the feature vector.

[0030] The optimization is specifically as follows:

[0031] By iteratively learning the initial teacher eigenvector matrix P and the initial student eigenvector matrix Q through minimizing the root mean square error RMSE loss function, the update direction is determined; the parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q are updated according to the direction of the partial derivatives through the stochastic gradient descent algorithm; the derivative of the minimization root mean square error RMSE loss function is obtained as follows:

[0032]

[0033] According to the principle of stochastic gradient descent algorithm, the iterative formula for obtaining the teacher eigenvector matrix P and the student eigenvector matrix Q is as follows:

[0034]

[0035] Among them, α represents the learning rate of the stochastic gradient descent algorithm; λ represents the regularization parameter; by adjusting the parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q, the optimized teacher eigenvector matrix P' and the optimized student eigenvector matrix Q' are obtained.

[0036] First, students choose a teacher based on the recommendation list, and then students who choose the same teacher are matched and ranked and recommended to that teacher.

[0037] A graduate teaching assistant teacher-student dual selection recommendation system oriented to personalized needs, including:

[0038] Similarity confirmation module: used to determine the weight of matching effect indicators using the hierarchical analysis algorithm and calculate the similarity of teacher and student needs based on the multi-attribute similarity recommendation algorithm of weighted content;

[0039] The similarity score of the needs of teachers and students is calculated by using the singular value decomposition and stochastic gradient descent algorithm on the score matrix to obtain the predicted score;

[0040] Recommendation list generation module: For each teacher, it uses the predicted rating matrix to obtain the students with the highest predicted ratings, and selects the top M students with the highest predicted ratings as the recommendation list.

[0041] The beneficial effects of the present invention are as follows: the present invention introduces the hierarchical analysis method based on the cosine similarity algorithm, comprehensively considering multiple dimensions such as the matching degree of research direction and the fit of course requirements; by constructing a two-level indicator system of "target layer-criterion layer", core indicators such as research direction, teaching ability, and academic background are incorporated into the decision-making framework, making the weight distribution more in line with the underlying logic of academic matching, and significantly improving the accuracy of teacher-student dual selection matching.

[0042] It can be seen from the above technical solutions that, compared with the existing technology, the present invention provides a dual-selection recommendation method and system for graduate teaching assistants based on personalized needs, which can not only improve the accuracy of teacher-student matching, but also help new teachers quickly integrate into the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of the personalized demand-oriented graduate assistant teacher-student dual-selection recommendation method of the present invention;

[0044] Figure 2 This is a structural diagram of the graduate assistant teacher-student dual-selection recommendation system for personalized needs of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] Specifically, the present invention adopts a content-based multi-attribute similarity calculation method, combines the basic information of teachers and students (such as research direction, teaching ability, academic background, course requirements, etc.) and historical behavior data (such as course requirements, teaching evaluation, graduation year, etc.) to make recommendations, adopts hierarchical analysis method to determine the weight of each indicator, establishes a content-based recommendation algorithm, and calculates the similarity of teacher and student needs; in order to solve the problem of sparsity of rating data, the traditional collaborative filtering algorithm is improved by introducing singular value matrix decomposition technology combined with stochastic gradient descent optimization learning model to achieve the purpose of predicting ratings, and recommends users according to the predicted rating ranking.

[0047] The teacher-student dual-selection recommendation method primarily relies on basic information provided by both the teacher and the student. Without collecting relevant information about new users, it's difficult to leverage historical behavior for recommendations. This is a common cold start problem in personalized matching systems. Newly registered teachers and students often register their attributes, such as their major, research area, teaching experience, and learning interests. This information can be used to calculate similarity and make content-based recommendations. By analyzing the needs of both teachers and students, we can identify highly compatible teachers or students and recommend them to the user. This not only improves the accuracy of teacher-student matching but also helps new teachers quickly integrate into the platform.

[0048] The embodiment of the present invention discloses a method for recommending graduate assistants to teachers and students based on individual needs. Figure 1 ,The hierarchical analysis algorithm is used to determine the weight of the matching ,effect index, and the multi-attribute similarity recommendation algorithm based on ,weighted content calculates the similarity of teacher and student needs;

[0049] A rating matrix is established based on historical data. Each element in the matrix represents a student's rating of a teacher or a teacher's rating of a student. The rating matrix is calculated using singular value decomposition and stochastic gradient descent algorithms to obtain predicted ratings.

