KR102636868B1 - Mentor-mentee group matching system and mentor-mentee group matching method using machine learning - Google Patents

Abstract

The mentor-mentee group matching system using machine learning of the present invention applies the first method of unsupervised learning to the vector of the user's keyword of interest and the keyword vectors of the mentors and mentees to provide information about the user. a first grouping unit that performs first group matching of mentors and mentees; And after mentoring through the first group matching, a second group matching of mentor-mentees for the user is performed by applying a second method of unsupervised learning based on the mutual satisfaction of each mentor and each mentee. and a second grouping unit that performs the second method, and the second method is different from the first method. Through this, the present invention provides a mentor-mentee group matching system using machine learning for optimal mentor-mentee group matching.

Description

Mentor-mentee group matching system and mentor-mentee group matching method using machine learning {MENTOR-MENTEE GROUP MATCHING SYSTEM AND MENTOR-MENTEE GROUP MATCHING METHOD USING MACHINE LEARNING}

The present invention relates to a mentor-mentee group matching system, and more specifically, to a mentor-mentee group matching system and a mentor-mentee group matching method using various methods of machine learning.

A mentor is a person who helps a mentee in various aspects. A mentor is a person who cares for, trusts, and encourages the mentee. A good mentor wants to be around their mentee, is experienced, and likes to help their mentee succeed in life.

A mentee is a person who develops and develops his or her own capabilities with the help of a mentor. In college, a mentee is a learner who lacks basic knowledge of their major and seeks help from a mentor to develop learning skills to supplement this, adapt to college life, and obtain information about career paths and employment.

Mentoring refers to activities in which a mentor influences a mentee. The type of mentoring is divided into 1:1 mentoring, peer mentoring, and group mentoring depending on how the relationship between mentor and mentee is formed. 1:1 mentoring refers to a relationship where an experienced mentor teaches one-on-one to inexperienced people who are in a stage of learning or transition. Peer mentoring (or group study) refers to a relationship in which colleagues of similar level support and guide one another. Group mentoring is a form in which several mentees work together under one or more experienced mentors for a specific purpose. The advantage is that we can exchange ideas and information and receive feedback as a group.

A system and method for optimal grouping of mentors and mentees is required.

Registered Patent Publication 10-2408115 Public Patent Publication 10-2022-0077790

The present invention is intended to solve the above problems, and the purpose of the present invention is to provide an optimal mentor-mentee group matching system and a mentor-mentee group matching method using machine learning.

The above and other objects and advantages of the present invention will become apparent from the following description of preferred embodiments.

The above purpose is to provide a mentor-mentee group matching system and a mentor-mentee group matching method using machine learning.

The mentor-mentee group matching system using machine learning according to an embodiment of the present invention,

A first grouping unit that performs a first group matching of mentors and mentees for the user by applying a first method of unsupervised learning to the vector of the user's keywords of interest and the keyword vectors of the mentors and mentees. ; and

After performing mentoring through the first group matching, a second group matching of mentor-mentees for the user is performed by applying the second method of unsupervised learning based on the mutual satisfaction of each mentor and each mentee. It includes a second grouping unit,

The second method is different from the first method, where each of the mentors and mentees has one keyword vector, and the keyword vector is generated by a word embedding method.

Preferably,

After carrying out mentoring through the second group matching, a third method of machine learning is applied based on the mutual satisfaction of each mentor and each mentee to perform a new group matching of mentor-mentee for the user, or

A third grouping unit that performs a new group matching of mentor-mentees for the user by applying the third method to the user's new interest keyword vector,

The third method is different from the first method and the second method.

Preferably,

The first method is GMM (Gaussian Mixture Model)-based soft-clustering,

The second method is performed using collaborative filtering.

Preferably,

After performing mentoring through the second group matching, the first grouping unit further merges keyword vectors of mentors and fellow mentees who participated in the second group matching, and applies the GMM (Gaussian Mixture Model) to the merged keyword vectors. The first group matching is performed by applying soft-clustering.

Preferably,

The third method is performed using a Recurrent Neural Network (RNN) model or a general RNN model, and the third method is performed based on the expected satisfaction obtained as a result of the collaborative filtering in addition to the mutual satisfaction. do.

Preferably,

The third grouping unit matches a new group of mentor-mentees for a new user by applying the global RNN (Recurrent Neural Network) based on the mutual satisfaction and the expected satisfaction for each mentor and each mentee each time mentoring is performed. or

The RNN (Recurrent Neural Network) is applied to the user's new keyword vector of interest to perform a new group matching of mentor-mentee for the user.

