KR102502454B1 - Real-time comment judgment device and method using ultra-high-speed artificial analysis intelligence - Google Patents

Abstract

A real-time comment judgment method using high-speed analysis artificial intelligence is disclosed. The comment judgment method of the present invention collects real-time comments, extracts morphemes, analyzes them in a quantified emotion analysis model, obtains the results of quantified emotion analysis, displays them, or configures them to be fed back to the broadcasting subject, thereby utilizing artificial intelligence capable of high-speed analysis. Since real-time comments can be analyzed and visualized or quantified and displayed through the technology of the present invention, there is an effect of effectively analyzing and managing satisfaction and immediately utilizing real-time feedback.

Description

Real-time comment judgment method using high-speed analysis artificial intelligence {REAL-TIME COMMENT JUDGMENT DEVICE AND METHOD USING ULTRA-HIGH-SPEED ARTIFICIAL ANALYSIS INTELLIGENCE}

The present invention relates to a real-time comment judgment method, and specifically, by using artificial intelligence capable of high-speed analysis, real-time comments are quantified as positive/negative or out of 5 points to effectively analyze and manage satisfaction, and can be immediately utilized as real-time feedback It is about a real-time comment judgment method using high-speed analysis artificial intelligence.

A social network service (SNS) is a one-person community that allows individuals to form personal networks while sharing various information such as personal daily lives, interests, and discussions on specific topics online.

In addition to SNS, in data-based sharing services such as mobile messengers, when users register posts such as photos, videos, and texts, acquaintances such as friends, seniors, and family members who have built a relationship with the user leave comments or simple emoticons at the bottom of the post. You can exchange your regards or your opinions.

In this way, comments posted in real time by viewers and listeners on YouTube, Afreeca TV, home shopping, performances, sports games, and various live broadcasting platforms, where the number is increasing, are very difficult for the broadcaster to manage due to the amount and speed, and are related to satisfaction. It is difficult to use feedback in real time.

Moreover, in the case of popular broadcasts, there are so many comments that it is impossible to check in real time, so even the broadcasters or viewers watching together cannot read them with fast scrolling.

In other words, it is difficult to know whether there are more positive comments or more negative comments.

KR Registration Patent No. 10-2078627 (2020.02.12)

Therefore, through the technology of the present invention using artificial intelligence capable of high-speed analysis, real-time comments can be quantified and displayed as positive / negative or out of 5 points, etc. do.

In order to solve this problem, the real-time comment determination method using ultra-high-speed analysis artificial intelligence that analyzes the comments broadcast in real time and gives feedback in real time in the device of the present invention includes the steps of (a) collecting comments included in posts, ( b) extracting morphemes from the comments and inputting at least one of the morphemes into a quantified sentiment analysis model, (c) obtaining quantified sentiment analysis results from the quantified sentiment analysis model by using artificial intelligence capable of high-speed analysis and (d) feeding back or displaying the obtained quantified emotion analysis result to the broadcasting entity of the device in real time.

In addition, step (a) includes collecting a plurality of comments included in the postings of the device, and step (b) recognizes and extracts slang words, abbreviations, and abbreviated expressions from the comments as morphemes, , Morphemes are extracted from all of the plurality of comments.

In addition, in step (c), nouns extracted from the morphemes, slang words, abbreviations, and abbreviated expressions extracted from the morphemes, and particles, verbs, adjectives, special characters, uppercase letters, and lowercase letters extracted from the morphemes and inputting at least one of emoticons into a quantitative analysis model.

And the step (d) compares the output value of the emotion analysis model with one or more predetermined reference values to classify and display the type of emotion of the comment, or satisfaction expressed as textual comments input in real time for a certain interval of time. can be quantified to visualize and display positive, neutral, and negative trends over time.

Therefore, according to the real-time comment judgment method using ultra-high-speed analysis artificial intelligence of the present invention, real-time comments can be visualized or quantified and displayed as positive/negative or out of 5 points through the technology of the present invention using artificial intelligence capable of high-speed analysis. Therefore, it has the effect of effectively analyzing and managing satisfaction and using it immediately as real-time feedback.

1 is a flowchart for explaining a real-time comment judgment method using ultra-high-speed analysis artificial intelligence according to an embodiment of the present invention;
2 is a diagram for explaining a method for generating a quantified emotion analysis model according to an embodiment of the present invention;
3 is a diagram showing an example of a quantification analysis model according to an embodiment of the present invention;
4 is a main configuration diagram for implementing a real-time comment judgment method using ultra-high-speed analysis artificial intelligence according to an embodiment of the present invention;
5 is a diagram showing output values of a quantified emotion analysis model as positive, negative, or neutral;
6 is a diagram illustrating the time required for comment analysis using an artificial intelligence model;
and
7 is a diagram illustrating an example of execution of comment analysis using an artificial intelligence model.

