KR20090103171A - Automatic ripple filtering method - Google Patents

Abstract

PURPOSE: A method for automatically filtering replies is provided to filter malignant replies which are inserted in a bulletin board so that they cannot be displayed according to dangerous degrees of each person. CONSTITUTION: Persons who participate in an experiment subjectively evaluate replies(S11). A word database of the replies is generated based on average evaluation indexes of each word which are obtained by an average evaluation index of a common word between a subject and the replies(S12). A user inclination index is measured through a depression test(S13). Evaluation indexes for a newly inputted subject and replies are automatically calculated based on the average evaluation indexes of each word(S14).

Description

Automatic ripple filtering method

The present invention relates to a method for filtering comments, and more particularly, to an automatic filtering method for Internet comments causing personal feelings of shame and shame among various causes of suicide.

The suicides of singers Euni and talent Jung Da-bin are celebrities, so they are known to be a very suicidal event, and they are just the tip of the iceberg of the total population who die from suicide. A comparison of the deaths of young people under the age of 35 who died of suicide and AIDS over the decade between 1987 and 1996 found that nearly 15,000 young people died of suicide than young people who died of AIDS. In 1998, 1.8 percent of the world's 55 million fatalities were suicidal. This suicide was the second leading cause of death among women in their 15-44 year olds, with the second leading cause of death for men. These suicides include serious psychosis, genetic traits, impulsive or violent temperaments, mental illness, manic-depressive illness, emotional ups and downs, difficulties in life events and situations, economic life and occupation, There are a number of causes, including court confrontations, incurable illnesses and disabilities, conditions of shame and disgrace. Among these, suicides caused by external causes include school violence, poor grades, and shame and scorn of constant exposure to Internet comments.

Several studies related to medicine and biology have revealed that the biologics that change due to suicide-causing factors include serotonin levels, 5-HIAA levels in cerebrospinal fluid, noradrenaline levels, and omega-3 essential fatty acids. Long-term investigations have been conducted in a number of subjects to show violent and impulsive behavior when serotonin and 5-HIAA levels are low, node adrenergic neurons are significantly reduced in number, and omega-3 essential fatty acid intake is reduced. The data showed that the risk of suicide increased significantly.

The present invention is to solve the problems of the conventional product as described above, the object is to provide a method for automatically filtering comments to prevent the risk of suicide due to comments among other factors.

In order to achieve the above object, the automatic comment filtering method according to the present invention includes (i) subjects who participated in the experiment subjectively to the comments and the words obtained by the average of the evaluation index for the common words between the topic and comments Creating a word database of said comment with an average rating index; (ii) measuring a user disposition index through the depression test; (iii) automatically evaluating an evaluation index on newly inputted topics and comments based on the word-based average rating index; And (iv) automatically filtering the comments according to the user propensity index.

 Preferably, in the step (i), the words of the comment is the mean rating index for each word is a positive word set consisting of a set of values that the evaluation index is positive, a malicious word set consisting of a set of values that the negative evaluation index, And positive or malicious words are classified according to categories, and are divided into a separate set of malicious words irrespective of a subject and stored in the word database. More preferably, step (iii) comprises: (iii-1) separating each word included in the comments; (iii-2) summing all the evaluation scores of the words included among the words existing in the word database of step (i) among the separated words; (iii-3) determining a result value obtained by dividing the sum of the evaluation indexes by the number of words belonging to the word database as the evaluation index of the comments.

According to another embodiment of the present invention, the method for automatically filtering comments includes (a) an average evaluation index for each word obtained by subjects who participate in the experiment subjectively to the comments and the evaluation index average for the common words between the topic and the comments. Creating a word database of the comment; (b) measuring a user disposition index through measuring suicide values; (c) automatically evaluating an evaluation index on newly inputted topics and comments based on the word-based average rating index; And (d) automatically filtering the comments according to the user propensity index.

Preferably, the user tendency index by measuring suicide value of step (b) is Where α represents the mean of the four physiological suicides of a normal person, and w is a constant that is appropriately adaptively adjusted through the neural network to yield an appropriate rate based on the malignant or benign intensity of the comment. .

The present invention can prevent the death caused by suicide by filtering so that malicious comments posted on the bulletin board is not displayed according to the individual risk.