Preferably,

The second grouping unit, when the total expected satisfaction level for each member (mentor + mentee) of the mentoring group that has previously conducted mentoring, is higher than a predefined standard, a third user (the third user is in the mentoring group that has previously conducted mentoring) We recommend the content created through the mentoring process mentioned above to those who are not affiliated with the organization.

The mentor-mentee group matching method using machine learning according to an embodiment of the present invention,

Performing the first group matching of mentor-mentees for the user by applying GMM (Gaussian Mixture Model)-based soft-clustering to the vector of the user's keywords of interest and the keyword vectors of the mentors and mentees. step; and

After performing mentoring through the first group matching, performing a second group matching of mentor-mentees for the user by applying collaborative filtering based on mutual satisfaction with each mentor and each mentee; and

Mentor-mentee for new users by applying a global RNN (Recurrent Neural Network) based on the mutual satisfaction of each mentor and each mentee each time mentoring is performed and the expected satisfaction obtained as a result of the collaborative filtering. A group matching upgrade step of performing new group matching, or performing new group matching of mentor-mentee for the user by applying RNN (Recurrent Neural Network) to the new interest keyword vector of the user,

Each of the mentors and mentees has one keyword vector, and the keyword vector is generated by a word embedding method.

By using the mentor-mentee group matching system and the mentor-mentee group matching method using machine learning according to the present invention, group matching is continuously upgraded each time mentoring is performed or the user's keyword of interest changes. It works.

However, the effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

In order to more fully understand the drawings cited in the detailed description of the present invention, a brief description of each drawing is provided.
Figure 1 is a schematic diagram of a mentor-mentee group matching system using machine learning of the present invention.
Figure 2 is a diagram showing an operation flowchart of the first grouping unit of the mentor-mentee group matching system using machine learning of the present invention.
Figure 3 is a diagram showing an operation flowchart of the second grouping unit of the mentor-mentee group matching system using machine learning of the present invention.
Figure 4 is a diagram showing an operation flowchart of the third grouping unit of the mentor-mentee group matching system using machine learning of the present invention.
Figure 5 is a diagram showing a flowchart of the mentor-mentee group matching method using machine learning of the present invention.

Hereinafter, the present invention will be described in detail with reference to embodiments of the present invention and drawings. These examples are merely presented as examples to explain the present invention in more detail, and it will be apparent to those skilled in the art that the scope of the present invention is not limited by these examples. .

Additionally, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains, and in case of conflict, this specification including definitions The description will take precedence.

In order to clearly explain the proposed invention in the drawings, parts unrelated to the description have been omitted, and similar reference numerals have been assigned to similar parts throughout the specification. And, when it is said that a part "includes" a certain component, this means that it does not exclude other components, but may further include other components, unless specifically stated to the contrary. Additionally, “unit” as used in the specification refers to a unit or block that performs a specific function.

Identification codes (first, second, etc.) for each step are used for convenience of explanation. The identification codes do not describe the order of each step, and each step does not clearly state a specific order in context. It may be carried out differently from the order specified above. That is, each step may be performed in the same order as specified, may be performed substantially simultaneously, or may be performed in the opposite order.

Figure 1 is a schematic diagram of a mentor-mentee group matching system using machine learning of the present invention.

The mentor-mentee group matching system using machine learning of the present invention includes a first grouping unit 110 and a second grouping unit 130.

The first grouping unit 110 applies soft-clustering based on the Gaussian Mixture Model (GMM) of unsupervised learning to the vector of the user's keywords of interest and the keyword vectors of mentors and mentees. Thus, the first group matching of mentor-mentee for the user is performed.

Mentors and mentees each have one keyword vector, and the keyword vector is generated by a word embedding method. During soft-clustering, the keyword vectors of all users, including mentors and mentees, are mapped to the Semantic Space, and each user is clustered within this space.

Keyword vectors are created using a word embedding method using a neural network. In the field of natural language processing, embedding refers to the result of converting natural language used by humans into a vector, a numeric form that machines can understand, or the entire series of processes. The simplest form of embedding is to use the word frequency as a vector. In a term-document matrix, rows and columns correspond to documents.

A word-document matrix is an example of the simplest form of embedding.

Gaussian Mixture Model (GMM), as its name suggests, is a clustering algorithm that mixes multiple Gaussian distributions. The basic idea of GMM is to express the complex probability distribution that exists in reality by mixing K Gaussian distributions. At this time, K must be set directly by the person who wants to analyze the data.

Gaussian Mixture Model (GMM) is widely used for unsupervised learning (clustering) in machine learning.

