KR20200087312A - Device and method for evaluating the communication qualities of authors based on machine learning models - Google Patents

Abstract

According to one embodiment of the present invention, a device for evaluating the communication qualities of authors based on a machine learning model may include: a collection unit for collecting survey data for evaluating the communication qualities of online text authors; a communication quality information generation unit for generating communication quality information for each online text author based on the survey data; a feature extraction unit for grouping a plurality of online text authors based on the communication quality information for each online text author, and extracting features from the online text of and author information on the grouped online text authors; a model building unit for constructing a model capable of evaluating the communication quality information on the online text authors by using a preset machine learning method based on the extracted features; and an evaluation unit for receiving new online text and evaluating the communication quality information on authors of the new online text through the built model.

Description

DEVICE AND METHOD FOR EVALUATING THE COMMUNICATION QUALITIES OF AUTHORS BASED ON MACHINE LEARNING MODELS}

The present application relates to an apparatus and method for evaluating communication qualities of authors based on a machine learning model.

The social network service enables individuals to access various information and provides an environment for interaction with various users. Social network services are closely related to the daily lives of modern people, and their use is also increasing due to the generation and provision of various information.

On the other hand, the social network service has a problem in that the guarantee of the reliability and accuracy of information is not clear due to the characteristics that anyone can access, and misleading information or wrong information can be easily and quickly distributed. However, since it is difficult to confirm the reliability of all information provided to the reader through the social network service, there is a limitation in determining the reliability of information based on the reader's individual intellectual ability, experience, and knowledge. In addition, conventional reliability evaluation techniques have been limited to evaluating the reliability of the online post itself by a simple method, and there have been no attempts to evaluate the reliability of the online post written by the author based on the reliability of the author of the online post itself.

The background technology of the present application is disclosed in Korean Patent Publication No. 10-2014-0076667.

The present application is to solve the above-mentioned problems of the prior art, the machine learning model based on the machine's learning model based on the reader's evaluation of the online text, the machine learning model based author's communication quality evaluation device that can evaluate the communication qualities of the author And to provide a method.

An object of the present invention is to provide a device and method for evaluating communication qualities of authors based on a machine learning model capable of evaluating communication qualities of new online text authors as a solution to the aforementioned problems of the prior art.

However, the technical problems to be achieved by the embodiments of the present application are not limited to the technical problems as described above, and other technical problems may exist.

As a technical means for achieving the above technical task, a method for evaluating communication qualities of an author based on a machine learning model according to an embodiment of the present application includes: (a) collecting questionnaire data for evaluating communication qualities of an online text author , (b) generating online text author communication qualities information based on the questionnaire data, (c) grouping a plurality of online text authors based on the online text author communication qualities information, (d) the A step of extracting features from online text and author information of a plurality of grouped online text authors, and (e) a model capable of evaluating communication qualities of authors of online texts using a preset machine learning method based on the extracted features. And (f) receiving a new online text and evaluating the communication quality information of the author of the new online text through the constructed model.

According to an embodiment of the present application, in step (a), the questionnaire data includes a response to a quality parameter for evaluating communication quality, and the quality parameter includes: source reliability, interpersonal attraction, conversation ability, Interaction intentions.

According to an embodiment of the present application, in step (b), communication quality information is generated by further considering characteristic data of a questionnaire respondent, and the physician is based on statistical analysis using the questionnaire data and characteristic data of a questionnaire respondent. Communication quality information is generated, and the correlation between the characteristic data and the communication quality information is calculated through multivariate linear regression analysis. The characteristic data of the questionnaire respondents include gender, age, SNS usage period, and SNS of the questionnaire respondent. It may include at least one of frequency and nature of use.

According to an embodiment of the present application, in step (c), candidate online text authors are selected based on a resampling technique among full online text authors, and the candidate online is based on preset reference values for the communication quality information. Each of the text authors can be divided into upper and lower groups.

According to an embodiment of the present application, step (d) extracts the feature based on the author information, syntax information, similarity information, emotion information, and linguistic technique, and step (e) inputs the feature The above model can be constructed based on the classification model.

According to one embodiment of the present application, the linguistic technique includes linguistic survey and word count analysis and cosine similarity, and the linguistic survey and word count analysis includes at least one of emotion, style, survey, and word from the online text. Extraction, and the cosine similarity may extract at least one of hashtag, punctuation, emoticon, link, and retweet from the online text.