The terms or words used in this specification and claims are not limited to the usual or dictionary meanings, and the inventor can properly define the concept of the term in order to explain his or her invention in the best way. Based on this, it should be interpreted as a meaning and concept consistent with the technical spirit of the present invention.

Throughout the specification, when a certain component is said to "include", it means that it may further include other components without excluding other components unless otherwise stated. In addition, terms such as “… unit”, “… unit”, “module”, and “device” described in the specification mean a unit that processes at least one function or operation, which is implemented as a combination of hardware and/or software. It can be.

Throughout the specification, the term "and/or" should be understood to include all possible combinations from one or more related items. For example, the meaning of “a first item, a second item and/or a third item” may be presented from two or more of the first, second or third item as well as the first, second or third item. A combination of all possible items.

Throughout the specification, identification codes (e.g., a, b, c, ...) for each step are used for convenience of explanation, and the identification code does not limit the order of each step, and each step Unless the specific order is clearly stated in context, it may occur in a different order from the specified order. That is, each step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

1 is a flowchart for explaining a real-time comment judgment method using ultra-high-speed analysis artificial intelligence according to an embodiment of the present invention. As shown, the comment judgment method of the present invention quantifies comments using ultra-high-speed analysis artificial intelligence in real time. It is characterized in that it is performed through a quantified emotion analysis model programmed for the emotion analysis method, and it is displayed or fed back.

In other words, the present invention relates to a method for determining comments in real time using ultra-high-speed analysis artificial intelligence that analyzes comments being broadcast in real time on a predetermined device and gives feedback in real time. Extracting morphemes from the comments collected in step S110 and inputting at least one of the extracted morphemes into a quantified sentiment analysis model (S120), quantitative sentiment analysis using artificial intelligence capable of high-speed analysis from the quantified sentiment analysis model Obtaining a result (S130) and displaying the quantified emotion analysis result obtained in step S130 and providing feedback to the broadcasting entity of the device in real time (S140).

To this end, as shown in the configuration diagram of FIG. 4, the present invention includes a comment collection unit 110 that collects comments from posts and a morpheme extraction input unit 120 that extracts morphemes from the comments collected by the comment collection unit 110; A quantification analysis model unit 130 that analyzes the morphemes extracted from the morpheme extraction input unit 120 through a quantification sentiment analysis model, and a display unit 140 that visually displays the analyzed result or gives feedback to the broadcasting entity of the device in real time. can be done through

The comment collection unit 110, the morpheme extraction input unit 120, the quantification analysis model unit 130, and the display/feedback unit 140 can be made in one terminal such as a PC and a smartphone.

That is, in the step of collecting comments included in the postings (S110), the comment collection unit 110 operates to collect comments from the postings being broadcast in real time on the device.

The comment collection unit 110 is a device that collects comments included in posts in the device. When a user registers a post such as a photo, video, article, etc. Acquaintances, such as friends, seniors, and family members who have built a network with , collect comments from various information by leaving comments or simple emoticons at the bottom of posts.

In addition, you can collect comments on posts from various sites. For example, one or more comments on a specific post posted in a site such as a social network service (SNS), news articles, blogs, reservation sites, and shopping malls may be collected.

When collecting comments, information on the site where the post was posted may be collected. For example, information on the source of the post may be collected among SNS, news site, blog, reservation site, and shopping mall. Information on the topic or category of the posting, for example, information on categories such as politics, current events, movies, entertainment, sports, games, and travel may also be collected. Information on the content type of the post, text, photo, video, etc. may be collected.

In addition, the comment collection unit 110 may collect one or more plural comments included in the postings of the device and allow the morpheme extraction input unit 120 to extract morphemes from all of the plural comments.

The morpheme extraction input unit 120 extracts morphemes from the replies collected in the comment collection unit 110 and allows the quantified sentiment analysis model unit 130 to analyze the sentiment of the comments using artificial intelligence.

The morpheme extraction unit 120 recognizes and extracts slang words, abbreviations, and abbreviated expressions from comments as morphemes, and may extract morphemes from all of the plurality of comments. ) is entered.

For reference, the abbreviated language can be "Jjang, Kwang, ^^, TT" and the like.

In addition, the morpheme extraction input unit 120 inputs at least one of the extracted nouns and the extracted postpositions, verbs, adjectives, special characters, uppercase letters, lowercase letters, and emoticons from among the morphemes into the quantitative analysis model unit 130 for analysis. .

The quantitative analysis model unit 130 analyzes the morphemes extracted from the morpheme extraction input unit 120 with artificial intelligence to obtain a quantified emotion analysis model as a result of emotion analysis.