1 is a structural diagram showing the configuration of an automatic comment evaluation method according to a first embodiment of the present invention.

2 is a structural diagram showing the configuration of an automatic comment evaluation method according to a second embodiment of the present invention.

3 is a view showing an example of a noun extraction form that can be applied to the present invention.

4 is a graph showing the error of the automatic evaluation index in accordance with an embodiment of the present invention.

Figure 5 is a graph showing the result of configuring the training DB with the total data according to an embodiment of the present invention.

6 is a view showing an example of a post sent by the reader of the average rating index according to an embodiment of the present invention.

7 is a diagram illustrating an example of a post sent by a reader having a low propensity index according to an embodiment of the present invention.

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

1 is a structural diagram showing the configuration of an automatic comment evaluation method according to a first embodiment of the present invention.

Devices that measure body-related levels can be divided into non-blood-mechanical devices, such as weight scales and blood pressure gauges, and body-blooded devices such as blood glucose meters that measure blood sugar after the body pierces blood. The four major levels of suicide related serotonin, 5-HIAA, noradrenaline, and omega-3 essential fatty acids belong to the hematopoietic apparatus and should be measured by analysis of cerebrospinal fluid and blood components. Of course, if the wave associated with these values can be measured with an ultrasonic meter, it can also be used as a non-bodied device.

If you can measure four major numerical changes for external events such as life events or circumstances affecting suicide, economic life, court confrontation, incurable illness, shame, and shame, You can use it. Such a suicide numerical measuring instrument has not been developed yet, and the present invention intends to develop a theory under the assumption that there is a virtual measuring instrument.

We will use the four major numerical changes as a measure of shame and scorn. In other words, suppose that you can measure the level of serotonin, 5-HIAA, the number of noradrenaline, and the decrease of omega-3 essential fatty acids. If their figures change for each newspaper article or comment on their topic, then the increase or decrease of these figures can quantify the degree of malignancy and positivity. Depending on the individual's personality and the degree of stress and mood of the day, if the individual's current state reads the comments A and these numbers remain unchanged or increase, then they are positive. If these numbers decrease, they are more likely to be malicious. . In reality, however, it is impossible to judge the propensity of comments using such a measuring apparatus. Accordingly, the present invention proposes a method of classifying the propensity of comments in a realistically possible manner.

Relationship to virtual machines and subjective classification

In the present invention, articles related to themselves or posts related to themselves or posts related to themselves are referred to as topics. In addition, two indexes are introduced to develop the explanation. The first score is the value that a person evaluates or automatically evaluates in the comments. The second score is the value measured by a virtual meter or in the form of a questionnaire.

You can assign a rating index to a large number of comments running on a topic. This evaluation index has a positive value for positive comments that have a positive tendency to the topic, such as praise or encouragement in the topic, and negative for malicious comments that have a negative tendency for content such as human attacks or negative opinions. Define to have a value. 1 and 2, subjects who participated in the experiment subjectively evaluate the comments regardless of the topic. At this time, a positive comment having a positive tendency has a positive value, and a malicious comment having a negative tendency due to content of a human attack or a negative opinion has a negative value (steps S11 and S21).

This evaluation index can be determined by multiplying the propensity index measured by the suicide virtual measurer by a constant constant. The propensity index calculated by the virtual meter may be converted into an evaluation index y through Equation 1 below.

Here, α value represents the average value of four physiological suicide values of a normal person. The w (weight) value is a constant that is adaptively appropriately adjusted through neural networks such as SOM and BP so that an appropriate ratio is obtained according to the malignant or positive strength of the comment. If the difference between the propensity index x and the average propensity index α is small in Equation 1, the evaluation index is set to 0 as the median. Here, when the propensity index x decreases, the evaluation index also has a small value. Therefore, reading comments and having a negative value for the propensity index increases the likelihood of suicide due to mental or other external factors.

However, Equation 1 above is an abstract equation since the suicide virtual counter is currently unavailable. As similar to the virtual measurement value, the experimenter can read the comments on the topic and use the value that evaluates the malignant or positive recognition as a subjective score. Rate the most malicious comments as -5 and the most positive comments as +5. Depending on the subject group of experimenters, the figures will be different, but in games such as gymnastics, the average score can be applied except for the highest and lowest scores. Therefore, the propensity index can be determined by the average value of scores excluding 10% of the low score group and 10% of the high score group. If the number of subjects is large, such as research research, subjective judgment errors on comments may be reduced.