A method of constructing a model for evaluating communication qualities according to an embodiment of the present application includes: (b) collecting questionnaire data for evaluating communication qualities of an online text author of the online text by a survey respondent, ( c) generating communication quality information based on the questionnaire data, (d) extracting author information of the online text and features from the online text, and (e) classifying the learning model using the features as input. On the basis of this, it may include the step of constructing a model capable of evaluating the communication quality information of the author of the online text.

According to an embodiment of the present application, in step (a), the questionnaire data includes a response to a quality parameter for evaluating communication quality, and the quality parameter includes: source reliability, interpersonal attraction, conversation ability, Interaction intentions.

According to an embodiment of the present application, step (d) extracts the features based on the author information, syntax information, similarity information, emotion information, and linguistic techniques, and the linguistic techniques include language research and word calculation analysis And cosine similarity, wherein the language survey and word calculation analysis extract at least one of emotions, styles, surveys, and words from the online text, and the cosine similarity is a hashtag, punctuation, emoticon from the online text. , Link, Retweet can extract at least one.

The apparatus for evaluating communication qualities of authors based on a machine learning model according to an embodiment of the present disclosure includes a collection unit that collects survey data for evaluating communication qualities of online text authors, and communication qualities information for each online text author based on the survey data Communication feature information generating unit for generating, grouping a plurality of online text authors based on the communication feature information for each of the online text authors, and extracting features from online text and author information of the grouped plurality of online text authors Extraction unit, receiving a model construction unit and a new online text to build a model capable of evaluating the communication qualities information of the author of the online text in a preset machine learning method based on the extracted features, and through the constructed model It may include an evaluation unit for evaluating the communication quality information of the author of the new online text.

According to an embodiment of the present application, the questionnaire data includes a response to a quality parameter for evaluating communication quality, and the quality parameter includes source reliability, interpersonal attraction, communication ability, and interaction intention. Can.

According to an embodiment of the present application, the feature is extracted based on the author information, syntax information, similarity information, emotion information, and linguistic techniques, and the linguistic technique includes language survey and word calculation analysis, and cosine similarity, The language survey and word calculation analysis extracts at least one of emotion, style, survey, and word from the online text, and the cosine similarity is at least one of hashtag, punctuation, emoticon, link, and retweet from the online text. Either can be extracted.

The above-described problem solving means are merely exemplary and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description of the invention.

According to the above-described problem solving means of the present application, an apparatus and method for evaluating communication qualities of authors based on a machine learning model capable of evaluating communication qualities of an online text author through a machine learning model based on readers' evaluation of online text can do.

According to the above-described problem solving means of the present application, it is possible to provide an apparatus and method for evaluating communication qualities of authors based on a machine learning model capable of evaluating communication qualities of new online text authors.

1 is a view showing the configuration of the author's communication evaluation system based on a machine learning model according to an embodiment of the present application.
FIG. 2 is a diagram showing the configuration of an author's communication evaluation apparatus based on a machine learning model according to an embodiment of the present application.
3 is a diagram illustrating an example of a question about a quality parameter for evaluating a communication quality of an apparatus for evaluating communication quality of an author based on a machine learning model according to an embodiment of the present application.
4 and 5 are diagrams illustrating examples of online texts of an author's communication quality evaluation device based on a machine learning model according to an embodiment of the present application.
FIG. 6 is a diagram illustrating a multivariate regression analysis result of a feature parameter of an author's communication feature evaluation apparatus based on a machine learning model according to an embodiment of the present application.
7 is a diagram illustrating a flow of a method for evaluating communication qualities of an author based on a machine learning model according to an embodiment of the present application.
8 is a diagram illustrating a flow of a method for building a model for evaluating communication qualities according to an embodiment of the present application.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those skilled in the art to which the present application pertains may easily practice. However, the present application 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 application in the drawings, parts irrelevant to the description are omitted, and like reference numerals are assigned to similar parts throughout the specification.

Throughout this specification, when a part is "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element in between. do.

Throughout this specification, when one member is positioned on another member “on”, “on top”, “top”, “bottom”, “bottom”, “bottom”, it means that one member is on another member This includes cases where there is another member between the two members as well as when in contact.