Detailed technical descriptions of the various methods of machine learning in the present invention are omitted since they are obvious and widely known to those skilled in the art.

Merge the vector of the user's keyword of interest with the keyword vectors of mentors and mentees, and apply GMM (Gaussian Mixture Model)-based soft-clustering to the merged keyword vectors to perform first group matching. You can.

The second grouping unit 130 performs mentoring through first group matching and then applies collaborative filtering of unsupervised learning based on the mutual satisfaction of each mentor and each mentee to provide a mentor to the user. -Perform second group matching of mentees.

After performing mentoring through second group matching, the first grouping unit 110 additionally merges the keyword vectors of the mentors and fellow mentees who participated in the second group matching, and performs a GMM (Gaussian Mixture Model) on the merged keyword vectors. First group matching can be performed by applying soft-clustering.

The user's keyword vector is updated by adding and merging keyword vectors after mentoring. Even before conducting mentoring, simply by saving the user of interest, the keyword vectors of the users of interest can be merged and the user's keyword vector can be updated.

The mentor-mentee group matching system using machine learning of the present invention may further include a third grouping unit (not shown).

The third grouping department performs mentoring through second group matching and then applies machine learning's RNN (Recurrent Neural Network) based on the mutual satisfaction of each mentor and each mentee to perform a new group matching of mentor-mentee for the user. , or a third method is applied to the user's new interest keyword vector to perform a new group matching of mentor-mentee for the user.

Each time mentoring is performed, the third grouping unit applies RNN (Recurrent Neural Network) based on the mutual satisfaction of each mentor and each mentee to match a new group of mentors and mentees for the user, or to match the user's new interest. A new group matching of mentors and mentees for users is performed by applying RNN (Recurrent Neural Network) to keyword vectors.

Figure 2 is a diagram showing an operation flowchart of the first grouping unit of the mentor-mentee group matching system using machine learning of the present invention.

The first group matching of the first grouping unit includes generating a keyword vector (S210).

The user's keywords are extracted from the interest keywords and career plan of the user's profile card. However, weight is given to keywords of interest. Keyword vector creates a user keyword vector using a word embedding method using a neural network. When word embedding, all corpuses imported into the platform are utilized. Word embedding methods include an embedding layer learning method using a corpus and a word-document matrix method. Learn the embedding layer from the word-matrix method. And when learning RNN, there are three: one-hot vector (reflecting the user's major), the user's latest interest keyword vector using the word embedding method, and the user's previous interest keyword vector in the hidden layer. The vectors are integrated and received as input.

The first group matching of the first grouping unit measures Similarity (Cost) by measuring the Sum of Squared Euclidean distance (S230).

The first group matching of the first grouping unit creates a Semantic Vector Space (S250).

The first group matching of the first grouping unit performs the K-means algorithm by calculating the cost in S230 (S270). Measure the change in cost (elbow method) by changing K. The elbow part is calculated by the number of clusters. The centers of each cluster obtained from the K-means algorithm are calculated as the initial average value for each cluster of the GMM model. In the K-means algorithm, the variance of the entire dataset is calculated as the initial variance of each cluster in GMM. The initial mixture weights of the GMM model are calculated in proportion to the number of data points for each cluster obtained from the K-means algorithm.

The first group matching of the first grouping unit is performed using GMM (Gaussian Mixture Model)-based soft-clustering (EM algorithm) (S290). The first group matching is performed by exposing the user to the cluster within the group to which the user belongs.

If the user lists or separately displays interested mentors or interested colleagues on the profile card, the user's keyword vector and the keyword vectors of the interested mentors and interested colleagues are merged, and the merged keyword vectors are subjected to GMM (Gaussian Mixture Model)-based soft-clustering ( EM algorithm) is applied to perform the first group matching. However, when merging keywords, greater weight can be given to the user's interests by adding weight to the user's keyword vector.

Figure 3 is a diagram showing an operation flowchart of the second grouping unit of the mentor-mentee group matching system using machine learning of the present invention.

The second group matching of the second grouping unit of the mentor-mentee group matching system collects mentor/mentee satisfaction results based on mutual evaluation after the user's mentoring performance (S310). Different weights can be assigned in the order of highest score in the satisfaction evaluation.

The second group matching of the second grouping unit performs collaborative filtering (S330). An example of collaborative filtering may be the Row Rank Matrix decomposition method.

The second group matching of the second grouping unit preferentially recommends mentors/mentees with high expected satisfaction as a result of collaborative filtering to the user (S350).