Automatic word classification system

In general, search engines can be divided into text-based and content-based searches. Portal search sites like Empas and Naver are text-based search engines that search based on the words you enter. They also search for pictures or videos based on the title or short description given by the owner of the media file. MPEG-7, which is the basis for video retrieval, applies both text-based and content-based retrieval. The MPEG-7 text-based search engine that searches pictures and videos by text is based on the description of subjects classified by human subjectively. In contrast, classification and retrieval based on unique characteristics of the image itself in a desired image may be classified into an MPEG-7 content-based search engine. For example, content-based retrieval detects each frame as a main frame when the features differ from the previous frame in a sudden scene change, and the characteristic information such as RGB color information and object information in this frame is determined by an absolute value. You can get it. That is, the value obtained from the feature information is always obtained regularly from the function of extracting each feature. Thus, the same scene always gets the same value.

Subjective classification of a topic or comment can be likened to the text-based retrieval of MPEG-7. In other words, it can be said to be a subjective classification engine that classifies people based on the index. This classification method requires a large number of judgment personnel for all topics and comments, and requires a lot of time. In addition, the judgment on the propensity for comments should always be evaluated for new words due to the characteristics of comments updated in real time on the Internet.

Automatic classification of topic or comment requires a basic basis for judging whether it is malignant or benign based on its own criteria for words or sentences. However, judging whether a sentence whose meaning is unknown through a discrimination system is malicious or benign is difficult to set the criterion, and thus cannot be implemented. Therefore, the present invention intends to propose the following method for selecting a criterion selection method for classifying benign and malignant.

In the state of classifying each word included in a comment, it is difficult to always obtain the same value from a function f (x) that determines whether each word is malicious or positive for an existing topic. First, we need to build a word DB (database) to be used as a criterion for evaluation. When constructing the word DB, the evaluation index of each word depends on subjective evaluation values evaluated by the experimenters. The larger the number of test subjects and the higher the number of evaluations for each word, the higher the reliability. Therefore, a word with a low number of evaluations has low reliability in the evaluation index, and thus, a more detailed criterion should be applied in the comment judgment. Determining whether the comment is malignant or benign requires judgment based on the statistics surveyed on the basis of the person, not on the machine, so it is impossible to obtain a regular evaluation index for common words through the machine. Therefore, the reliability of the word evaluation index is bound to be increased by huge data collection and time.

How basic words are organized

Basic words use nouns used in topics and comments. Irrespective of the topic text, the words DB are generated by integrating all the words to which the indexes subjectively evaluated by the people who participated in the experiment (steps S12 and S22). The rating index of each comment gives the same score to all nouns used in that comment. The average score is calculated for each word by the average score of the common words between the comments and the main words. If the evaluation index is composed of positive values, a positive word set is formed, and a negative value set can be composed of a malicious word set. Positive or malicious words can be classified according to categories, but malicious words irrelevant to the subject are stored in a separate set. For example, this topic is about a pictorial pictorial, storing malicious words that are irrelevant to the subject, such as plastic suspicion, rumors, and aggression.

Next, a method of automatically assigning a word evaluation index to the overall evaluation index for one comment will be described below. To get a score for every word used in a comment, you need to construct a broad experimental group to get a long-term subjective score. This is not possible in time or technically. Therefore, the present invention intends to use the word automatic evaluation for the newly input comments based on the evaluation index of each word calculated due to the subjective evaluation index.

How to automatically give word evaluation index

Each word in the malignant and benign word set has a rating index. Index evaluation for unrated general comments can be easily obtained as the average rating index divided by the total number of rating indexes of malignant and benign words included in the article. However, the set of words obtained from the subjective evaluation index is limited. Automatically assigning the evaluation index of the comment based on these words is that if the comment is not related to the current words, the number of words that can determine the evaluation index is small, the index itself becomes meaningless. The newly added word is to give the sum of the evaluation indexes of the comments to which it belongs as an average value divided by the number of these comments.

