Analysis and recommendation method of tutoring service based on big data information processing

Abstract

The present invention relates to the technical field of training service recommendation, and discloses a training service analysis and recommendation method based on big data information processing, comprising: collecting first feature data of a current user and second feature data of N coaches; preprocessing the first feature data and the second feature data respectively to obtain third feature data and fourth feature data; constructing graph structure data based on the third feature data and the fourth feature data, the graph structure data including: nodes, initial features of nodes and edges between nodes; constructing and training a training service recommendation model, wherein the input of the training service recommendation model is the graph structure data, and the output value represents the matching value between the current user and the N coaches; sorting the matching values ​​between the current user and the N coaches in descending order, and obtaining the top M coaches as a training recommendation list.

Description

Technical Field

[0001] The present invention relates to the technical field of training service recommendation, and more specifically, to a training service analysis and recommendation method based on big data information processing. Background Art

[0002] With the gradual improvement of people's health awareness, the importance of sports has become increasingly prominent. In the face of the current situation of lack of professional guidance and companionship, some people may fail to achieve the expected results when participating in sports due to improper methods, and may even suffer injuries. Therefore, providing professional and high-quality sports training services has become an important way and means to meet the public's health needs.

[0003] However, it is not easy to make a suitable choice among the numerous coaching services. Users often face the problem of information asymmetry during the selection process. For example, the coach's teaching qualifications and service quality are not transparent, which may lead to the selected coach not matching their own conditions.

[0004] Therefore, there is an urgent need for a training service analysis and recommendation method to solve the above problems. Summary of the Invention

[0005] The present invention provides a method for analyzing and recommending training services based on big data information processing, which solves the technical problems in related technologies such as mismatched training partners and low service quality caused by information asymmetry when users select training partners.

[0006] The present invention provides a training service analysis and recommendation method based on big data information processing, comprising: step S101, collecting first feature data of the current user and second feature data of N coaches, where N is a custom parameter and is set according to the number of coaches;

[0007] The first characteristic data includes: age, gender, occupation, sports preference and injury status;

[0008] The second characteristic data includes: age, gender, years of work experience, sports expertise, number of professional certificates, and service ratings;

[0009] Step S102, pre-processing the first feature data and the second feature data to obtain third feature data and fourth feature data;

[0010] Step S103, constructing graph structure data according to the third feature data and the fourth feature data;

[0011] Graph structure data includes: nodes, initial features of nodes, and edges between nodes;

[0012] The nodes include: a first node and a second node;

[0013] The first node establishes a mapping relationship with the current user;

[0014] The second node establishes a mapping relationship with the coach;

[0015] The initial feature of the first node is represented by the third feature data of the current user that establishes a mapping relationship with the first node;

[0016] The initial feature of the second node is represented by the fourth feature data of the coach that establishes a mapping relationship with the second node;

[0017] Step S104, constructing and training a training service recommendation model;

[0018] The input of the coaching service recommendation model is graph structure data, and the output value represents the matching value between the current user and N coaches;

[0019] Step S105: sort the matching values ​​between the current user and the N coaches in descending order, and obtain the first M coaches as a training partner recommendation list, where M is a custom parameter.

[0020] Furthermore, the pre-processing steps include:

[0021] Step S201, filling missing values ​​in the first feature data and the second feature data by using the most frequently occurring value therein;

[0022] Step S202, performing replacement processing on the outliers in the first feature data and the second feature data by calculating the mean of all non-outliers;

[0023] Step S203 : normalizing the first feature data and the second feature data using a z-score normalization method.

[0024] Furthermore, gender includes: male and female, and gender is represented by real number coding; occupation includes: doctor, teacher, engineer, student and service personnel, and occupation is represented by one-hot coding; sports preference and sports specialty are the same feature, including: basketball, billiards, running, fitness and table tennis, represented by one-hot coding; injury status includes: with injury and without injury, and injury status is represented by real number coding; service rating is represented by real number coding, and is represented by the rating of coaches by users collected by the platform.

[0025] Furthermore, the ways to construct edges include:

[0026] Build an edge between the first node and the second node, including:

[0027] When a user follows a coach, a one-way edge is created between the user and the coach, that is, from the user to the coach;

[0028] When the user's sport preference is the same as the coach's sport specialty, a bidirectional edge is constructed between the user and the coach, and the weight is represented by the same number in the sport preference and the sport specialty;

[0029] When a user has made an appointment or taken a course with the instructor, a one-way edge is created between the user and the instructor, that is, from the user to the instructor, and the weight is represented by the number of appointments or lessons;

[0030] When a user interacts with a coach, a bidirectional edge is created between them. The weight is represented by the number of interactions. The types of interactions include comments, likes, and private messages.

