KR20240171604A - System and method for recommending educational contents to develop social skills and program stored in recording medium - Google Patents

Abstract

According to an embodiment of the present invention, a system for recommending educational content comprises: an inference model generation unit for receiving an embedding vector generated from text data of a 1:1 conversation and generating an inference model trained to output an evaluation result for each social evaluation item of a 1:1 conversation person; a text data generation unit for acquiring a conversation image between 1:1 conversation persons and generating text data from the acquired 1:1 conversation image; a word embedding unit for dividing the generated text data into word units, embedding the word-based text data, and converting the word-based text data into an embedding vector value; an evaluation result inference unit for inputting the embedding vector value of the text data of a subject to be analyzed into the generated inference model, and outputting an evaluation result for each social evaluation item of the subject to be analyzed by the inference model; and a content recommendation unit for recommending educational content matched with the subject to be analyzed based on the output evaluation result for each social evaluation item.

Description

{SYSTEM AND METHOD FOR RECOMMENDING EDUCATIONAL CONTENTS TO DEVELOP SOCIAL SKILLS AND PROGRAM STORED IN RECORDING MEDIUM}

The present invention relates to an educational content recommendation technology, and more specifically, to an educational content recommendation system for improving social skills, an educational content recommendation method, and a program stored in a recording medium.

Infants or children can develop social skills by forming social relationships within groups such as kindergartens, schools, and academies. In addition, infants or children can improve their social skills through interactions with parents, brothers, and sisters within the home.

When infants or children pay attention to what others say, listen to what they say, understand it, and determine whether the other person understands what they are saying from their reactions, this is a broader ability than just language. In the process of communication, infants or children adjust topics, correct mistakes in conversation, and use the context of the situation to decide how and what to say.

There are research results showing that when infants or children pair up and interact with each other, strong social relationships are formed, and when two children pair up 1:1 and engage in peer pair activities during kindergarten or school activities, it has a more positive effect on the infants' kindergarten adaptability, sociality, and language skills than when they simply engage in group activities.

Meanwhile, as the birth rate has recently decreased, there are more cases of one child per household than two or more children, and as a result, there are fewer opportunities for pair activities within the household. For this reason, measures are needed to improve social skills in infants and children.

The problem to be solved by the present invention is to provide an educational content recommendation system, an educational content recommendation method, and a program that can evaluate a child's social ability and improve the child's insufficient social ability.

As a means for solving the above-described problem, an educational content recommendation system according to one disclosure of the present invention comprises: an inference model generation unit which receives an embedding vector generated from text data of a 1:1 conversation and generates an inference model learned to output an evaluation result for each social evaluation item of a 1:1 conversationalist; a text data generation unit which acquires a conversation video between 1:1 conversationalists and generates text data from the acquired 1:1 conversation video;

The method may include a word embedding unit that divides the generated text data into word units, embedding the words, and converting them into embedding vector values; an evaluation result inference unit that inputs the embedding vector values of the text data of the analysis subject into the generated inference model, and outputs the evaluation results for each social evaluation item of the analysis subject by the inference model; and a content recommendation unit that recommends educational content that matches the analysis subject based on the output evaluation results for each social evaluation item.

1:1 conversation partners can be matched based on user information including the user's residential area, interests, age, tendencies, and usage purpose, and matching performance information including the number of matches, cumulative usage time, and the other party's evaluation score.

The evaluation results for each social assessment item may include a satisfaction or failure for each social assessment item, including (1) discourse management ability item, (2) adjustment and application ability according to situation or listener context item, and (3) communication intention ability.

The above inference model may include a learning model based on a Bidirectional LSTM algorithm.

The above word embedding can be performed based on CNN and Glove vector embedding algorithms.

The embedding vector value of the text data of the analysis subject can be generated based on the speech text of the analysis subject in a 1:1 conversation and the speech text of the 1:1 conversation between the analysis subject and the conversation partner.