Throughout the present specification, when a part “includes” a certain component, it means that the component may further include other components, not to exclude other components, unless specifically stated to the contrary.

1 is a diagram showing the configuration of a communication evaluation system of an author based on a machine learning model according to an embodiment of the present application, and FIG. 2 is a configuration of a communication evaluation device of an author based on a machine learning model according to an embodiment of the present application It is a diagram showing.

Referring to FIG. 1, the communication quality evaluation system 10 may include a communication quality evaluation device 100 and a social network server 200. The social network server 200 may include a database that stores online text written by all online text authors. In addition, the communication quality evaluation device 100 and the social network server 200 may be connected to a network. The network refers to a connection structure capable of exchanging information between each node such as terminals and servers, and examples of such a network include a 3rd Generation Partnership Project (3GPP) network, a Long Term Evolution (LTE) network, and 5G. Network, World Interoperability for Microwave Access (WIMAX) network, Internet, Local Area Network (LAN), Wireless Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), wifi network, Bluetooth (Bluetooth) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, but is not limited thereto.

Referring to FIG. 2, the communication feature evaluation device 100 includes a collection unit 110, a communication feature information generating unit 120, a feature extraction unit 130, a model building unit 140, and an evaluation unit 150 It may include.

The collection unit 100 may collect online text from the social network server 200 through the network. In addition, the collection unit 100 may collect questionnaire data for evaluating communication qualities of online text authors. If you describe the questionnaire for evaluating the communication qualities of online text authors before collecting the survey data, the tweets for the June to August 2016 Olympic Games, that is, the online text and January to April 2017, Zika, Ebola, Chi An online text author who collected about 12,000,000 online texts via Twitter Search API (https://dev.twitter.com/rest/public) and wrote less than 10 of the above tweets for the Kungunya virus related tweets. To filter. After filtering, 15,260 authors were calculated, and 1,000 authors and tweets of each author were randomly selected from the authors.

The survey is conducted through SurveyMonkey, and the collection unit 110 may collect survey data from SurveyMonkey. Prior to the survey, the survey respondents were informed of the goals and procedures of the survey, and the consent of the survey was requested. The questionnaire can be roughly divided into two sections, and the first section provides survey respondents with questionnaires about gender, age, SNS usage period, SNS usage frequency, and personality of the survey respondents, and each question has 1 to 5 points You can respond on the Likert scale. For example, in the case of age, it can be divided into 1 point in 10s, 2 points in 20s, 3 points in 30s, 4 points in 40s, and 5 points in 50s. You can conduct a questionnaire in a manner. The above personality questionnaire used five personality traits, and the personality traits were extroverted (social level), sympathetic (organized, thorough), conscientious (organized, thorough), neurosensitive (emotional stability level), and openness. (Level of creativity and desire for knowledge and new experiences).

In the second section of the survey, each survey respondent is given an online text written by the same online text author, and a questionnaire respondent is asked about the quality parameters for communication quality assessment. As described above, each question can be answered on a scale of 1 to 5 points. The collection unit 110 may collect characteristic data of the survey respondents from the questionnaire of the first section, and may collect questionnaire data from the questionnaire of the second section. That is, the characteristic data of the survey respondent may include at least one of the gender, age, SNS usage period, SNS usage frequency, and personality of the survey respondent. In addition, the questionnaire data may include a response to a quality parameter for evaluating communication quality.

3 is a diagram illustrating an example of a question about a quality parameter for evaluating a communication quality of an apparatus for evaluating communication quality of an author based on a machine learning model according to an embodiment of the present application.

The quality parameters are, for example, source credibility (SC), interpersonal attraction (IA), communication competence (CC), and intent to interact ( INT)). Referring to FIG. 3, each quality parameter includes two or more questions, and each question is based on the Likert scale, 1 point (not entirely agreed), 2 points (not agreed), 3 points (normal), You can respond by choosing one of four (agree) or five (agree entirely). In addition, referring to Figure 3, according to the results of Cronbach's α (Cronbach's α), it can be seen that the interrelationship between questions of each feature parameter is high. For example, in interpersonal attractiveness, the questions of Ia1 and Ia2 are related to social attractiveness, and Ia3 and Ia4 are related to business attractiveness.