The quantification analysis model unit 130 may compare the output value of the quantification sentiment analysis model with one or more predetermined reference values, classify the type of emotion of the comment, and display it on the display/feedback unit 140, or at regular intervals. It is also possible to quantify the comments entered in real time during the period, visualize the trend of positive, neutral and negative over time and display them on the display/feedback unit 140, and quantify the real-time comments into positive/negative or 5-point scales and display/display them. It can also be displayed on the feedback unit 140.

Referring to the drawing in which the output value of the quantified sentiment analysis model of FIG. 5 is displayed as positive, negative, or neutral, comments input during a set time at regular intervals are quantified and the trend is shown by time, and the positive, neutral, and negative Comments are displayed, and the weight is indicated by the number of circles.

In the figure, it indicates that there are many negative comments during the set time.

Referring to the drawing, when the set time elapses, the comments are analyzed for the set time again and displayed with positive, neutral and negative weights, and the emotion analysis result within the set time is displayed in real time in a streaming format.

That is, the present invention is a system that analyzes posted comments in real time and gives feedback in real time, rather than analyzing registered comments.

Referring to the diagram illustrating the time required for comment analysis using the artificial intelligence model of FIG. 6 , it can be seen that the present invention is a technology capable of real-time analysis.

6 shows that when using the artificial intelligence model programmed by the present invention using the latest computer (i9-9000, Nvidia GTX2080 super GPU), sentiment analysis through morphological analysis of 5383 thousand sentences takes 512 ms, so 100 per case This is shown by an execution example that took only ms.

In addition, the quantification analysis model unit 130 may calculate a probability corresponding to each type of emotion of a comment, calculate a satisfaction score of the comment based on the calculated probability, and display the result on the display unit 140 .

Referring to a drawing illustrating an example of an execution of comment analysis using an artificial intelligence model of FIG. 7 , a type of emotion expressed in each comment and calculated as a probability is exemplified.

For example, if a comment is posted saying, "It was a very fun and informative class, and I liked how he took good care of the students," it is displayed as a 99.21% positive review, so that the broadcaster can view the content of the comment in real time. is to make it understandable.

The comment quantification analysis method and display method will be described in detail below.

Referring back to FIG. 1 , when comments are collected in step S110, the morpheme extraction unit 120 refines the text of the comments collected in the comment collection unit 110, extracts morphemes, and extracts at least one of the extracted morphemes. It is input to the quantification analysis model unit 130 (S120).

Comments can be collected in the form of emoticons, slang words, abbreviations, and abbreviated expressions other than text.

For example, the collected comments may be refined in a predetermined method. As the text is refined, it may be classified into at least one of nouns, postpositions, verbs, adjectives, special characters, uppercase and lowercase letters of the alphabet, and the like.

When at least one of the morphemes is input to the quantification analysis model unit 130 according to a predetermined criterion, the quantification analysis model unit 130 analyzes emotion using a predetermined quantification emotion analysis model.

For example, a quantified sentiment analysis model extracts the contents of a post, performs a supervised learning method through logistic regression analysis to analyze the post, learns the contents of the comment, and new comments based on a machine learning learning model such as deep learning. Based on the result of the regression equation, the emotion of the content is a model for determining whether a comment is positive, negative, or has other emotions.

A method for generating a quantified sentiment analysis model will be described below.

Also, among the classified morphemes, only nouns may be input, or at least one of postpositions including nouns, verbs, adjectives, special characters, capital letters of the alphabet, and emoticons may be input.

In step S140, the quantified emotion analysis result is obtained from the quantified emotion analysis model and displayed or fed back to the broadcaster in real time.

For example, the quantitative emotion analysis result may include at least one of emotion types and emotion scores. Types of emotions include positive and negative emotions, but neutral emotions that are not biased may be included, and a quantified emotion analysis result in which positive and negative emotions are subdivided may be obtained. The quantification analysis model may output a quantification sentiment analysis result as a numerical value.

For example, if a value of 75% positive and 25% negative is output, the emotion type can be output as positive based on the value, and 75 points can be output as an emotional score, or an award based on a star rating or a predetermined criterion. It may be displayed as , middle, or lower.

For example, if a noun is entered into a quantified sentiment analysis model, the quantified sentiment analysis model can determine whether the content of the comment is positive or negative or contains other emotions based on the nouns included in the comment, and additionally, the type of emotion. The degree corresponding to may be expressed numerically as a score.

Hereinafter, a method of deriving a quantified emotion analysis result from a quantified emotion analysis model will be described with reference to the drawings.