[0031] Build an edge between the second node and the second node, including:

[0032] When different coaches have the same sports expertise, a bidirectional edge is constructed between the coaches, and the weight is represented by the same quantity in the sports expertise.

[0033] Furthermore, the training service recommendation model includes: a first hidden layer, a second hidden layer, a third hidden layer and an extraction layer;

[0034] The first hidden layer inputs the initial feature of the first node of the graph structure data and outputs the first updated feature;

[0035] The second hidden layer includes N first units, the nth first unit inputs the initial feature of the nth second node of the graph structure data, and outputs the second updated feature, where 1≤n≤N;

[0036] N first units share weight parameters and bias parameters;

[0037] The first updated feature and the second updated feature have the same size;

[0038] The third hidden layer inputs graph structure data and outputs a graph update matrix, where each row vector of the graph update matrix corresponds to a third updated feature of the first node or the second node;

[0039] The extraction layer includes N extractors, the nth extractor is used to extract the third updated feature corresponding to the nth second node and input it into the first classifier, and the classification space of the first classifier represents the matching value between the current user and the nth coach.

[0040] Furthermore, the first hidden layer and the second hidden layer are used to unify the dimensions of the initial features of the first node and the second node.

[0041] Furthermore, the calculation formula of the first hidden layer is:

[0042] ;

[0043] in represents the first updated feature of the first hidden layer output, represents the initial features of the first node of the first hidden layer input, 、 、 and are the first weight parameter, the second weight parameter, the first bias parameter, and the second bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

[0044] Furthermore, the calculation formula of the nth first unit of the second hidden layer is:

[0045] ;

[0046] in represents the second updated feature output by the nth first unit of the second hidden layer, represents the initial feature of the second node of the nth first unit input of the second hidden layer, 、 、 and are the third weight parameter, the fourth weight parameter, the third bias parameter, and the fourth bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

[0047] Furthermore, the calculation formula of the third hidden layer is:

[0048] ;

[0049] ;

[0050] ;

[0051] ;

[0052] Where S represents the graph update matrix output by the third hidden layer, represents the third updated feature of node i, and Both represent the update characteristics of the node, represents the normalized attention weight between node i and its adjacent node j, represents the set of nodes adjacent to node i, ReLU represents the ReLU activation function, represents the attention coefficient between node i and its adjacent node j, represents the attention coefficient between node i and its adjacent node k, LeakyReLU represents the LeakyReLU activation function, exp represents the exponential function, T represents the transposition operation, Represents a splicing operation, 、 and They represent the fifth weight parameter, the sixth weight parameter and the seventh weight parameter respectively. The update characteristics of the above nodes include: the first update characteristic of the first node and the second update characteristic of the second node.

[0053] Furthermore, it should be noted that the weight parameters and bias parameters in the accompanying service recommendation model are all learnable parameters. During the training process of the accompanying service recommendation model, the gradient information of each weight parameter and bias parameter is calculated through the loss function and the chain rule, and the weight parameters and bias parameters in the accompanying service recommendation model are updated according to the gradient information through back propagation and gradient descent algorithm.

[0054] Furthermore, the loss function of the tutoring service recommendation model is the cross entropy loss function.

[0055] Furthermore, sample labels of training samples used to train the accompanying training service recommendation model are obtained through manual labeling.

[0056] The beneficial effect of the present invention is that it constructs a training service recommendation model to analyze the interaction between users and coaches, as well as the relationship between coaches, to obtain the matching value of each coach for the user, and then recommend high-quality and highly matched training services to users, thereby improving user satisfaction.