The communication feature information generation unit 120 may generate communication feature information for each online text author based on the questionnaire data. Communication quality information through the questionnaire data may be a variable dependent on the above four quality parameters. However, as shown in the results of the Kronbach alpha of FIG. 3, since the question of each feature parameter has a high reliability and is closely related to each other, it is an effective approach to calculate communication feature information from the response to the feature parameter. It can be said.

In addition, characteristic data of survey respondents may influence communication qualities information. For example, the elderly respondents had a negative perception of the communication qualities of online text authors, and the female respondents had a more positive perception of all questions than the male respondents. Accordingly, the communication quality information generating unit 120 may generate communication quality information by further considering characteristic data of the survey respondent.

The communication feature information generating unit 120 may generate communication feature information based on statistical analysis using questionnaire data and characteristic data of the survey respondents. In addition, the communication feature information generating unit 120 may calculate a correlation between characteristic data and communication feature information through multivariate linear regression analysis. Among the various techniques for analyzing the correlation, in the case of multivariate linear regression analysis, to measure the effectiveness of the regression results, DurbinWatson statistical test, variance inflation factor (VIF) test, and additional measurement and calculation of SPSS statistical analysis This may be required. Illustratively, the DurbinWatson statistical test means measuring the autocorrelation of the residuals in a regression analysis. The result of the DurbinWatson statistic can be calculated as a value of 0-4. When the result is 2, it means that there is no auto-correlation in the sample, and the closer the result is to 0, the more the positive auto-correlation. In addition, the closer the result is to 4, the more the negative auto-correlation.

Variance expansion coefficient (VIF) testing means measuring the severity of multicollinearity in a least-squares regression analysis. The variance expansion coefficient can provide an index to measure how much the variance of the regression coefficient estimated by collinearity increased. In general, when the coefficient of dispersion expansion exceeds 4, further investigation is required, and when it exceeds 10, it can be interpreted that it exhibits considerable multi-layer properties.

SPSS statistical analysis using a stepwise linear regression feature can recompress multiple independent variables while simultaneously removing non-critical variables. This can lead to multiple regressions each time the smallest correlation variable is removed. Consequently, SPSS statistical analysis is the analysis that leaves the independent variable that best represents the distribution.

As described above, in generating communication quality information, the communication quality information generating unit 120 may consider correlation between characteristic data and communication quality information through multivariate linear regression analysis. According to the correlation between the characteristic data and the communication quality information, it can be confirmed that the characteristic data of the questionnaire respondents may affect the communication quality information. Therefore, it can be said that it is important to consider characteristic data of survey respondents when generating communication quality information. The communication feature information generating unit 120 may generate communication feature information that is calculated as one score through an average of responses to the feature parameters. At this time, since the survey data by the survey respondent is generated based on the score based on the criteria of the individual survey respondent, the criteria of such individuals may affect the overall communication quality information. Therefore, consensus needs to be considered for the online text used in the questionnaire. According to one embodiment of the present application, the communication feature information generation unit 120 may consider the standard deviation σ of the average score of the communication feature information by questionnaire. In addition, the communication feature information generation unit 120 may consider non-verbal elements (eg, hashtags, punctuation marks, emoticons, links (URLs), and retweets) of online text as independent variables.

4 and 5 are diagrams illustrating examples of online texts of an author's communication quality evaluation device based on a machine learning model according to an embodiment of the present application.

FIG. 4 shows three online texts with different communication qualities information, and FIG. 5 shows online text with relatively high communication qualities information. Referring to Figure 4, Figure 4 (a) shows an online text composed of a number of hashtags. If a tweet is created with a hash tag without mention, it may be calculated with low communication quality information. The communication quality information of FIG. 4(a), that is, the score was calculated as 2.59 and the standard deviation σ was calculated as 1.0. This means that many survey respondents gave a low score to the online text, while a few survey respondents gave a high score. FIG. 4(b) shows a tweet including both the repeated mention and the hash tag, and in this case, as shown in FIG. 4(a), it may be calculated with low communication quality information. (Score 2.98, standard deviation σ 1.3) Fig. 4(c) shows similar tweets on the same subject. In the case of (c) of FIG. 4, a low score may be calculated because it is recognized as a duplicate and meaningless tweet to the questionnaire respondent. (Score 2.98, standard deviation σ 1.2)

On the other hand, FIG. 5 shows tweets calculated with relatively high communication quality information (score 3.75, standard deviation σ 0.55). Referring to FIG. 5, survey respondents are high in tweets that do not contain hashtags or repeated mentions. You can confirm that you gave the score. As described above, since the non-verbal element affects communication feature information, it can be said that it is important to consider the non-verbal element in calculating communication feature information.