The above spoken text may be text data converted from a 1:1 conversation that takes place in a predetermined situation and over a predetermined period of time.

The above content recommendation unit can recommend labeling content corresponding to social evaluation items that were evaluated as unsatisfactory among the evaluation results.

As another means for solving the above-described problem, a video recommendation method according to one disclosure of the present invention may include an inference model generation step of receiving an embedding vector generated from text data of a 1:1 conversation and generating an inference model learned to output an evaluation result for each social evaluation item of a 1:1 conversationalist; a text data generation step of acquiring a conversation video between 1:1 conversationalists and generating text data from the acquired 1:1 conversation video; a word embedding step of dividing the generated text data into word units and converting it into an embedding vector value by word-based embedding; an evaluation index inference step of inputting an embedding vector value of text data of an analysis subject into the generated inference model and outputting an evaluation result for each social evaluation item of the analysis subject by the inference model; and a content recommendation step of recommending educational content matching the analysis subject based on the output evaluation result for each social evaluation item.

As another means for solving the above-described problem, a program according to one disclosure of the present invention may include a program stored in a recording medium for performing the above-described educational content recommendation method by a computer.

According to an embodiment of the present invention, an educational content recommendation system, an educational content recommendation method, and a program can be provided, which can evaluate a child's social ability with high accuracy and improve insufficient social ability by inputting an embedding vector generated from text data of a 1:1 conversation, generating an inference model trained to output an evaluation result for each social evaluation item of a 1:1 conversationalist, acquiring a conversation video between the 1:1 conversationalist, generating text data from the acquired 1:1 conversation video, dividing the generated text data into word units, converting it into an embedding vector value through word-based embedding, inputting the embedding vector value of the text data of the analysis subject into the generated inference model, outputting the evaluation result for each social evaluation item of the analysis subject by the inference model, and recommending educational content matching the analysis subject based on the output evaluation result for each social evaluation item.

FIG. 1 is a diagram showing the connection relationship between an educational content recommendation system (100) according to one embodiment of the present invention and other components.
Figure 2 is a block diagram showing the configuration of an educational content recommendation system (100) according to one embodiment of the present invention.
Figure 3 is a diagram illustrating the collection of speech data and generation of word embeddings.
Figure 4 is a diagram illustrating a method for generating an inference model.
FIG. 5 is a schematic diagram illustrating an operation method of an educational content recommendation system according to one embodiment of the present invention.
FIG. 6 is a flowchart illustrating a video recommendation method performed by an educational content recommendation system according to one embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

The terms used in the present invention are selected from the most widely used general terms possible while considering the functions of the present invention, but they may vary depending on the intention of engineers working in the field, precedents, the emergence of new technologies, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meanings thereof will be described in detail in the description of the relevant invention. Therefore, the terms used in the present invention should be defined based on the meanings of the terms and the overall contents of the present invention, rather than simply the names of the terms.

When a part of the specification is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated. In addition, terms such as "part", "module", etc., described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware or software, or a combination of hardware and software.

Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily practice the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in order to clearly describe the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are assigned similar drawing reference numerals throughout the specification.

FIG. 1 is a diagram showing the connection relationship between an educational content recommendation system (100) according to one embodiment of the present invention and other components.

Referring to FIG. 1, the educational content recommendation system (100) can be connected to a plurality of conversational terminals (200) and expert terminals (300) through a network. The educational content recommendation system (100) can evaluate the social ability of a conversationalist who has performed a 1:1 conversation, and recommend educational content for improving the insufficient social ability based on the evaluation result. In addition, the educational content recommendation system (100) can construct a learning model that infers the evaluation result by using the expert evaluation result for the 1:1 conversation as sample data, and can evaluate the social ability of the conversationalist using the constructed learning model.