[0057] Figure 1 is a flow chart of the method for analyzing and recommending training services based on big data information processing of the present invention;

[0058] Figure 2 is a flow chart of preprocessing characteristic data of users and coaches according to the present invention;

[0059] Figure 3 It is a schematic diagram of the method of constructing edges in the graph structure data of the present invention. DETAILED DESCRIPTION

[0060] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0061] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, but do not exclude other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0062] like Figures 1 to 3 As shown, the training service analysis and recommendation method based on big data information processing includes: step S101, collecting the first characteristic data of the current user and the second characteristic data of N coaches, where N is a custom parameter and N is set according to the number of coaches;

[0063] The first characteristic data includes: age, gender, occupation, sports preference and injury status;

[0064] The second characteristic data includes: age, gender, years of work experience, sports expertise, number of professional certificates, and service ratings;

[0065] Step S102, pre-processing the first feature data and the second feature data to obtain third feature data and fourth feature data;

[0066] Step S103, constructing graph structure data according to the third feature data and the fourth feature data;

[0067] Graph structure data includes: nodes, initial features of nodes, and edges between nodes;

[0068] The nodes include: a first node and a second node;

[0069] The first node establishes a mapping relationship with the current user;

[0070] The second node establishes a mapping relationship with the coach;

[0071] The initial feature of the first node is represented by the third feature data of the current user that establishes a mapping relationship with the first node;

[0072] The initial feature of the second node is represented by the fourth feature data of the coach that establishes a mapping relationship with the second node;

[0073] Step S104, constructing and training a training service recommendation model;

[0074] The input of the coaching service recommendation model is graph structure data, and the output value represents the matching value between the current user and N coaches;

[0075] Step S105: sort the matching values ​​between the current user and the N coaches in descending order, and obtain the first M coaches as a recommended list for sparring partners, where M is a custom parameter. Preferably, M is set to 10.

[0076] In one embodiment of the present invention, the pre-processing step includes:

[0077] Step S201, filling missing values ​​in the first feature data and the second feature data by using the most frequently occurring value therein;

[0078] Step S202: replace the outliers in the first feature data and the second feature data by calculating the mean of all non-outliers. For example, if the age exceeds a reasonable range, is a negative number, or exceeds 100 years old, replace it with the mean of the remaining ages in the first feature data.

[0079] Step S203 : normalizing the first feature data and the second feature data using a z-score normalization method.

[0080] In one embodiment of the present invention, gender includes: male and female, and gender is represented by real number coding, for example, male is represented by 1; occupation includes: doctor, teacher, engineer, student and service industry personnel, and occupation is represented by one-hot coding, for example, teacher is represented by [0, 1, 0, 0, 0]; sports preference and sports specialty are the same feature, including: basketball, billiards, running, fitness and table tennis, and are represented by one-hot coding, for example, running is represented by [0, 0, 1, 0, 0], and basketball and billiards are represented by [1, 1, 0, 0, 0]; injury status includes: injury and no injury, and injury status is represented by real number coding, for example, injury status is represented by 1; service rating is represented by real number coding, and is represented by the rating of coaches by users collected by the platform.

[0081] In one embodiment of the present invention, the method of constructing an edge includes:

[0082] Build an edge between the first node and the second node, including:

[0083] When a user follows a coach, a one-way edge is constructed between the user and the coach, that is, from the user to the coach. Preferably, the weight is set to 1;

[0084] When the user's sports preference is the same as the coach's sports specialty, a bidirectional edge is constructed between the user and the coach, and the weight is represented by the same number in the sports preference and sports specialty. For example, if the user's sports preference is basketball and billiards, and the coach's sports specialties are billiards and running, and the user's and the coach's sports preferences and sports specialties both include billiards, then an edge is constructed between the user and the coach, and the weight is set to 1;

[0085] When a user has made an appointment or taken a course with the instructor, a one-way edge is created between the user and the instructor, that is, from the user to the instructor, and the weight is represented by the number of appointments or lessons;

[0086] When a user interacts with a coach, a bidirectional edge is created between them. The weight is expressed by the number of interactions, including comments, likes, and private messages. Preferably, if the number of interactions is 1-3, the weight is 1; if the number of interactions is 4-10, the weight is 3; and if the number of interactions is more than 10, the weight is 5. For example, if the current user has privately messaged the coach 3 times, an edge is created between the current user and the coach, and the weight is set to 1.

[0087] Build an edge between the second node and the second node, including:

[0088] When different coaches have the same sports expertise, a bidirectional edge is constructed between the coaches, and the weight is represented by the same quantity in the sports expertise.