The feature extracting unit 130 may group a plurality of online text authors based on communication quality information for each online text author. For example, the feature extraction unit 130 may be divided into an upper group having a high score and a lower group having a low score according to communication quality information. A more accurate model can be built by constructing a model for evaluating communication qualities in consideration of criteria for classifying high and low scores based on communication quality information. The construction of the model will be described later.

The feature extraction unit 130 may select a candidate online text author based on a resampling technique (SMOT) from among all online text authors. Also, the feature extracting unit 130 may classify each of the candidate online text authors into an upper group and a lower group based on a preset reference value for communication feature information. According to the above-mentioned questionnaire, each questionnaire can respond on the Likert scale. It can be assumed that a score of 2.5 or less indicates a negative response of the questionnaire, and a score of 3.5 or more indicates a positive response of the questionnaire. Accordingly, the feature extracting unit 130 may classify online text authors having communication qualities of 3.5 points or more into upper groups, and online text authors having communication qualities of 2.5 points or less into lower groups. As a result of classifying the entire online text authors into upper and lower groups, the lower group accounted for 20% and the upper group accounted for 80%. The feature extracting unit 130 selects candidate online text authors based on the Synthetic Minority Over-sampling Technique (SMOTE), which is one of the resampling techniques, to select the same number of candidate online text authors in each group. Can be screened. The feature extracting unit 130 may over-sample from the lower group to select candidate online text authors at the same ratio as the upper group and the lower group. The composite minity oversampling technique is a technique using the KNN approach, so a detailed description is omitted.

Also, the feature extracting unit 130 may extract features from online text and author information of a plurality of grouped online text authors. The extracted features can be used to build a model for evaluating communication qualities. The feature extraction unit 130 may extract features from online text based on five criteria. Specifically, the feature extraction unit 130 may extract features based on author information, syntax information, similarity information, emotion information, and linguistic techniques. The author information may include at least one of the author's name, author profile mention, age, number of followers, number of followers, followers, posts, and author's favorite posts (for example, posts with'Like' on Twitter). . The syntax information may include at least one of the number of retweets of online text, favorites, hashtags, punctuation marks (question marks, exclamation marks, etc.) in online text, the number of words, and the length of words. The similarity information means that the online text written by the same author is converted into a TF-IDF vector and the cosine similarity is calculated to calculate the similarity of the subject. The emotion information can be calculated based on natural language processing using TextBlob. The emotion score has a range of -1 to 1, -1 means a negative emotion score, and 1 means a positive emotion score.

Linguistics techniques can extract features through linguistic inquiry and word count analysis (LIWC) and cosine similarity. The language survey and word calculation analysis may extract at least one of emotion, style, survey, and word from the online text. According to linguistic survey and word count analysis, text can be extracted from online text, but since non-verbal factors cannot use language survey and word count analysis, features can be extracted from non-verbal factors through cosine similarity. have. Cosine similarity can extract at least one of hashtags, punctuation marks, emoticons, links (URLs), and retweets from online text. In addition, cosine similarity can be calculated by converting online text written by the same author into a TF-IDF vector.

FIG. 6 is a diagram illustrating a multivariate regression analysis result of a feature parameter of an author's communication feature evaluation apparatus based on a machine learning model according to an embodiment of the present application.

Referring to FIG. 6, the feature extraction unit 130 may calculate a correlation between features extracted through multivariate regression analysis linguistics technique and cosine similarity and communication quality information. For example, the more words (features) of 6 characters or more, the higher the quality parameters (source reliability, interpersonal attractiveness, conversational ability, and interaction intention) can be awarded. Using six or more words means factors related to education and social class in tweets, and using these words can have a positive impact on the questionnaire for evaluating communication qualities. In addition, the more frequently a definite article is used, the higher the quality of communication information can be calculated. Specifically, the higher the number of communication qualities of a relatively high score, the more tweets representing interests in concept nouns, objects, and objects.