A plurality of conversation partners can access the educational content recommendation system (100) through a network via conversation partner terminals (200). The educational content recommendation system (100) can match 1:1 conversation partners according to a user's request. The user can conduct a 1:1 conversation via video with the matched 1:1 conversation partner during a set situation and set time presented by the educational content recommendation system (100), and after the conversation, can watch the educational content recommended by the educational content recommendation system (100).

An education expert can access an education content recommendation system (100) using an expert terminal (300). The education content recommendation system (100) provides a 1:1 conversation video to the education expert, and the education expert can view the provided 1:1 conversation video and perform an evaluation of each conversationalist according to each evaluation item of social ability. The education content recommendation system (100) labels the evaluation results of the education expert on the 1:1 conversation video, and a learning model can be learned to infer the evaluation results using the labeled evaluation results.

The educational content recommendation system (100) can generate an inference model that is trained to output evaluation results for each social evaluation item of a 1:1 conversationalist. In addition, the educational content recommendation system (100) can acquire a conversation video between 1:1 conversationalists and generate text data from the acquired 1:1 conversation video. The generated text data is divided into word units and can be converted into an embedding vector value by word-based embedding.

The educational content recommendation system (100) can input the embedding vector value of the text data of the subject of analysis into the generated inference model, and output the evaluation results of each social evaluation item of the subject of analysis by the generated inference model. In addition, the educational content recommendation system (100) can recommend educational content that matches the subject of analysis based on the output evaluation results of each social evaluation item.

Figure 2 is a block diagram showing the configuration of an educational content recommendation system (100) according to one embodiment of the present invention.

Referring to FIG. 2, the educational content recommendation system (100) may include an inference model generation unit (110), a text data generation unit (120), a word embedding unit (130), an evaluation result inference unit (140), and a content recommendation unit (150). In FIG. 2, the inference model generation unit (110), the text data generation unit (120), the word embedding unit (130), the evaluation result inference unit (140), and the content recommendation unit (150) are illustrated as separate configurations, but may be configured as an integrated functional unit that performs two or more functions (for example, one functional unit that performs the functions of the inference model generation unit and the evaluation result inference unit).

The inference model generation unit (110) can receive an embedding vector generated from text data of a 1:1 conversation and generate an inference model trained to output evaluation results for each social evaluation item of a 1:1 conversationalist.

In one embodiment, an embedding vector can be generated from text data of a 1:1 conversation by word embedding, and the generated embedding vector and the evaluation results of each social evaluation item of the 1:1 conversation by an expert can be learned as sample data by an inference model of a deep learning algorithm. The expert can label the evaluation results of each conversation party of a plurality of 1:1 conversations as satisfied or unsatisfied for each social evaluation item. The inference model can be learned by the evaluation results of the expert and the embedding vector of the corresponding 1:1 conversation.

In another embodiment, the inference model may calculate the similarity between the embedding vector generated for the text data of the 1:1 conversation and the embedding vector generated by the text data of the example conversation script written by the education expert, and output the evaluation result for each evaluation item of the social ability according to the calculated similarity. That is, the education expert may write an example conversation script that satisfies the evaluation criteria for each social evaluation item, and the inference model may calculate the similarity between the first embedding vector value converted from the written example conversation script and the second embedding vector converted from the text data of the 1:1 conversation, and if the similarity is lower than a predetermined value, it may be determined that the social ability for the corresponding evaluation item is insufficient.

The evaluation results for each social evaluation item may include satisfaction or non-satisfaction for each social evaluation item, including (1) discourse management ability item, (2) adjustment and application ability item according to situation or listener context item, and (3) communication intention ability. Table 1 shows specific examples (higher items and sub-items) of social evaluation items by experts. Experts can perform evaluation by giving each interlocutor in a 1:1 conversation a score from 0 to 3 for each sub-item as shown in Table 1. If the total score of the sub-item is less than a given value, it may be determined that the social ability of the corresponding higher item is not satisfied.

[Table 1]

The inference model can include a learning model based on the Bidirectional LSTM algorithm. LSTM extracts information in the forward direction (from left to right), but it can also extract information and use it in the reverse direction, which is called Bidirectional LSTM. When using Bidirectional LSTM, it can obtain better performance than LSTM because it can extract more information from sequence data.