Through this, users can be recommended mentors or mentees who fall into a similar category to the mentor or mentee they liked while conducting mentoring. Furthermore, content originating from the group to which such mentor or mentee belongs can be provided. In other words, the second grouping unit is a third user (the third user belongs to a mentoring group that has already conducted mentoring) whose total expected satisfaction level for each member (mentor + mentee) of the mentoring group that has already conducted mentoring is higher than the predefined standard. You can recommend content created from mentoring that has already been conducted.

Figure 4 is a diagram showing an operation flowchart of the third grouping unit of the mentor-mentee group matching system using machine learning of the present invention.

The third grouping unit's new group matching is performed by applying RNN (Recurrent Neural Network) based on the mutual satisfaction of each mentor and each mentee each time mentoring is performed to perform a new group matching of mentor-mentee for the user, or A new group matching of mentor-mentee for the user is performed by applying RNN (Recurrent Neural Network) to the user's new keyword vector of interest (S410).

The new group matching of the third grouping unit can be performed based on expected satisfaction in addition to mutual satisfaction by applying RNN (Recurrent Neural Network) to perform new group matching of mentor-mentee for the user.

First, the neural network is updated by training the user's individual RNN based on the expected satisfaction obtained after collaborative filtering and the already progressed satisfaction. If the user's keyword vector changes, the RNN is applied to recommend mentoring. In addition, it is possible to match mentoring groups according to the keyword vectors of new users who have never mentored through a general RNN model trained with the keyword vectors of all users.

Within the platform, there is an individual RNN assigned to each user and a general RNN that exists only once within the entire platform.

Both types of RNN are periodically trained based on the results of collaborative filtering based on the satisfaction evaluation of all mentoring taking place within the platform. (In other words, the evaluation and results are set as output values and the RNN is trained.) This is because collaborative filtering itself reflects the mentoring results of all users within the platform.

At this time, in the case of an individual RNN, it is learned based on a specific user's keyword input and 'the user's mentoring satisfaction and expected satisfaction results for other users as a result of collaborative filtering according to the satisfaction evaluation of all users in the platform.'

At this time, in the case of the general RNN, it is a model learned based on all users' keyword vectors and the user's mentoring satisfaction + expected satisfaction with other users as a result of collaborative filtering, and the purpose is the (new) user's keywords. This is to conduct highly accurate group mapping without mentoring using only vectors.

1. The purpose of the individual RNN is to infer a more accurate group mapping using only the update information when an existing user updates the 'keyword vector' by modifying the corpus or saving the user of interest, etc. without continuing mentoring. am.

2. The purpose of General RNN is to perform highly accurate group mapping using only the user's keyword vector, even if a new user has never been mentored. This method is expected to increase the accuracy of group mapping (network recommendation) for new users compared to the existing first method.

This is expected to be effective when sufficient data within the platform is accumulated.

If the user is a college student, each of the current major and the major of interest is expressed as a one-hot vector (three one-hot vectors arranged side by side according to plan A, B, and C).

The user's interest keyword vector (word embedding) and three one-hot vectors mapped to each of the user's plans A, B, and C are input to the RNN. At this time, the user's RNN module is called RNN-1. (If there are a total of N users in the platform, RNN-2, RNN-3, …, RNN-N modules exist for each user.)

Input the hidden layer of the previous state as RNN-1 input.

Connect the input vector and the hidden layer of the current state through a neural network.

The user's expected satisfaction with all mentors and other mentees within the platform derived from the three-step collaborative filtering is set as the result of RNN-1.

Connect the current hidden layer and the result vector to a neural network and learn RNN-1.

Each time a user conducts a new mentoring/group study and conducts a mutual evaluation, RNN-1 learns through the changed collaborative filtering results.

If the user modifies keywords/career plan, etc. in the profile card without conducting additional mentoring/group study, RNN-1 recommends a mentor/mentee based on the expected results by inputting the keyword.

There may also be an RNN-General module that receives input from all users within the platform. In the case of the RNN-G module, it is a general model that learned the input and output of all users in the platform.

When new users are introduced, mentors/mentees are recommended based on the results (expected satisfaction) that appear after inputting the new user's keyword vector and one-hot vector into the RNN-G.

For users within a certain range of expected satisfaction results, connections are recommended in the order of keywords and users with high similarity.

If there is overlap with the recommended mentors/mentees as a result of soft-clustering in the first stage, the priority is increased and the keywords are listed in order of similarity.

Figure 5 is a diagram showing a flowchart of the mentor-mentee group matching method using machine learning of the present invention.

The mentor-mentee group matching method using machine learning of the present invention applies GMM (Gaussian Mixture Model)-based soft-clustering to the vector of the user's keyword of interest and the keyword vectors of the mentors and mentees. It includes performing a first group matching of mentor-mentee for the user (S510).