The model building unit 140 may build the model based on the classification model using the extracted feature as an input. Specifically, the model building unit 140 may build a model capable of evaluating the communication quality information of the author of the online text using a preset machine learning method based on the extracted features. For example, the model building unit 140 may build a model capable of evaluating communication quality information through machine learning based on a random forest algorithm among the classification models. In addition to the random forest algorithm, the model building unit 140 can build a model capable of evaluating communication quality information based on algorithms such as Logistic Regression (LR), Support Vector Machine (SVM), and Adaptive boosted Decision Trees (ADT). have. However, as a result of calculating the accuracy, precision, and detection rate of the models based on the respective algorithms, the random forest algorithm showed the best performance, and the model building unit 140 communicated through machine learning based on the random forest algorithm. Build a model to evaluate information. The model capable of evaluating the communication quality information constructed by the model building unit 140 is different from the model based on the random forest algorithm and the dependent variable, and the labeling process can be completely newly performed.

The evaluation unit 150 may receive the new online text and evaluate the communication quality information of the author of the new online text through the constructed model. The evaluation unit 150 extracts features from the new online text and the author information of the new online text, and the communication quality information of the author of the new online text through a model capable of evaluating communication quality information using the extracted features as input. You can calculate the score.

7 is a diagram illustrating a flow of a method for evaluating communication qualities of an author based on a machine learning model according to an embodiment of the present application, and FIG. 8 is a model for constructing a model for evaluating communication qualities according to an embodiment of the present application It is a diagram showing the flow of the method.

The method of evaluating the communication qualities of the author based on the machine learning model shown in FIG. 7 and the method of constructing a model for evaluating the communication qualities shown in FIG. 8 are the communication qualities evaluation devices described through FIGS. 1 to 6 It can be performed by (100). Therefore, even if omitted, the description of the communication quality evaluation apparatus 100 through FIGS. 1 to 6 may be applied to FIGS. 7 and 8 as well.

Referring to FIG. 7, in step S710, the collection unit 100 may collect online text from the social network server 200 through the network. In addition, the collection unit 100 may collect questionnaire data for evaluating communication qualities of online text authors. For example, the survey is conducted through SurveyMonkey, and the collection unit 110 may collect survey data from SurveyMonkey. The questionnaire data may include a response to a quality parameter for evaluating communication quality. The quality parameters are, for example, source credibility (SC), interpersonal attraction (IA), communication competence (CC), and intent to interact ( INT)).

In step S720, the communication feature information generating unit 120 may generate communication feature information for each online text author based on the questionnaire data. In addition, the communication feature information generating unit 120 may generate communication feature information by further considering characteristic data of the survey respondent. The characteristic data of the survey respondent may include at least one of the gender, age, SNS usage period, SNS usage frequency, and personality of the survey respondent. In addition, the communication quality information generating unit 120 may generate communication quality information based on statistical analysis using questionnaire data and characteristic data of the survey respondents. In addition, the communication feature information generating unit 120 may calculate a correlation between characteristic data and communication feature information through multivariate linear regression analysis.

In step S730, the feature extraction unit 130 may group a plurality of online text authors based on the communication quality information for each online text author. For example, the feature extraction unit 130 may be divided into an upper group having a high score and a lower group having a low score according to communication quality information. The feature extraction unit 130 may select a candidate online text author based on a resampling technique (SMOT) from among all online text authors. In addition, the feature extracting unit 130 may classify each of the candidate online text authors into an upper group and a lower group based on a preset reference value for communication quality information.

In step S740, the feature extraction unit 130 may extract features from online text and author information of a plurality of grouped online text authors. The feature extraction unit 130 may extract features from online text based on five criteria. Specifically, the feature extraction unit 130 may extract features based on author information, syntax information, similarity information, emotion information, and linguistic techniques. Linguistics techniques can extract features through linguistic inquiry and word count analysis (LIWC) and cosine similarity. The language survey and word calculation analysis may extract at least one of emotion, style, survey, and word from the online text. According to linguistic survey and word count analysis, text can be extracted from online text, but since non-verbal factors cannot use language survey and word count analysis, features can be extracted from non-verbal factors through cosine similarity. have. Cosine similarity can extract at least one of hashtags, punctuation marks, emoticons, links (URLs), and retweets from online text. In addition, cosine similarity can be calculated by converting online text written by the same author into a TF-IDF vector.