The text data generation unit (120) can acquire a video of a conversation between 1:1 conversation partners and generate text data from the acquired video of a 1:1 conversation partner. Matching of 1:1 conversation partners can be determined based on information of a user connected to the system.

In one embodiment, 1:1 conversation partners may be matched based on user information including the user's residence, interests, age, tendencies, and usage purposes, and matching performance information including the number of matches, cumulative usage time, and the other party's evaluation score. The matching probability may be higher between users who have the same residence, interests, age, tendencies, and usage purposes. In addition, in cases where the matching probability by the similarity of the user information is the same, the matching probability may be higher for users who have larger values for the number of matches, cumulative usage time, and the other party's evaluation score.

The text data generation unit (120) can extract voice from a 1:1 conversation video and convert the extracted voice into text. The text data generation unit (120) can determine part-of-speech tags through morphological analysis of the text and extract text data (script data) through entity recognition (natural language processing).

The word embedding unit (130) can divide the generated text data into word units, embed it on a word basis, and convert it into an embedding vector value. The text data of a 1:1 conversation is divided into word units by a morphological analyzer, and the text divided into word units can be converted into a vector value (real value) by considering the similarity between each word.

In one embodiment, word embedding can be performed based on CNN and Glove vector embedding algorithm. CNN is an algorithm that applies a filtering technique to original data and then performs a classification operation on the filtered data by adding new layers called convolutional layer and pooling layer before the fully-connected layer. Glove vector embedding algorithm is an algorithm that calculates in advance how frequently input and output word pairs of data appear simultaneously within one window and creates a loss function so that the inner product and logP of the input and output word embedding vectors become close to each other. Using Glove vector embedding algorithm reduces redundant calculations, thereby improving learning speed and increasing inference accuracy even for a small number of sample data.

The embedding vector value of the text data of the analysis subject can be generated based on the speech text of the analysis subject in a 1:1 conversation and the speech text of the 1:1 conversation between the analysis subject and the conversation partner. In other words, the embedding vector value of the text data of the analysis subject can include the speech text of the analysis subject twice and the speech text of the conversation partner once. According to this configuration, a different embedding vector value can be obtained for each conversation partner in a 1:1 conversation, and by inputting a different embedding vector value for each conversation partner into the inference model, an evaluation result of the social ability of each conversation partner can be obtained.

The utterance text of the subject of analysis during a 1:1 conversation may be a text data of a 1:1 conversation that takes place in a predetermined situation and for a predetermined time. In one embodiment, a specific situation may be given during the 1:1 conversation (for example, a time of 1 minute is given to greet each other). By presenting a specific situation and a predetermined conversation time in this way, the amount of input data input to the word embedding and inference model can be minimized.

The evaluation result inference unit (140) can input the embedding vector value of the text data of the analysis subject into the generated inference model, and output the evaluation result for each social evaluation item of the analysis subject by the inference model. For example, when a 1:1 conversation is performed between child A and child B, the text data of the 1:1 conversation is converted into an embedding vector value by word embedding, and when this embedding vector value is input into the generated inference model, the evaluation result for each social evaluation item of child A, the analysis subject, can be output.

The content recommendation unit (150) can recommend educational content that matches the analysis subject based on the evaluation results for each output social evaluation item. In addition, the content recommendation unit (150) can recommend labeling content corresponding to a social evaluation item that is evaluated as unsatisfactory among the evaluation results.

For example, if child A, who is the subject of the analysis, is evaluated as having a lack of discourse management ability (greeting and attempting to approach), a video content, “How to Greet Well” A, which is pre-labeled as a video content for improving discourse management ability, may be recommended. If there are multiple items evaluated as being lacking in the subject of the analysis, one or more video contents for improving the most lacking ability may be recommended depending on the size of the evaluation value of each item.

