KR102269606B1 - Method, apparatus and computer program for analyzing new contents for solving cold start - Google Patents

Abstract

The present invention relates to a method, apparatus and computer program for generating a modeling vector of new learning content. The present invention provides a method of generating a modeling vector of a new problem, for any problem, step a of vectorizing one or more problem characteristic information representing the characteristics of the problem, respectively, combining the vectorized problem characteristic information to obtain problem metadata step b of creating a, step c of learning the data analysis framework by applying one or more problem metadata to the data analysis framework, generating new problem metadata for the new problem through steps a to b, and a step d of generating a modeling vector of the new problem by applying the new problem metadata to the learned data analysis framework. According to the present invention, it is possible to model the characteristics of the problem by itself, so that it is possible to analyze and utilize the new problem without collecting the user's solution result for the new problem.

Description

NEW CONTENT ANALYSIS ANALYSIS METHOD, DEVICE AND COMPUTER PROGRAM FOR ANALYZING NEW CONTENTS FOR SOLVING COLD START

The present invention relates to a new learning content analysis method, apparatus, and computer program, and more specifically, new learning content for machine learning modeling that enables the analysis of features by itself in the case of a new problem that is difficult to analyze because there is no solution result data It relates to analytical methods, apparatus and computer programs.

In recent years, there is an increasing trend in the number of services provided by learning problem-solving results to identify and provide problem characteristics and user characteristics, or based on this, personalized content.

For example, in the case of Japanese Patent Registration No. 4447411 (Title of Invention: Learner Acquisition Characteristics Analysis System, Method and Program, Publication Date: March 16, 2006), automatically selected according to the score or deviation value and affiliation type for each learner It provides result information such as textbooks, and allows each learner to know their own learning tendency and the learning tendency of the type of affiliation close to each learner or the characteristics of mistakes in items.

In the case of Korean Patent Registration No. 10-1773065 (title of invention: method, device and computer program for providing personalized educational content), one or more problem solving result data for a plurality of users is collected to determine the degree of similarity between problems Calculate and index the calculation results into the problem.

That is, the analysis of learning data to which the conventional machine learning is applied is based on the user's problem solving results for each problem. Therefore, in the case of a newly introduced user or problem, analysis results cannot be provided before data on the user or problem is accumulated, and it is also impossible to put the characteristics into the analyzed problem pool. However, since it takes a considerable amount of time to collect the user's problem solving results, a new method for modeling a new problem is required.

An object of the present invention is to enable immediate utilization of a new problem by modeling the characteristics of the problem by itself.

Another object of the present invention is to reflect the meaning of each word included in the problem in the context in problem modeling in analyzing the content of the problem.

The present invention for achieving this object is a method of generating a modeling vector of a new problem, for any problem, step a of vectorizing one or more problem characteristic information representing the characteristics of the problem, vectorized problem characteristic information Step b to create problem metadata by combining It is characterized in that it comprises a step d of generating a modeling vector of the new problem by generating metadata and applying the new problem metadata to the learned data analysis framework.

According to the present invention as described above, it is possible to model the characteristics of the problem by itself, so that it is possible to analyze and utilize the new problem without collecting the user's solution result for the new problem.

In addition, according to the present invention, in analyzing the content of the problem, the meaning of each word included in the problem in the context can be reflected in problem modeling.

1 is a flowchart illustrating a method for generating a new problem modeling vector according to an embodiment of the present invention;
2 is a view for explaining a vectorization method of problem content according to an embodiment of the present invention;
3 is a view for explaining a framework learning method using solution result data according to an embodiment of the present invention;
4 is a diagram for explaining a framework learning method according to an embodiment of the present invention.

The above-described objects, features and advantages will be described below in detail with reference to the accompanying drawings, and accordingly, those skilled in the art to which the present invention pertains will be able to easily implement the technical idea of the present invention. In describing the present invention, if it is determined that a detailed description of a known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to refer to the same or similar elements, and all combinations described in the specification and claims may be combined in any manner. And unless otherwise specified, it is to be understood that references to the singular may include one or more, and references to the singular may also include the plural.