In step S750, the model building unit 140 may build the model based on the classification model using the extracted feature as an input. Specifically, the model building unit 140 may build a model capable of evaluating the communication quality information of the author of the online text using a preset machine learning method based on the extracted features. For example, the model building unit 140 may build a model capable of evaluating communication quality information through machine learning based on a random forest algorithm among the classification models. The model capable of evaluating the communication quality information constructed by the model building unit 140 is different from the model based on the random forest algorithm and the dependent variable, and the labeling process can be completely newly performed.

In step S760, the evaluation unit 150 may receive the new online text and evaluate the communication quality information of the author of the new online text through the constructed model. The evaluation unit 150 extracts features from the new online text and the author information of the new online text, and the communication quality information of the author of the new online text through a model capable of evaluating communication quality information using the extracted features as input. You can calculate the score.

Referring to FIG. 8, in step S810, the collection unit 100 may collect online text from the social network server 200 through the network.

In step S820, the collection unit 100 may collect questionnaire data for evaluation of communication qualities of the online text author. For example, the survey is conducted through SurveyMonkey, and the collection unit 110 may collect survey data from SurveyMonkey. The questionnaire data may include a response to a quality parameter for evaluating communication quality. The quality parameters are, for example, source credibility (SC), interpersonal attraction (IA), communication competence (CC), and intent to interact ( INT)).

In step S830, the communication feature information generating unit 120 may generate the communication feature information of the online text author based on the questionnaire data. In addition, the communication feature information generating unit 120 may generate communication feature information by further considering characteristic data of the survey respondent. The characteristic data of the survey respondent may include at least one of the gender, age, SNS usage period, SNS usage frequency, and personality of the survey respondent. In addition, the communication quality information generating unit 120 may generate communication quality information based on statistical analysis using questionnaire data and characteristic data of the survey respondents. In addition, the communication feature information generating unit 120 may calculate a correlation between characteristic data and communication feature information through multivariate linear regression analysis.

In step S840, the feature extraction unit 130 may extract features from the author information and online text of the online text. Specifically, the feature extraction unit 130 may extract features based on author information, syntax information, similarity information, emotion information, and linguistic techniques. Linguistics techniques can extract features through linguistic inquiry and word count analysis (LIWC) and cosine similarity. The language survey and word calculation analysis may extract at least one of emotion, style, survey, and word from the online text. According to linguistic survey and word count analysis, text can be extracted from online text, but since non-verbal factors cannot use language survey and word count analysis, features can be extracted from non-verbal factors through cosine similarity. have. Cosine similarity can extract at least one of hashtags, punctuation marks, emoticons, links (URLs), and retweets from online text. In addition, cosine similarity can be calculated by converting online text written by the same author into a TF-IDF vector.

In step S850, the model building unit 140 may build the model based on the classification model using the extracted feature as an input. Specifically, the model building unit 140 may build a model capable of evaluating the communication quality information of the author of the online text using a preset machine learning method based on the extracted features. For example, the model building unit 140 may build a model capable of evaluating communication quality information through machine learning based on a random forest algorithm among the classification models. The model capable of evaluating the communication quality information constructed by the model building unit 140 may differ from the model based on the random forest algorithm and the dependent variables, and the labeling process may be entirely newly performed.

According to one embodiment of the present application, a method for evaluating communication qualities of an author based on a machine learning model and a method for constructing a model for evaluating communication qualities are implemented in a form of program instructions that can be performed through various computer means. It can be recorded on a readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention or may be known and usable by those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specifically configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes made by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

The above description of the present application is for illustrative purposes, and a person having ordinary knowledge in the technical field to which the present application belongs will understand that it is possible to easily change to other specific forms without changing the technical spirit or essential characteristics of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the claims, which will be described later, rather than the detailed description, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts should be interpreted to be included in the scope of the present application.