In addition, by evaluating the discourse management ability (greeting and approaching) of the subject of analysis who performed a 1:1 conversation in more detail, different video contents can be provided according to the detailed evaluation items of “asking the partner’s name” and “telling my name.”

For example, in the case of subject A, if the discourse management ability is evaluated as being insufficient and the detailed item "asking a partner's name" is evaluated as being insufficient, video content a can be recommended for improving the ability related to "asking a partner's name." In addition, in the case of subject B, if the discourse management ability is evaluated as being insufficient and the detailed item "telling my name" is evaluated as being insufficient, video content b can be recommended for improving the ability related to "telling my name."

Figure 3 is a diagram illustrating the collection of speech data and generation of word embeddings.

Referring to FIG. 3, a child user can access the system (100) through a terminal and be matched with another conversation partner by a child matching system. The child user can conduct a 1:1 video conversation with the conversation partner child for a set period of time (e.g., 20 minutes), and voice can be extracted from the 1:1 video and the extracted voice can be converted into text. Text data converted from the 1:1 video can be stored as big data in a database.

The analysis specialist (education specialist) can provide child language analysis indicators and metadata for 1:1 video sample data. The analysis specialist provides indicators such as vocabulary, speech, speed, and tension, and metadata such as the degree of judgment, and the evaluation results of the child's social ability can be derived based on the provided analysis indicators and metadata.

Figure 4 is a diagram illustrating a method for generating an inference model.

Voice data can be extracted from the child's matching video data and a dialogue script can be stored. Meaningless dialogues in the dialogue script can be detected as garbage and removed, and a word embedding model can be generated from meaningful dialogues.

As a result of Bidirectional LSTM, a representation configuration of each word composed of hidden states can be obtained, and an inference model can be created by learning by expert labeling data or learning based on similarity with expert example scripts.

FIG. 5 is a schematic diagram illustrating an operation method of an educational content recommendation system according to one embodiment of the present invention.

A child who accesses the educational content recommendation system (100) is matched to a 1:1 conversation by a matching system, and when the 1:1 conversation is performed, the speech script is word-embedded, the child's speech intention is predicted by an inference model, and an evaluation of social ability can be performed. The educational content recommendation system (100) recommends a solution video based on the evaluation result of the child's social ability by a video recommendation algorithm, and a child who watches the solution video can improve his or her social ability. An expert performs an evaluation of the child's social ability for the 1:1 conversation, and the evaluation result of the expert is fed back to the learning of the inference model, so that the accuracy of the inference model can be improved.

FIG. 6 is a flowchart illustrating a video recommendation method performed by an educational content recommendation system according to one embodiment of the present invention.

Referring to FIG. 6, the image recommendation method may include an inference model generation step (S110), a text data generation step (S120), a word embedding step (S130), an evaluation result inference step (S140), and a content recommendation step (S150).

In the inference model generation step (S110) performed by the inference model generation unit (110), an inference model trained to receive an embedding vector generated from text data of a 1:1 conversation and output the evaluation results for each social evaluation item of a 1:1 conversationist can be generated.

In the text data generation step (S120) performed by the text data generation unit (120), a conversation video between 1:1 conversation partners can be acquired, and text data can be generated from the acquired 1:1 conversation video.

In the word embedding step (S130) performed by the word embedding unit (130), the generated text data can be divided into word units and converted into an embedding vector value through word-based embedding.

In the evaluation result inference step (S140) performed by the evaluation result inference unit (140), the embedding vector value of the text data of the analysis subject is input into the generated inference model, and the evaluation result for each social evaluation item of the analysis subject can be output by the inference model.

In the content recommendation step (S150) performed by the content recommendation unit (150), educational content matching the analysis subject can be recommended based on the evaluation results for each output social evaluation item.