[0050] For each teacher and student, the teacher and student with the highest predicted score is obtained through the predicted score matrix, and the top M teachers and students with the highest predicted scores are selected as the recommendation list.

[0051] Specifically, the present embodiment includes the following steps:

[0052] Step 1: Calculate the similarity of needs between teachers and students and screen a partial list of candidates;

[0053] Step 2: Create a scoring matrix based on the matched historical data;

[0054] Step 3: Singular value decomposition of the rating matrix is the product of three matrices, wherein the three matrices are respectively a student feature matrix, a teacher feature matrix, and a diagonal matrix; using the stochastic gradient descent algorithm, the element parameters in the matrix are optimized by minimizing the root mean square error (RMSE) loss function to obtain the optimized student feature matrix and the optimized teacher feature matrix, so that the data is more reasonable, and the predicted rating matrix is calculated;

[0055] Step 4: Generate a recommendation list: For each user, use the predicted score matrix to obtain the teachers or students with the highest predicted scores from some of the objects in the candidate list, and select the top M teachers or students with the top M predicted scores as the recommendation list.

[0056] The present invention adopts the hierarchical analysis method to determine the weights of the four attributes. Our goal is to calculate the weights of each factor in the criterion layer, so we set up two levels: the target layer and the criterion layer.

[0057] The specific steps are as follows:

[0058] The similarity of the needs of teachers and students is calculated using a multi-attribute similarity recommendation algorithm based on weighted content, and a partial candidate list with a need similarity greater than 0.6 is screened out based on the need similarity result.

[0059] The similarity of teacher and student needs specifically includes the following steps:

[0060] Step 1-1: Establish a hierarchical relationship of evaluation indicators, including the target layer and the criterion layer:

[0061] Target level: To achieve accurate matching and recommendation between teachers and students for graduate teaching assistant positions.

[0062] Standard layer:

[0063] Teacher evaluation indicators: research direction (C1), teaching ability (C2), academic background (C3);

[0064] Student evaluation indicators: research direction preference (S1), course requirements (S2), expected supervisor type (S3), and graduation year (S4).

[0065] Step 1-2: Establish a judgment matrix for the criterion layer, where the rows and columns represent the elements of the criterion layer. The elements in the judgment matrix represent the relative importance of two elements, which is determined through questionnaires. By comparing the elements of the criterion layer pairwise, the importance of each element to the target layer is assessed. This method uses a 1-9 scale (1 = equally important, 9 = absolutely important). The teacher criterion layer judgment matrix is shown in Table 1, and the student criterion layer judgment matrix is shown in Table 2:

[0066] Table 1 Teacher Criteria Level Judgment Matrix

[0067] <![CDATA[C1 (Research direction)]]> <![CDATA[C2 (Teaching ability)]]> <![CDATA[C3 (Academic background)]]> <![CDATA[C1]]> 1 3 2 <![CDATA[C2]]> 1 / 3 1 1 / 2 <![CDATA[C3]]> 1 / 2 2 1

[0068] Table 2 Student Criteria Level Judgment Matrix

[0069] <![CDATA[S1 (Research direction preference)]]> <![CDATA[S2 (Course Requirements)]]> <![CDATA[S3(Expected supervisor type)]]> <![CDATA[S4 (Graduation Year)]]> <![CDATA[S1]]> 1 5 9 9 <![CDATA[S2]]> 1 / 5 1 3 5 <![CDATA[S3]]> 1 / 9 1 / 3 1 3 <![CDATA[S4]]> 1 / 9 1 / 5 1 / 3 1

[0070] Steps 1-3: Calculate the eigenvalues and eigenvectors of the judgment matrix; obtain the maximum eigenvalue and corresponding eigenvector; the maximum eigenvalue is tested for consistency to ensure the reliability of the evaluation indicators or factor weights: if the consistency ratio (CR) of the judgment matrix is ≤ 0.1, it means that the weights are reasonable and the subsequent similarity calculation is valid;

[0071] Teacher's maximum eigenvalue: λ max1 =3.053

[0072] Teacher criterion weight: ω T =[0.54, 0.16, 0.30]

[0073] Student's largest eigenvalue: λ max2 =4.12

[0074] Student criterion weight: ω S =[0.58,0.10,0.04,0.07]

[0075] Use the largest eigenvalue λ max1 and λ max2 To calculate the consistency index (CI1 and CI2), the formula for calculating CI is:

[0076]

[0077] The calculated results are CI1=3.053, CI2=0.04;

[0078] The table shows RI1 = 0.58, RI2 = 0.90, which are then used to calculate the consistency ratio (CR):

[0079]

[0080] The calculation results are: CR1≈0.06<0.1, CR2≈0.04<0.1;

[0081] Since the values of CR1 and CR2 are less than 0.1, it can be concluded that the consistency of the judgment matrix is within an acceptable range. Therefore, it can be considered that the weight distribution of the teacher criterion layer judgment matrix and the student criterion layer judgment matrix is reasonable.