Claims (12)

In the method of evaluating the communication qualities of authors based on machine learning models,
(a) collecting questionnaire data for evaluating communication qualities of online text authors;
(b) generating communication quality information for each online text author based on the questionnaire data;
(c) grouping a plurality of online text authors based on the communication quality information for each online text author;
(d) extracting features from online text and author information of the grouped online text authors; And
(e) constructing a model capable of evaluating the communication quality information of the author of the online text using a preset machine learning method based on the extracted feature; And
(f) receiving a new online text and evaluating the communication quality information of the author of the new online text through the constructed model,
A method of evaluating communication qualities of authors based on a machine learning model comprising a. According to claim 1,
Step (a) is,
The questionnaire data includes a response to a quality parameter for evaluating communication quality,
The qualities parameters include source reliability, interpersonal attractiveness, conversational ability, and interaction intentionality. The method of evaluating communication qualities of authors based on a machine learning model. According to claim 1,
Step (b) is,
Communication characteristics are generated by further considering characteristic data of survey respondents,
The communication quality information is generated based on the statistical analysis using the questionnaire data and characteristic data of the survey respondents,
A correlation between the characteristic data and the communication quality information is calculated through multivariate linear regression analysis,
The characteristic data of the survey respondents,
A method for evaluating communication qualities of authors based on a machine learning model, which includes at least one of gender, age, SNS usage period, SNS usage frequency, and personality of the survey respondent. According to claim 1,
Step (c) is,
Among the online text authors, candidate online text authors are selected based on the resampling technique.
A method for evaluating communication qualities of authors based on a machine learning model, wherein each of the candidate online text authors is classified into a high group and a low group based on a preset reference value for the communication qualities. According to claim 1,
Step (d) is,
Extracting the feature based on the author information, syntax information, similarity information, emotion information, and linguistic techniques,
Step (e) is,
A method for evaluating communication qualities of authors based on a machine learning model, wherein the model is constructed based on a classification model using the feature as an input. The method of claim 5,
The linguistic technique,
Includes language research and word count analysis and cosine similarity,
The language survey and word calculation analysis extracts at least one of emotion, style, survey, and word from the online text,
The cosine similarity is to extract at least one of hashtags, punctuation marks, emoticons, links, and retweets from the online text, and a method for evaluating communication qualities of authors based on a machine learning model. In the method of building a model for evaluating communication qualities,
(a) collecting online text;
(b) collecting survey data for evaluating communication qualities of the online text author of the online text by a survey respondent;
(c) generating communication quality information based on the questionnaire data;
(d) extracting author information of the online text and features from the online text; And
(e) constructing a model capable of evaluating the communication quality information of the author of the online text based on the classification learning model using the feature as an input,
How to build a model for evaluating communication qualities, including. The method of claim 7,
Step (a) is,
The questionnaire data includes a response to a quality parameter for evaluating communication quality,
The qualities parameters include source reliability, interpersonal attractiveness, conversational ability, and interaction intention. A method for building a model for evaluating communication qualities. The method of claim 7,
Step (d) is,
Extracting the feature based on the author information, syntax information, similarity information, emotion information, and linguistic techniques,
The linguistic technique,
Includes language research and word count analysis and cosine similarity,
The language survey and word calculation analysis extracts at least one of emotion, style, survey, and word from the online text,
The cosine similarity is to extract a hash tag, punctuation marks, emoticons, links, and retweets from the online text, and build a model for evaluating communication qualities. In the machine learning model based author's communication quality evaluation device,
A collection unit that collects questionnaire data for evaluating communication qualities of online text authors;
A communication feature information generating unit generating communication feature information for each online text author based on the questionnaire data;
A feature extraction unit for grouping a plurality of online text authors based on the communication feature information for each of the online text authors and extracting features from online text and author information of the grouped online text authors;
A model building unit for constructing a model capable of evaluating the communication quality information of the author of the online text in a preset machine learning method based on the extracted features; And
An evaluation unit for receiving a new online text and evaluating the communication quality information of the author of the new online text through the constructed model,
Apparatus for evaluating communication qualities of authors based on a machine learning model comprising a. The method of claim 10,
The questionnaire data includes a response to a quality parameter for evaluating communication quality,
The qualities parameters include source reliability, interpersonal attractiveness, conversational ability, and interaction intention. The machine learning model-based author communication qualities evaluation device. The method of claim 10,
The feature extraction unit,
Extracting the feature based on the author information, syntax information, similarity information, emotion information, and linguistic techniques,
The linguistic technique,
Includes language surveys and word counting analysis and cosine similarity,
The language survey and word calculation analysis extracts at least one of emotion, style, survey, and word from the online text,
The cosine similarity is to extract at least one of hashtags, punctuation marks, emoticons, links, and retweets from the online text, a machine learning model-based author communication quality evaluation device.

Images