In this method, the reliability of the evaluation index of the existing word set influences the determination of the evaluation index of the new word. The evaluation index that is automatically assigned to the first word that appears in a comment has very low reliability. However, in the case of words continuously exposed to comments of similar tendency in a plurality of comments, the reliability of the evaluation index may be adjusted autonomously and set to a suitable value.

Reliability test method of automatic evaluation index

There should be a procedure for verifying the accuracy of the evaluation scores that are subjectively measured in existing topics and comments, and the scores that are automatically obtained as described above for new topics and comments. Examination of the evaluation index obtained by the automatic classification method may use the error between the computer-generated automatic evaluation index and the subjective evaluation index average evaluated by humans.

For inspection, when the total number of topics and comments is T, select N topics and comments. Here, the range of N is 1 ≦ N ≦ T / 2. For each comment, several tests are conducted to examine a total of M subjective scores. For each comment, the final evaluation index is given as the average value obtained by dividing the sum of the evaluation scores except 0.9 and M by the highest and lowest 5% points. At this time, 50% of comments on the same topic are divided into Group A and the rest into Group B. Group A is used as an evaluation index that constitutes a word DB that is the basis for word classification and automatic computer judgment. Group B is used to automatically obtain an evaluation index through a computer from the evaluation index of the stored word DB. At this time, the reliability of the automatically evaluated index can be checked through the error between the automatic evaluation index measured in the comments belonging to group B and the subjective evaluation index given by the experimenter.

Measure risk level (user propensity index) through a survey test (S23)

The virtual measurement system for suicide risk is a conceptual abstract system at present. Therefore, the user shaping index was measured by applying a risk value through a depression test as a measurable model for the simulation of the present invention (step S23). The figures are based on the Higher Depression Test, a professional psychological test site sponsored by the Department of Mental Health, Department of Health and Human Services. Depending on the object to be inspected and the number of test items, more precise values can be measured. Although there are many variables depending on the incorrect measurement result due to unsatisfactory answers and the contents of test questions by age, the assumption is that the analysis of propensity close to the average value is possible with the correct answer. The first five of the 30 questions are:

1. I have an appetite.

2. I believe it is best to shut your mouth when you are in trouble.

3. He suffers from the sickness coming up several times a week.

4. It's hard to concentrate on one thing.

5. The neck is rarely painful.

Comments have a high number of malicious posts regardless of risk. Anyone who reads will have a score of -4 or higher that will hurt or raise the risk. In the present invention, regardless of the number of malicious comments will be used together with the filter function existing in the original bulletin board, the posts of the other propensity index will be filtered according to the measured risk value.

Word DB (database) creation (S12, S22) and comment automatic evaluation method (S14, S24)

The automatic comment evaluation method (S14, S24) of the present invention is a method of automatically assigning an evaluation index to a newly input topic and comment based on each word evaluation index such as a subjective search engine. It is necessary to store the evaluation score for each word in the DB in the DB based on the subjective evaluation index of the experimenter for the comment (S12, S22). This DB is the basic data for automatically measuring the evaluation index of each comment. Based on the average rating index of each comment, the words included in the comments are added to the evaluation score in the database, and the number of comments included is increased by one. For example, suppose the word plastic had a total score of 11 and included 3 before. If the newly entered comment has an evaluation score of 2 points and the word 'molding' is included, the total evaluation score of this word is 13 points and the number is 4. Later, the final evaluation score of this word is 3.25 (= 13/4), which is an average value.

The method of evaluating the new comment (S14, S24) is to separate each word included in the comment, so as to sum all the evaluation scores of the words existing in the word DB. If the words included in the new comment are A, B, C, D, E, and the words belonging to the word DB are A, B, D, the evaluation scores of A, B, D are added to the word DB and the number of words in the DB is added. Calculate the rating index for this comment.

Linkage of Bulletin Board View Function and Virtual Measurement System

Social awareness of the serious harm of bad comments, malignant comments, is creating a campaign to vote as a positive comment for the public interest. However, due to the diversity of humanity as well as the various combinations of individual genetic DNAs, malicious comments, either heroic or isolated social dissatisfaction on the Internet, will not disappear regardless of this movement.