Conventionally, in order to analyze content and users, the concepts of the subject are manually defined by experts, and the experts individually judge and tag the concepts included in each problem for the subject. After that, each user analyzes the learner's ability based on the result of solving the tagged problems for a specific concept. However, this method has a problem in that the tag information is determined according to the subjectivity of the person. The reliability of the result data was not high because the criteria for evaluating the difficulty level were different for each person.

Recently, as machine learning using artificial neural networks has been used in various fields, efforts to exclude human intervention in data processing using machine learning frameworks are spreading. In other words, it collects the log of user problem solving results, constructs a multidimensional space composed of users and problems, and calculates vectors for each user and problem by assigning values to the multidimensional space based on whether the user got the problem right or wrong. to model users and/or problems in a way that

In this way, giving a point in a multidimensional space corresponding to a characteristic of a user or a problem can be called embedding or vectorization. In addition, a point on the multidimensional space given in response to an arbitrary object in this way may be called a modeling vector, a vectorized object, or an object vector. That is, if the object to be embedded is a problem, a point in the multidimensional space representing the problem feature can be named a problem modeling vector, a vectorized problem, or a problem vector. A point in the multidimensional space can be named as a user modeling vector, a vectorized user, or a user vector.

By vectorizing user characteristics or problem characteristics in this way, it becomes possible to mathematically calculate the location of a specific problem in the whole problem, other problems that can be clustered into groups similar to the above problems, and the similarity between other problems and the specific problem. In addition, it is possible to cluster users or problems based on attributes corresponding to the dimensions of the vector. In the present invention, it is not possible to limit and interpret which characteristics or properties the user vector or problem vectors include, but for example, the problem vector may include what concept the problem is composed of (concept diagram).

Hereinafter, a method for generating a modeling vector capable of representing characteristics of an object according to an embodiment of the present invention will be described in detail. For convenience of description, the present specification will focus on an example in which an object to be embedded as a modeling vector is a new problem. However, it should be noted that this is only an example, and the present invention may be applied to various objects such as users, problems, views, and learning contents.

1 is a diagram for explaining a method of generating a modeling vector of a new problem according to an embodiment of the present invention.

Referring to FIG. 1 , for an arbitrary problem, the server may vectorize one or more problem characteristic information indicating the characteristic of the problem, respectively ( S100 ). Here, the problem characteristic information includes the problem content, problem subject (field), image included in the problem, sound source characteristics (nationality, intonation, frequency of the speaker who recorded the sound source), sound source length, problem length, and combination problem (in one passage) Among the number of problems in which multiple problems are combined, the type of unit (part) to which the problem belongs, the type of language (English, Chinese, Japanese, Korean, etc.), and the type of the problem (listening, reading, writing, grammar, etc.) It may include at least one.

For example, when the characteristics of the sound source are to be vectorized, the sound source characteristics may be embedded in a manner of substituting an average frequency value for each dimension in each dimension. Of course, each of the problem characteristic information such as the length of the sound source and the type of part may be embedded in a preset method. The vectorized problem characteristic information may have different dimensions. For example, since the length of the sound source does not include various characteristics, it may be embedded in a low-dimensional vector. However, in the case of sound source characteristics or problem content, the dimension can be greatly increased depending on how vectorized it is.

Let's look at the example of FIG. 2 in which the problem content is vectorized.

The content of the problem contains multiple words. The server may assign an arbitrary n-dimensional vector to each of the plurality of words. The server may obtain a low-dimensional second vector by applying the first vector assigned to the first word to the encoding framework 130 ( S103 ). The server may obtain an n-dimensional third vector by applying the second vector to the decoding framework 150 ( S105 ), and the third vector is arbitrarily assigned to a second word included within a preset distance from the first word. The encoding framework 130 and the decoding framework 150 are trained to correspond to the fourth vector.

Here, the encoding framework 130 and the decoding framework 150 have a deep neural network structure, and model a vector value through a combination of a nonlinear transformation method.