[0089] In one embodiment of the present invention, the training service recommendation model includes: a first hidden layer, a second hidden layer, a third hidden layer and an extraction layer;

[0090] The first hidden layer inputs the initial feature of the first node of the graph structure data and outputs the first updated feature;

[0091] The second hidden layer includes N first units, the nth first unit inputs the initial feature of the nth second node of the graph structure data, and outputs the second updated feature, where 1≤n≤N;

[0092] N first units share weight parameters and bias parameters;

[0093] The first updated feature and the second updated feature have the same size;

[0094] The third hidden layer inputs graph structure data and outputs a graph update matrix, where each row vector of the graph update matrix corresponds to a third updated feature of the first node or the second node;

[0095] The extraction layer includes N extractors, the nth extractor is used to extract the third updated feature corresponding to the nth second node and input it into the first classifier, and the classification space of the first classifier represents the matching value between the current user and the nth coach.

[0096] The first hidden layer and the second hidden layer are used to unify the dimensions of the initial features of the first node and the second node.

[0097] In one embodiment of the present invention, the calculation formula of the first hidden layer is: in represents the first updated feature of the first hidden layer output, represents the initial features of the first node of the first hidden layer input, 、 、 and are the first weight parameter, the second weight parameter, the first bias parameter, and the second bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function. In one embodiment of the present invention, the calculation formula of the nth first unit of the second hidden layer is:

[0098] ;

[0099] in represents the second updated feature output by the nth first unit of the second hidden layer, represents the initial characteristics of the second node of the nth first unit input, 、 、 and are the third weight parameter, the fourth weight parameter, the third bias parameter, and the fourth bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

[0100] It should be noted that the first node of the first hidden layer input The initial feature is a 1*5 feature vector, and the first weight parameter The first feature matrix is ​​5*16, and the second weight parameter is The second feature matrix is ​​16*32, and the first updated feature is obtained by calculation is a 1*32 feature vector; the second node of the nth first unit input of the second hidden layer The initial feature is a 1*6 feature vector, and the third weight parameter is the third characteristic matrix of 6*16, and the fourth weight parameter The fourth characteristic matrix of 16*32 is calculated to obtain the second updated characteristic of the nth first unit output of the second hidden layer It is a 1*32 feature vector.

[0101] In one embodiment of the present invention, the calculation formula of the third hidden layer is:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] Where S represents the graph update matrix output by the third hidden layer, represents the third updated feature of node i, and Both represent the update characteristics of the node, represents the normalized attention weight between node i and its adjacent node j, represents the set of nodes adjacent to node i, ReLU represents the ReLU activation function, represents the attention coefficient between node i and its adjacent node j, represents the attention coefficient between node i and its adjacent node k, LeakyReLU represents the LeakyReLU activation function, exp represents the exponential function, T represents the transposition operation, Represents a splicing operation, 、 and represents the fifth weight parameter, the sixth weight parameter, and the seventh weight parameter, respectively. The update features of the above-mentioned nodes include: the first updated feature of the first node and the second updated feature of the second node. It should be noted that the weight parameters and bias parameters in the accompanying training service recommendation model are all learnable parameters. During the training process of the accompanying training service recommendation model, the gradient information of each weight parameter and bias parameter is calculated using the loss function and the chain rule. The weight parameters and bias parameters in the accompanying training service recommendation model are updated based on the gradient information through backpropagation and gradient descent algorithms.

[0107] In one embodiment of the present invention, the loss function of the tutoring service recommendation model is a cross entropy loss function.

[0108] In one embodiment of the present invention, sample labels of training samples used to train the accompanying training service recommendation model are obtained by manual labeling.