After performing mentoring through first group matching, it includes a step (S530) of performing second group matching of mentor-mentees for users by applying collaborative filtering based on mutual satisfaction with each mentor and each mentee. do.

Based on the mutual satisfaction of each mentor and each mentee each time mentoring is performed and the expectations obtained as a result of the above collaborative filtering, a general RNN (Recurrent Neural Network) is applied to create a new mentor-mentee for a new user. A group matching upgrade step (S550) of performing group matching or performing a new group matching of the mentor-mentee for the user by applying RNN (Recurrent Neural Network) to the user's new interest keyword vector, and mentor. Each of the mentees and mentees has one keyword vector, and the keyword vector is generated by a word embedding method.

In this specification, only a few examples of various embodiments performed by the present inventors are described, but the technical idea of the present invention is not limited or limited thereto, and of course, it can be modified and implemented in various ways by those skilled in the art.

Claims (8)

A first grouping unit that performs a first group matching of mentors and mentees for the user by applying a first method of unsupervised learning to the vector of the user's keywords of interest and the keyword vectors of the mentors and mentees. ; and
After performing mentoring through the first group matching, a second group matching of mentor-mentees for the user is performed by applying a second method of collaborative filtering based on the mutual satisfaction of each mentor and each mentee. Includes a second grouping unit,
The second method is different from the first method, each of the mentors and mentees has one keyword vector, and the keyword vector is generated by a word embedding method,
After carrying out mentoring through the second group matching, a third method of machine learning is applied based on the mutual satisfaction of each mentor and each mentee to perform a new group matching of mentor-mentee for the user, or
A third grouping unit that performs a new group matching of mentor-mentees for the user by applying the third method to the user's new interest keyword vector,
The third method is different from the first method and the second method,
The third method is performed using a Recurrent Neural Network (RNN) model or a general RNN model, and the third method is performed based on the expected satisfaction obtained as a result of the collaborative filtering in addition to the mutual satisfaction. become,
The third grouping unit matches a new group of mentor-mentees for a new user by applying the global RNN (Recurrent Neural Network) based on the mutual satisfaction and the expected satisfaction for each mentor and each mentee each time mentoring is performed. or
By applying the RNN (Recurrent Neural Network) to the user's new interest keyword vector, a new group matching of mentor-mentee for the user is performed,
When learning, the RNN receives as input a one-hot vector reflecting the user's major, the user's new interest keyword vector, and the user's previous interest keyword vector in a hidden layer, Mentor-mentee group matching system using machine learning. delete In claim 1,
The first method is performed using GMM (Gaussian Mixture Model)-based soft-clustering,
Mentor-mentee group matching system using machine learning. In claim 3,
After performing mentoring through the second group matching, the first grouping unit further merges keyword vectors of mentors and fellow mentees who participated in the second group matching, and applies the GMM (Gaussian Mixture Model) to the merged keyword vectors. A mentor-mentee group matching system using machine learning that performs the first group matching by applying soft-clustering. delete delete In claim 1,
The second grouping unit, when the total expected satisfaction level for each member (mentor + mentee) of the mentoring group that has previously conducted mentoring, is higher than a predefined standard, a third user (the third user is in the mentoring group that has previously conducted mentoring) A mentor-mentee group matching system using machine learning that recommends content generated from the mentoring that has already been conducted above to those who are not affiliated. The first grouping unit applies GMM (Gaussian Mixture Model)-based soft-clustering to the vector of the user's keywords of interest and the keyword vectors of the mentors and mentees to determine the number of mentor-mentees for the user. 1. Performing group matching; and
After mentoring is performed through the first group matching by the second grouping unit, collaborative filtering is applied based on the mutual satisfaction of each mentor and each mentee to match the second group of mentors and mentees for the user. Steps to perform; and
A new RNN (Recurrent Neural Network) is applied based on the mutual satisfaction of each mentor and each mentee each time mentoring is performed by the third grouping department and the expected satisfaction obtained as a result of the collaborative filtering. Group matching upgrade that performs a new group matching of mentor-mentees for the user, or performs a new group matching of mentor-mentees for the user by applying RNN (Recurrent Neural Network) to the user's new keyword vector of interest. Including steps;
Each of the mentors and mentees has one keyword vector, and the keyword vector is generated by a word embedding method,
When learning, the RNN receives as input a one-hot vector reflecting the user's major, the user's new interest keyword vector, and the user's previous interest keyword vector in a hidden layer, Mentor-mentee group matching method using machine learning.