The encoding framework 130 and the decoding framework 150 continuously update the weights so that an arbitrarily set initial weight value approaches an optimal weight using the input value. The encoding framework 130 may lower the dimension of the vector, and the decoding framework 150 is characterized by reconstructing the low-dimensionally encoded vector back to the original dimension.

For example, if the content of the problem contains the sentence “I like an apple”, the server returns I=(1,0,0,0), like =(0,1,0,0), an=(0) ,0,1,0) and apple=(0,0,0,1) can be given as a 4D vector. In embedding each word, one purpose of the server of the present invention is to reflect the relationship of I with other words in the sentence to the modeling vector.

Assume that the weight matrix of the encoding framework 130 is w1 and the weight matrix of the decoding framework 150 is w2. When the first vector (v1 = (1,0,0,0)) corresponding to I is input to the encoding framework 130 , the second vector v2 output from the encoding framework 130 is as follows. can indicate

That is, if the first four-dimensional vector is applied to the encoding framework 130 , a two-dimensional second vector can be obtained. The above is an example for explaining the present invention, and the number of words actually used in the problem may be hundreds of thousands or more. In this case, the order of the first vector input to the encoding framework 130 is hundreds of thousands of orders. , and depending on how the weight matrix of the encoding framework 130 is set, the order of the second vector output from the encoding framework 130 may be converted to a low dimension of one thousandth or ten thousandths.

The initial value of the weight matrix w2 of the decoding framework 150 may be a transpose matrix of the weight matrix w1 of the encoding framework 130 . That is, if w1 is an m x n matrix, w2 may be an n x m matrix. Therefore, when the decoding framework 150 is applied to the low-dimensional (k-dimensional) second vector, an n-dimensional third vector can be output. That is, the third vector can be expressed as follows.

The server of the present invention may update the weight w2 of the decoding framework 150 so that the decoding framework 150 corresponds to another word adjacent to the third matrix v3. For example, w2 is updated so that the third vector v3=(v31,v32,v33,v34) becomes (0,1,0,0) assigned to like, and v3 is assigned to an (0,0, The decoding framework 150 can be trained by updating w2 so that it becomes 1,0) and updating w2 so that v3 corresponds to (0,0,0,1) corresponding to apple.

In step 100 of vectorizing the problem content, the server may repeat steps 101 to 105 described above for each of one or more second words included within a preset distance from the first word in order to vectorize the first word. For example, in embedding of each word, it is assumed that three words in the direction behind the corresponding word are used for embedding. In the sentence I like an apple, in order to embed I, the server sequentially applies the first vector (1,0,0,0) given to I to the encoding framework 130 and the decoding framework 150, A third vector is obtained. The server uses the encoding framework 130 and the decoding framework 150 (hereinafter referred to as 'framework 100') so that the third vector corresponds to the second vector (0,1,0,0) given to like. In the next step, the framework 100 is trained so that the third vector corresponds to the second vector (0,0,1,0) given to an, and in the next step, the third vector is assigned to apple The framework 100 may be trained to correspond to the second vector (0,0,0,1) (in a direction to reduce the difference).

And the server inputs the first vector to the encoding framework 130 that has been learned in this way, and converts the low-dimensional second vector mapped to the first vector in the encoding framework 130 to the vector value of the word I, that is, It can be set as a vector value of the word I.

The server can obtain the vector value of each word like, an, and apple in the same way that the vector value of I was obtained. In this case, the vector value of each word finally obtained may be a low-dimensional vector corresponding to the second vector.

The server of the present invention does not train the framework 100 so that the output value of the framework 100 becomes the same as the input value, but the output value of the framework 100 corresponds to an arbitrary vector value assigned to adjacent words By learning the framework 100 as much as possible, it is different from the conventional method in that the relationship between each word and other words can be reflected in the embedding.