[0109] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A training service analysis and recommendation method based on big data information processing, characterized in that: include: Step S101, collecting the first characteristic data of the current user and the second characteristic data of N coaches, where N is a custom parameter and is set according to the number of coaches; The first characteristic data includes: age, gender, occupation, sports preference and injury status; The second characteristic data includes: age, gender, years of work experience, sports expertise, number of professional certificates, and service ratings; Step S102, pre-processing the first feature data and the second feature data to obtain third feature data and fourth feature data; Step S103, constructing graph structure data according to the third feature data and the fourth feature data; Graph structure data includes: nodes, initial features of nodes, and edges between nodes; The nodes include: a first node and a second node; The first node establishes a mapping relationship with the current user; The second node establishes a mapping relationship with the coach; The initial feature of the first node is represented by the third feature data of the current user that establishes a mapping relationship with the first node; The initial feature of the second node is represented by the fourth feature data of the coach that establishes a mapping relationship with the second node; The methods for constructing edges include: Build an edge between the first node and the second node, including: When a user follows a coach, a one-way edge is created between the user and the coach, that is, from the user to the coach; When the user's sport preference is the same as the coach's sport specialty, a bidirectional edge is constructed between the user and the coach, and the weight is represented by the same number in the sport preference and the sport specialty; When a user has made an appointment or taken a course with the instructor, a one-way edge is created between the user and the instructor, that is, from the user to the instructor, and the weight is represented by the number of appointments or lessons; When a user interacts with a coach, a bidirectional edge is created between them. The weight is represented by the number of interactions. The types of interactions include comments, likes, and private messages. Build an edge between the second node and the second node, including: When different coaches have the same sports expertise, a bidirectional edge is constructed between the coaches, and the weight is represented by the same quantity in the sports expertise; Step S104, constructing and training a training service recommendation model; The input of the coaching service recommendation model is graph structure data, and the output value represents the matching value between the current user and N coaches; The training service recommendation model includes: a first hidden layer, a second hidden layer, a third hidden layer and an extraction layer; The first hidden layer inputs the initial feature of the first node of the graph structure data and outputs the first updated feature; The second hidden layer includes N first units, the nth first unit inputs the initial feature of the nth second node of the graph structure data, and outputs the second updated feature, where 1≤n≤N; N first units share weight parameters and bias parameters; The first updated feature and the second updated feature have the same size; The third hidden layer inputs graph structure data and outputs a graph update matrix, where each row vector of the graph update matrix corresponds to a third updated feature of the first node or the second node; The calculation formula for the third hidden layer is: Where S represents the graph update matrix output by the third hidden layer, represents the third updated feature of node i, and Both represent the update characteristics of the node, represents the normalized attention weight between node i and its adjacent node j, represents the set of nodes adjacent to node i, ReLU represents the ReLU activation function, represents the attention coefficient between node i and its adjacent node j, represents the attention coefficient between node i and its adjacent node k, LeakyReLU represents the LeakyReLU activation function, exp represents the exponential function, T represents the transposition operation, Represents a splicing operation, 、 and represent a fifth weight parameter, a sixth weight parameter, and a seventh weight parameter, respectively, and the update characteristics of the above-mentioned nodes include: a first update characteristic of the first node and a second update characteristic of the second node; represents a stacking operation, which means stacking N third updated features to form a matrix; The extraction layer includes N extractors, the nth extractor is used to extract the third updated feature corresponding to the nth second node and input it into the first classifier, and the classification space of the first classifier represents the matching value between the current user and the nth coach; Step S105: sort the matching values ​​between the current user and the N coaches in descending order, and obtain the first M coaches as a training partner recommendation list, where M is a custom parameter.

2. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: The preprocessing steps include: Step S201, filling missing values ​​in the first feature data and the second feature data by using the most frequently occurring value therein; Step S202, performing replacement processing on the outliers in the first feature data and the second feature data by calculating the mean of all non-outliers; Step S203 : normalizing the first feature data and the second feature data using a z-score normalization method.

3. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: Gender includes: Male and female, gender is represented by real number coding; Occupations include: doctors, teachers, engineers, students and service workers, and occupations are represented by one-hot coding; sports preferences and sports specialties are the same feature, including: basketball, billiards, running, fitness and table tennis, represented by one-hot coding; injury conditions include: with injury and without injury, and injury conditions are represented by real number coding; service ratings are represented by real number coding, which is represented by the user ratings of coaches collected by the platform.

4. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: The calculation formula for the first hidden layer is: ;in represents the first updated feature of the first hidden layer output, represents the initial features of the first node of the first hidden layer input, 、 、 and are the first weight parameter, the second weight parameter, the first bias parameter, and the second bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

5. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: The calculation formula for the nth first unit of the second hidden layer is: ;in represents the second updated feature output by the nth first unit of the second hidden layer, represents the initial feature of the second node of the nth first unit input of the second hidden layer, 、 、 and are the third weight parameter, the fourth weight parameter, the third bias parameter, and the fourth bias parameter, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

6. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: The loss function of the tutoring service recommendation model is the cross entropy loss function.

7. The method for analyzing and recommending training services based on big data information processing according to claim 1, characterized in that: The sample labels of the training samples used to train the accompanying service recommendation model are obtained through manual labeling.

Images