[0082] Step 1-4: Use the cosine similarity algorithm to evaluate the similarity between the needs of teachers and students.

[0083] Normalize teacher-student indicators (such as research direction correlation) to the interval [0,1];

[0084] The weighted similarity formula is used to calculate the similarity between the needs of teachers and students. The weighted similarity formula is as follows:

[0085]

[0086] where x Ti and x Si is the standardized value of teachers and students on the i-th indicator.

[0087] A scoring matrix is established based on the matching list of graduate teaching assistants and students in previous years, where the rows represent teacher-related information, the columns represent student-related information, and each element represents the demand similarity score between teacher-related information and student-related information.

[0088] In step 3, the score matrix is decomposed using singular value decomposition, which specifically includes the following steps:

[0089] Step 3-1: The matched historical data contains m = 50 teachers and n = 200 students. The rating matrix R is represented by a 50*200 matrix:

[0090]

[0091] Each value in the matrix represents the score given by teacher m to student n;

[0092] Step 3-2: Singular value decomposition decomposes the rating matrix into two lower-dimensional matrix multiplications:

[0093]

[0094] Where F = 10 is the latent feature dimension, P is a 50×f matrix, representing the teacher feature matrix; Q is a 200×f matrix, representing the student feature matrix;

[0095] Step 3-3: User u’s predicted score for matching object i is expressed as:

[0096]

[0097] f represents the dimension of the feature vector, and F represents the total number of dimensions of the feature vector.

[0098] The optimization is specifically as follows:

[0099] By iteratively learning the partial derivatives of the initial teacher eigenvector matrix P and the student eigenvector matrix Q by minimizing the root mean square error (RMSE) loss function, the update direction is determined. The parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q are updated according to the direction of the partial derivatives through the stochastic gradient descent algorithm.

[0100] Taking the derivative of the RMSE loss function to minimize the root mean square error, we get the following expression:

[0101]

[0102] According to the principle of stochastic gradient descent algorithm, the iterative formula for obtaining the teacher eigenvector matrix P and the student eigenvector matrix Q is as follows:

[0103]

[0104] Among them, α represents the learning rate of the stochastic gradient descent algorithm; λ represents the regularization parameter; by adjusting the parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q, the optimized teacher eigenvector matrix P' and the optimized student eigenvector matrix Q' are obtained.

[0105] First, students choose a teacher based on the recommendation list, and then students who choose the same teacher are matched and ranked and recommended to that teacher.

[0106] A graduate teaching assistant teacher-student dual selection recommendation system oriented to personalized needs, including:

[0107] Similarity confirmation module: used to determine the weight of matching effect indicators using the hierarchical analysis algorithm and calculate the similarity of teacher and student needs based on the multi-attribute similarity recommendation algorithm of weighted content;

[0108] Prediction score calculation module: used to build a score matrix based on historical data. Each element in the matrix represents the similarity score of the needs of teachers and students. The score matrix is calculated through singular value decomposition and stochastic gradient descent algorithm to obtain the predicted score.

[0109] Recommendation list generation module: For each teacher, it uses the predicted rating matrix to obtain the students with the highest predicted ratings, and selects the top M students with the highest predicted ratings as the recommendation list.

[0110] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0111] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is intended to be embodied in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for recommending graduate assistants to teachers and students based on personalized needs, characterized by: The following steps are involved: Step 1: Calculate the similarity of needs between teachers and students and screen a partial list of candidates; Step 2: Create a scoring matrix based on the matched historical data; Step 3, singular value decomposition scoring matrix is the product of three matrices, wherein the three matrices are respectively a student feature matrix, a teacher feature matrix and a diagonal matrix; using the stochastic gradient descent algorithm, the element parameters in the matrix are optimized by minimizing the root mean square error (RMSE) loss function to obtain an optimized student feature matrix and an optimized teacher feature matrix; and calculating the predicted scoring matrix based on the optimized student feature matrix and the optimized teacher feature matrix; Step 4: Generate a recommendation list: For each user, use the predicted score matrix to obtain the teachers or students with the highest predicted scores from some of the objects in the candidate list, and select the top M teachers or students with the top M predicted scores as the recommendation list.

2. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 1 is characterized in that: The similarity of the needs of teachers and students is calculated using a multi-attribute similarity recommendation algorithm based on weighted content, and a partial candidate list with a need similarity greater than 0.6 is screened out based on the need similarity result.

3. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 2 is characterized in that: The similarity of teacher and student needs specifically includes the following steps: Step 1-1: Establish a hierarchical relationship of evaluation indicators, including the goal layer and the criterion layer. The criterion layer is the specific evaluation indicators or factors for achieving the goal layer. For teachers, the elements of the criterion layer include their research direction, teaching ability, and academic background. For students, the elements of the criterion layer include research direction preferences, course requirements, desired supervisor type, and graduation year. Step 1-2: Create a judgment matrix for the criterion layer, where rows and columns represent the elements of the criterion layer. Elements in the judgment matrix represent the relative importance of two elements, determined through questionnaires. By comparing the elements of the criterion layer pairwise, evaluate their importance to the target layer. Steps 1-3: Calculate the eigenvalues and eigenvectors of the judgment matrix; obtain the maximum eigenvalue and corresponding eigenvector; the maximum eigenvalue is tested for consistency to ensure the reliability of the weights of the evaluation indicators or factors: if the consistency ratio CR of the judgment matrix is ≤ 0.1, it means that the weights are reasonable and the subsequent similarity calculation is valid; Step 1-4: Use the cosine similarity algorithm to evaluate the similarity between the needs of teachers and students.

4. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 1 is characterized in that: A scoring matrix is established based on the matching list of graduate teaching assistants and students in previous years, where the rows represent teacher-related information, the columns represent student-related information, and each element represents the demand similarity score between teacher-related information and student-related information.

5. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 1 is characterized in that: In step 3, the score matrix is decomposed using singular value decomposition, which specifically includes the following steps: Step 3-1: The matched historical data contains m teachers and n students. The rating matrix R is represented by an m×n matrix: Among them, r mn represents the score given by teacher m to student n; Step 3-2: Singular value decomposition decomposes the rating matrix into two lower-dimensional matrix multiplications: R=P T Q; Among them, P is an m×f matrix, which represents the teacher feature matrix; Q is an n×f matrix, which represents the student feature matrix; Step 3-3: The predicted score of user u for matching object i is expressed as: f represents the dimension of the feature vector, and F represents the total number of dimensions of the feature vector.

6. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 1 is characterized in that: The optimization is specifically as follows: By iteratively learning the initial teacher eigenvector matrix P and the initial student eigenvector matrix Q through minimizing the root mean square error RMSE loss function, the update direction is determined; the parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q are updated according to the direction of the partial derivatives through the stochastic gradient descent algorithm; the derivative of the minimization root mean square error RMSE loss function is obtained as follows: According to the principle of stochastic gradient descent algorithm, the iterative formula for obtaining the teacher eigenvector matrix P and the student eigenvector matrix Q is as follows: Among them, α represents the learning rate of the stochastic gradient descent algorithm; λ represents the regularization parameter; by adjusting the parameters of the teacher eigenvector matrix P and the student eigenvector matrix Q, the optimized teacher eigenvector matrix P' and the optimized student eigenvector matrix Q' are obtained.

7. The personalized demand-oriented graduate teaching assistant teacher-student dual selection recommendation method according to claim 1 is characterized in that: First, students choose a teacher based on the recommendation list, and then students who choose the same teacher are matched and ranked and recommended to that teacher.

8. A graduate assistant teacher-student dual selection recommendation system oriented to personalized needs, characterized by: include: Similarity confirmation module: used to determine the weight of matching effect indicators using the hierarchical analysis algorithm and calculate the similarity of teacher and student needs based on the multi-attribute similarity recommendation algorithm of weighted content; The similarity score of the needs of teachers and students is calculated by using the singular value decomposition and stochastic gradient descent algorithm on the score matrix to obtain the predicted score; Recommendation list generation module: For each teacher, it uses the predicted rating matrix to obtain the students with the highest predicted ratings, and selects the top M students with the highest predicted ratings as the recommendation list.

Images