Reactions also depend on the personality of the person who accepts the bad. Even though the outwardly active outgoing personality appears, those who are extroverted due to manic depression are more likely to be suicidal than those of general introverted personality. Also, elites who are always honored and oriented to the highest may be at higher risk of suicide due to unknown internal mental stress. People who belong to a high risk group may be mentally depressed or even thought to commit suicide even if they are exposed to malicious comments for a short time. These risks can be difficult to take action around if they are not obvious.

Thus, if you can automatically prevent malicious comments from appearing on bulletin boards for such risky groups, death by suicide could be prevented. However, individual risk measures are less visible when relying on itemized answers, such as a structured depression test. By using the virtual suicide value meter expressed in Equation 1, it can be periodically measured as a blood glucose meter used by people with diabetes. In the computer transmitted through the numerical value obtained through the communication, the comment displayed according to the evaluation index of the malicious comment may be adjusted. For example, if a person's current suicide tendency index is -3.5, then a comment of a's cosmetic index is -3.6 would not be shown, and a -3.5 or higher would be seen. In addition, if the rating score is negative and the malicious comments are read, the score should be added to the stored propensity index and adjusted to increase the suicide risk. Therefore, when you read the next comment, it changes to the ability to view the comment according to the adjusted risk value.

Risk figures and bulletin board

It is based on filtering by the default risk level and the rating score of each comment. However, when exposed to malicious comments several times, those who are at risk of suicide are more easily and strongly motivated to commit suicide. Since there is currently no system that can determine suicides based on responses to malicious and positive comments, we will apply a hypothetical increase or decrease to the risk value obtained from the survey test. In some people, if you keep seeing malicious comments, the impact of the posts can be dull, but in practice, there are many risk factors, so the value of the score for malicious comments is higher than the dynamic increase and decrease. It should be added to the propensity index of about -1, and to adjust the propensity index by adding +0.5 to the value of the evaluation index when viewing positive comments. For example, if you read good articles twice, endorphins increase, and if you read bad articles once, the stress factor builds up, the first measured value should continuously increase or decrease the tendency of good and bad articles. In other words, rather than increasing the number of good articles by 1 and reducing the number of bad articles by 1, the ratio is set by increasing 0.5 when reading good articles and 1 by reading bad articles. Percentage determination needs to be precisely scaled using real virtual measurement systems or through clinical trials. In the present invention, the ratio is fixed at 1: 0.5 to implement the actual algorithm. A ratio of 1: 1 may be applied, but a ratio of 1: 0.5 may not be appropriate because a single shock may last longer than several compliments.

Bulletin board implementation method according to relevance (S15, S25)

1 and 2, the comments evaluated through the word DB are automatically filtered according to the user propensity index measured in step S13 or S23 (S15 and S25). Referring to FIG. 1, the propensity index is adjusted while reading malignant and positive comments, and the comment is displayed again according to the adjusted index when the comment is shown again. Referring to FIG. 2, since the measured value is measured using an accurate value rather than an artificial test such as a depression test, the current state is best reflected. Comments on general newspaper articles are difficult to determine if the reader is relevant to the published article. Therefore, for the posts which are not the authors, filtering is uniformly applied only to the malicious comments in general, that is, the comments having the propensity index of -3 or less. Of course, the read button can be divided into two readers related to the general readers to be read, but the usability of these functions will not be high. Therefore, when the article is seen in the writings of other people, such as newspaper articles, only the comments with a propensity index of -3 or less for the comments should be filtered. Comments on subjects that are the subject of the author show that the evaluation index is less than (-inclination index + 1) / 2 according to the inclination index of the author. In other words, in the case of a high suicide risk score of -3, the criterion of the evaluation index is (+ 3 + 1) / 2 = 2 so that only positive comments with an evaluation index of 2 or more can be viewed. People who are at low risk of suicide can endure a little bit of bad comments, but people who are at high risk can have some bad comments.

Example

In the present invention, subjects were limited to general high schools, and subjective judgment data on subjects and comments posted on portal sites were collected. The data covered a variety of topics such as telme dancing, presidential elections, and unsold apartments. Twenty comments were selected for each topic in the discussion board of the portal site, rather than general newspaper articles, and used as experimental data. A group of 20 topics and comments allowed the experimenters to evaluate their subjects between the most malicious -5 points and the most positive +5 points. Ten people in each group participated in the evaluation. The basic extraction was solved programmatically to extract the nouns contained in each comment, but words could be added manually for the abbreviated nouns such as Nomhyun or the deformed languages such as Dipa (Die Fan) by the language destruction of netizens. It was configured to be. Some forms of extracting words from the simulation program are shown in FIG. 3.