For details on the inference model generation step (S110), the text data generation step (S120), the word embedding step (S130), the evaluation result inference step (S140), and the content recommendation step (S150), reference may be made to the descriptions given above for the inference model generation unit (110), the text data generation unit (120), the word embedding unit (130), the evaluation result inference unit (140), and the content recommendation unit (150).

The steps or processes described above may be performed by hardware components, software components, and/or a combination of hardware components and software components. For example, the steps or processes described in the embodiments may be performed using one or more general-purpose computers or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone, but those skilled in the art will recognize that the processing device may include multiple processing elements and/or multiple types of processing elements. For example, the processing device may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

The software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and/or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal waves, for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over network-connected computer systems, and stored or executed in a distributed manner. The software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination. The program commands recorded on the medium may be those specially designed and configured for the embodiment or may be those known to those skilled in the art and available for use. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, flash memories, etc. Examples of the program commands include not only machine language codes generated by a compiler but also high-level language codes that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiment, and vice versa.

Although the embodiments have been described above by way of limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, appropriate results can be achieved even if the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or are replaced or substituted by other components or equivalents.

Therefore, other implementations, other embodiments, and equivalents to the claims are also included in the scope of the claims described below.

Claims (10)

An inference model generation unit that receives an embedding vector generated from text data of a 1:1 conversation and generates an inference model trained to output evaluation results for each social evaluation item of a 1:1 conversationalist;
A text data generation unit that acquires a video of a conversation between 1:1 conversation participants and generates text data from the acquired video of the 1:1 conversation;
A word embedding unit that divides the generated text data into word units, embeds words, and converts them into embedding vector values;
An evaluation result inference unit that inputs the embedding vector value of the text data of the subject of analysis into the generated inference model and outputs the evaluation result for each social evaluation item of the subject of analysis by the inference model; and
Content recommendation section that recommends educational content that matches the analysis target based on the evaluation results for each printed social evaluation item
An educational content recommendation system including: In the first paragraph,
An educational content recommendation system in which 1:1 conversation partners are matched based on user information including the user's residential area, interests, age, tendencies, and usage purpose, and matching performance information including the number of matches, cumulative usage time, and the other party's evaluation score. In the first paragraph,
The evaluation results for each social evaluation item are (1) discourse management ability item, (2) adjustment and application ability according to situation or listener context item, and (3) communication intention ability. This is an educational content recommendation system that includes satisfaction or non-satisfaction for each social evaluation item. In the first paragraph,
The above inference model is an educational content recommendation system including a learning model based on a Bidirectional LSTM algorithm. In the first paragraph,
The above word embedding is an educational content recommendation system performed based on CNN and Glove vector embedding algorithms. In the first paragraph,
An educational content recommendation system in which the embedding vector value of the text data of the subject of analysis is generated based on the speech text of the subject of analysis during a 1:1 conversation and the speech text of the 1:1 conversation between the subject of analysis and the conversation partner. In Article 6,
The above speech text is an educational content recommendation system in which a 1:1 conversation that takes place in a predetermined situation and for a predetermined time is converted into text data. In the first paragraph,
The above content recommendation unit is an educational content recommendation system that recommends labeled content corresponding to social evaluation items that were evaluated as unsatisfactory among the evaluation results. An inference model generation step for generating an inference model that receives an embedding vector generated from text data of a 1:1 conversation and is trained to output evaluation results for each social evaluation item of a 1:1 conversationalist;
A text data generation step of acquiring a video of a conversation between 1:1 conversation participants and generating text data from the acquired video of the 1:1 conversation;
A word embedding step that divides the generated text data into word units, embeds them based on words, and converts them into embedding vector values;
An evaluation index inference step in which the embedding vector value of the text data of the analysis subject is input into the generated inference model, and the evaluation result for each social evaluation item of the analysis subject is output by the inference model; and
Content recommendation step that recommends educational content that matches the analysis target based on the evaluation results for each printed social evaluation item.
A method for recommending educational content including: A program stored in a recording medium for performing the educational content recommendation method described in Article 9 by a computer.