In the above example, the embedding method of a word included in one sentence has been described for convenience of explanation. However, in applying the present invention to actual learning content, the server provides an arbitrary vector value for each of hundreds of words included in the problem content. and the framework 100 can be trained to reflect all relationships with words included in one paragraph or the entire problem content. In addition, through this process, the server can model each word, each sentence, each paragraph, and each problem content as one low-dimensional vector.

In addition, for listening problems involving images and views that describe images, the server can model the relationship between images and views in the same way as above. In this case, the server sets the initial value in such a way that an arbitrary vector is given to the image feature information included in the image, and then learns the framework 100 so that the vector given to the view word and the vector given to the image correspond to each other. You can embed an image or view into a low-dimensional modeling vector in the following way.

In addition to vectorizing the problem content as described above, the server may map a preset vector value to the subject (field) of the problem, or vectorize the image using pixel values, descriptors, etc. of the image included in the problem. . In addition, the server may vectorize the problem length in a way that the problem length or section corresponds to a vector value. A language type, a problem type, and a unit (part) type may also correspond to a preset vector value.

Referring back to FIG. 1 , the server may generate problem metadata by combining various types of vectorized problem characteristic information ( S200 ). The problem metadata may be in the form of concatenating vectorized problem characteristic information. For example, for any problem A, the vectorized problem content is [0.3, 0.6, 0.3, 0.1], the vectorized source feature is [0.1, 0.2, 0.3], the vectorized problem length is [0.2], and the vectorized problem length is [0.2]. If the completed part information is [0.7], the problem metadata of problem A can be generated as [0.3, 0.6, 0.3, 0.1, 0.1, 0.2, 0.3, 0.2, 0.7].

Next, the server may learn the data analysis framework by applying one or more problem metadata to the data analysis framework (S300). The learning in step 300 may be performed using a problem with a solution result (refer to the description of FIG. 3 ) or may be performed only with a new problem (refer to the description of FIG. 4 ).

Referring to FIG. 3 , in step 300, the server applies the problem metadata to the data analysis framework so that the first problem modeling vector corresponds to the second modeling vector of the problem created by using the user's solution result data for the problem. A data analysis framework can be trained.

For example, in order to create a modeling vector of a new problem in which no solution result data exists, the server uses the problem that holds the solution result data among the problems constituting the problem database to implement the data analysis framework 300 can learn The data analysis framework 50 is a data analysis framework using conventionally used solution result data, and the data analysis framework 300 is a frame according to an embodiment of the present invention for modeling a problem using problem metadata. it is work

The server can acquire the solution result modeling vector 55 of the problem A by first applying the data analysis framework 50 to the problem A having the user's solution result data. And the server can obtain the metadata of the problem A by applying steps 100 to 200 of the present invention to the problem A, and apply the data analysis framework 300 to the metadata of the problem A corresponding to the problem A metadata A meta-modeling vector 305 can be obtained.

That is, the server obtains the modeling vector 55 as a result of solving the problem A in a conventional manner, and obtains the meta modeling vector 305 of the problem A according to an embodiment of the present invention that generates a modeling vector using metadata Thus, the data analysis framework 300 can be trained in a direction to reduce the difference between the two vector values.

After learning is completed, when a new problem is introduced, the server performs steps 100 to 200 for the new problem to generate new problem metadata, and applies the new problem metadata to the data analysis framework learned in step 300 to apply the new problem It is possible to generate a modeling vector of (S400). Therefore, in the case of a new problem generated later, a modeling vector can be generated by the problem itself without the user's solution result data, and when the modeling vector is generated, problems having similar features can be clustered based on this. Furthermore, according to the present invention, it is easy to provide a customized problem, to configure a diagnostic problem set for collecting user's solution result data, and the like.

Another embodiment of the step 300 of learning the data analysis framework 300 will be described with reference to FIG. 4 . 4 is an embodiment of a method for learning the data analysis framework 300 of the present invention without user pool result data. Therefore, according to the embodiment of FIG. 4 , the server may generate a modeling vector of a new problem only with a new problem without using any existing data.