Figure 3 shows an example of a noun extraction form according to an embodiment of the present invention.

In the upper part of FIG. 3, the read topic and comment data files are shown, and the second line shows subjective scores and average scores of subjects for the current comment. The next column shows a space separated string of current comments. The form on the right is a list of nouns that are automatically extracted, manually modified, and added. The classification of automatic nouns due to programming in simulations was limited to basic investigations such as silver, ,,,, and ramen. In fact, the exact noun separation requires a morpheme based algorithm, but in this experiment, only simple morpheme separation was solved by programming.

In this way, we tested the reliability of the automatic evaluation index for comments entered on the data stored as nouns and subjective scores. In order to measure the automatic word evaluation index, randomly select 10 comments on a topic to generate a word DB, and the 10 unselected comments are obtained from the automatic word evaluation index and the actual subjective evaluation data. The error was calculated to determine the reliability. Here, 10 randomly selected comments were divided into training data and the remaining comments into experimental data. Depending on the topic used in this example, the error range was large and small. Figure 4 is a graph representing the total average of the errors for the 10 comments evaluated in one bundle as a value. One value is the mean error for the sum obtained by performing 20 evaluations for each batch.

4 is a graph showing the error of the automatic evaluation index in accordance with an embodiment of the present invention.

As shown in FIG. 4, the range of the error at the value of 11 to −5 to 5 may be 0 to 2, which may be smaller than the range of the evaluation index. However, the collected experimental data had a small number of words, and in a comment containing 10 nouns, there was a case where only one noun matched in the word DB collected from the training data. Thus, the reliability of the error can be solved by a word DB that is sufficiently trained through time and manpower.

In another experiment, we generated a word DB using all the experimental data as training data. The same nouns are included in multiple comments, so when the noun's evaluation index is stored as an overall average, it differs from the original value. Unlike the previous experiment where the training data was randomly selected every time, this experiment uses the entire comment as input data and the training data is fixed. As before, when we experimented 20 times for each bundle, we did not generate different training data each time, so we constructed a word DB with one iteration. However, the experimental data were randomly extracted from 10 and calculated the error between the automatic evaluation index and the subjective evaluation index. In this experiment, the average of the total errors obtained by repeating a total of 20 times is represented in a graph in FIG. 5.

5 shows the results of configuring the training DB with all the data. Compared to the previous experiment, the nouns that occur in each comment do not exist in the word DB, so the error is significantly reduced, but due to the nouns included in the various comments, there is no satisfactory level of reliability. The more experimental data you have, the greater the number of duplicate nouns will reduce the margin of error. Of course, in the case of a large number of subjects, such as general research, it will be difficult to have a confidence level of more than 95%, depending on subjects' subjective criteria. By constructing a word DB with the experimental data as a whole, we measured the accuracy of the filtering algorithm that shows the comments on the topic according to the reader's personal inclination index. Participants in this experiment were 10 times the total number of questions divided by total number of questions according to the general depression test test item, and then subtracted -5 to equalize the range of -5 to +5, which is the evaluation index of existing comments. It was. Based on the program-measured value, we simulated comment filtering on the assumption that the experimenter who reads the current post is reading.

Referring to FIG. 1, after performing step S15, the user tendency index is corrected according to the comment point (step S16). 6 is a view showing an example of a post sent by the reader of the average rating index according to an embodiment of the present invention, the average index is 1.28 for the rating index of each comment evaluated through the word DB for the topic and comments This post only shows comments with a rating of 1.28 2 =-0.72. At this time, the evaluation index of each comment is larger than 1.28, which is the current reader's propensity index value, to modify the propensity index by adding (rating index-propensity index) / 10 to the propensity index. Conversely, if the valuation index is smaller than the propensity index, the (valuation index propensity index) / 2 is added. In other words, the comments positive for the current reader having a value larger than the propensity index increase the propensity index little by little, whereas the negative comments having a value smaller than the propensity index are rapidly lowered. In other words, the index was applied to the extent to which the reader suddenly receives mental stress due to malignancy.

Referring to FIG. 2, after performing step S25, the propensity index through the suicide value measurement is recalculated (step S26). FIG. 7 is a post seen when the propensity index is extremely low as a serious experimenter with depression. In this experiment, we can see that the generated word DB is affected a lot. The comment scores that generated the word DB were also subjective, and the initial depression test for the posts consisted of subjective evaluations of the experimenters. As mentioned in the first half of the specification, those with prospective stress who are expected to be around, and those who do not release internal stress through the depression test, are cases where this subjective test itself is useless. These are the types that can suddenly kill themselves one day for no reason. Therefore, in order to achieve more accurate filtering, it is necessary to filter comments by reading each comment by using the proposed suicide virtual meter, which is reflected in the tendency index.

The method according to the present invention is a basic theoretical invention which attempts to solve the current suicide in association with IT. In the medical field, medical engineering with advanced technologies such as electro-electronic computer engineering and physics is applied to practical work. As suggested by the present invention, if medical science and IT are combined, medical theory may be combined with the Internet or UCC to make human life more profitable. The present invention is an invention in which only a simple IT function is applied to medical theory, but if the actual suicide measurer is introduced in medical engineering, medical engineering and IT may be combined in reality. Child violence has a serious social anxiety factor that causes the subject to grow and cause another violence or suicide. If the proposed method is applied to hearing aids of children who are being abused, the sounds that can be used for child violence such as parents or relatives can be used to filter the sounds according to the stress of the current child. In the future, the medical engineering and IT fields, including the brain science field, will be studied more professionally and applied to real life.

Claims (9)

(i) generating a word database of the comments by means of word-by-word average ratings obtained by subjects evaluating the comments subjectively and evaluating average scores for common words between the topic and the comments; (ii) measuring a user disposition index through the depression test; (iii) automatically evaluating an evaluation index on newly inputted topics and comments based on the word-based average rating index; And (iv) filtering the comments automatically according to the user propensity index.  The method of claim 1, wherein the comment is a malicious comment or a positive comment according to the serotonin level, the 5-HIAA level of the cerebrospinal fluid, the number of node adrenergic systems, and the degree of reduction of omega-3 essential fatty acids for each comment on the topic. How to filter comments automatically.  According to claim 1, wherein the words of the comment in the step (i) is the average word-by-word average rating index is a positive word set consisting of a set of values that the evaluation index is positive, a malicious word consisting of a set of values that the negative evaluation index Positive or malicious words are classified according to a set and a category, and are automatically stored in the word database by dividing them into a separate set of malicious words irrelevant to a subject. The method of claim 1, wherein step (iii) (iii-1) separating each word included in the comments; (iii-2) summing all the evaluation scores of the words included among the words existing in the word database of step (i) among the separated words; and (iii-3) determining a result value obtained by dividing the sum of the evaluation indexes by the number of words belonging to the word database as the evaluation index of the comments. The method of claim 1, wherein in step (iv), when the post is viewed by a person other than the writer's subject, the inclination index of the comment is filtered to be a comment of -3 or less, and the subject of which the subject is the writer is Comments are automatically filtered to prevent only posts with an evaluation index of (-Disposition Index + 1) / 2 from being posted and only view positive comments with (-Disposition Index + 1) / 2 or higher, depending on the author's disposition index. . The method of claim 1, further comprising: (v) modifying the user propensity index according to a comment point. (a) subjecting the experiment subjectively to the comments, and generating a word database of the comments with an average rating index for each word obtained as an average of the evaluation indexes for the common words between the topic and the comments; (b) measuring a user disposition index through measuring suicide values; (c) automatically evaluating an evaluation index on newly inputted topics and comments based on the word-based average rating index; And and (d) automatically filtering the comments according to the user propensity index. According to claim 7, wherein the user tendency index through the suicide value measurement of step (b) is Where α represents the mean of four physiological suicides in a normal person, and w is a constant that is appropriately adaptively adjusted through the neural network to yield an appropriate rate based on the malignant or benign intensity of the comment. How to auto filter comments. The method of claim 7, further comprising: (e) recalculating a propensity index through the suicide measure.