Referring to FIG. 4, the server applies the problem metadata generated through steps 100 to 200 to the encoding framework to obtain a low-dimensional first problem vector (S303), and applies the first problem vector to the decoding framework. By applying the k-dimensional second problem vector can be obtained (S355). The server sets the encoding framework 330 and the decoding frame so that the second problem vector output from the data analysis framework 300 corresponds to the problem metadata that was initially input to the data analysis framework 300 (so that the difference is reduced). The work 350 may be trained.

As the learning progresses, the weights of each layer of the deep neural network constituting the encoding framework 330 and the decoding framework 350 are continuously updated, and the server applies the learning encoding framework 330 to the metadata of the new problem. By doing so, a new low-dimensional problem modeling vector can be derived (S400)

The reason for using a low-dimensional vector as a new problem modeling vector is that if a low-dimensional vector value is used, resources consumed for data processing (clustering, data classification, etc.) using the modeling vector can be reduced. In addition, in the case of learning an artificial neural network using a modeling vector, if the dimension of the vector value is low, the amount of learning is reduced.

In addition, in an embodiment of the present invention, each learning content is composed of vectorized metadata and embedded in a low-dimensional vector so that the content corresponds to a single vector. Through the above-described process, each feature of each problem Not only the modeling vector representing , but also the noise included in the metadata is removed.

The above-described method for generating a modeling vector of new learning content may be implemented in a server or a terminal through a new problem modeling vector generating program stored in a computer-readable medium to execute any one of the embodiments.

Some embodiments omitted in this specification are equally applicable when the implementing subject is the same. In addition, since the above-described present invention is capable of various substitutions, modifications and changes within the scope without departing from the technical spirit of the present invention for those of ordinary skill in the art to which the present invention pertains, the above-described embodiments and accompanying It is not limited by the drawing.

Claims (5)

A method for a server to generate a modeling vector of a new problem,
For an arbitrary problem, a step of vectorizing one or more problem characteristic information representing the characteristic of the problem, respectively;
b step of generating problem metadata by combining vectorized problem characteristic information;
c step of applying one or more problem metadata to a data analysis framework to train the data analysis framework;
a step d of generating new problem metadata for the new problem through the steps a to b, and applying the new problem metadata to the learned data analysis framework to generate a modeling vector of the new problem, ,
In step c,
Learning the data analysis framework so that the first problem modeling vector applying the problem metadata to the data analysis framework corresponds to the second modeling vector of the problem that has been previously generated using the user's solution result data for the problem A new problem modeling vector generation method comprising the step of: According to claim 1,
Step c is
step c-1 of applying the k-dimensional problem metadata to an encoding framework to obtain a low-dimensional first problem vector;
step c-2 of applying the first problem vector to a decoding framework to obtain a second problem vector of k-dimensionality;
a step c-3 of learning the encoding framework and the decoding framework by repeating steps c-1 to c-2 so that the second problem vector corresponds to the problem metadata;
and c-4 of setting the second problem vector at the time when learning is completed as a modeling vector of the problem metadata. According to claim 1,
When the problem characteristic information is a problem content, and the problem content includes a plurality of words,
Step a is
a-1 step of assigning an arbitrary n-dimensional vector to each of the plurality of words;
a-2 step of applying the first vector assigned to the first word to the encoding framework to obtain a low-dimensional second vector;
a-3 step of applying the second vector to a decoding framework to obtain an n-dimensional third vector;
a-4 step of learning the encoding framework and the decoding framework so that the third vector corresponds to a fourth vector arbitrarily assigned to a second word included within a preset distance from the first word;
a-5 step of repeatedly performing steps a-2 to a-4 for each of one or more second words to set a finally derived second vector as a vector value of the first word;
A new problem modeling vector creation comprising the step of performing steps a-2 to a-5 for each of the plurality of words to obtain a vector value for each word, and vectorizing the problem content using the vector value for each word Way. A new problem modeling vector generation application stored in a computer readable medium for executing the method of any one of claims 1 to 3 on a computer. A server for generating a new problem modeling vector according to any one of the methods of claims 1 to 3.

Images