2 is a diagram for explaining a method for generating a quantification analysis model according to an embodiment of the present invention. As shown, the quantification sentiment analysis model performs a supervised learning method through logistic regression analysis as described above. It can be created by learning the contents of

To this end, comments may be collected from posts posted on various sites, and various learning data 210 may be generated using each comment. The learning data 210 includes information about comments and their sources (posts and sites), information about commenters, etc., and the learning result 230 corresponding to the training data 210 of the quantification sentiment analysis model is output. Learning data 210 may be selectively input.

The learning data 210 outputs a learning result 230 through artificial intelligence learning 220, which is linear regression machine learning, and this learning result can be generated as a learning result 230, which is a deep learning-based emotion analysis model. there is. The learning result 230 may include emotion types and/or emotion scores corresponding to the comments. As in the embodiment, comments can be extracted and stored at specific points of time and generated as a quantified sentiment analysis model 240 through machine learning on the corresponding comments.

In addition, the quantified sentiment analysis model 240 applies a weight to a specific word or a specific part of speech based on the user's propensity to consume sites that post comments or in consideration of the age group and gender of the main user during learning. can be learned

For example, different weights may be applied and learned according to the main age group that consumes news articles or the gender that uses a shopping mall.

Alternatively, when the input is determined by the emotion analysis model, weights may be set for morphemes to be input.

Hereinafter, an embodiment using an artificial intelligence-based quantitative emotion analysis model will be described. It goes without saying that the quantified sentiment analysis model may be differently generated according to the input from which the quantified sentiment analysis model is learned.

3 is a diagram showing an example of a quantification sentiment analysis model according to an embodiment of the present invention. As shown, the collected comments are collected from comments on posts posted on the site at a specific time point, as described above. It can include site information, post information, etc. to be input to the sentiment analysis model.

Quantitative sentiment analysis models may correspond to machine-learned models by applying weights to specific words or specific parts of speech based on the propensity of users who consume sites that post comments or considering the age group and gender of the main user. can

Quantitative sentiment analysis models can receive only refined nouns as input through pre-processing of comments, and at least one keyword among investigations other than nouns, verbs, adjectives, special characters, uppercase and lowercase letters of the alphabet, emoticons, slang, abbreviations, and abbreviated expressions. can be input into a quantified emotion analysis model along with a noun.

For example, an embodiment that receives grammatical relationships including each noun included in a comment and a postposition linked to the noun as input, an embodiment that receives nouns and verbs of all comments, and receives special characters included in nouns and comments Examples and the like are possible, and a quantified emotion analysis model for each embodiment can be created differently.

As described above, emotion types and/or emotion scores may be included as outputs of the quantitative emotion analysis model.

For example, a quantified emotion analysis model may numerically output probabilities of emotions corresponding to inputs. It is possible to calculate the probability of each of the two extreme types of emotions of 'positive' and 'negative', or to calculate the probability of each of the emotions such as 'joy', 'lo', 'love', and 'rock', respectively. For the output of quantification sentiment analysis model can be trained in a different form.

Emotion types may be divided into positive, neutral, and negative based on a regression line according to a predetermined reference value in a quantified emotion analysis model. For example, when the learning result of the emotion analysis model for a specific comment is 'positive' 72% and 'negative' 25%, the emotion type is highly likely to be determined as 'positive'. In addition, the emotion score may be determined according to the probability learned from the emotion analysis model, and in the case of the above example, the score of 'positive 72 points' is determined, the difference between the probability of 'positive' and the probability of 'negative', or other Scores may be calculated by a calculation method.

The embodiments described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, the devices, methods and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate (FPGA). array), programmable logic units (PLUs), microprocessors, or any other device capable of executing and responding to instructions. A processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. You can command the device. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method for determining comments in real time using high-speed analysis artificial intelligence according to an embodiment of the present invention may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to 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 and DVDs, and magnetic media such as floptical disks. - To store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, etc.

A specially configured hardware device is included. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the present invention has been described in detail with respect to the specific embodiments described above, it is obvious to those skilled in the art that various changes and modifications are possible within the scope of the technical idea of the present invention, and it is natural that these changes and modifications fall within the scope of the appended claims.

110: comment collection unit 120: morpheme extraction input unit
130: quantification model analysis unit 140: display / feedback unit

Claims (8)

In the real-time comment judgment method using ultra-high-speed analysis artificial intelligence that analyzes comments being broadcast in real time on a device and gives feedback in real time,
(a) collecting a plurality of comments included in the posts of the device
(b) extracting morphemes from the comments and inputting at least one of the extracted morphemes into a quantitative sentiment analysis model;
(c) acquiring emotion analysis results by using artificial intelligence capable of high-speed analysis from the quantified emotion analysis model; and
(d) displaying the emotion analysis result or feeding back the acquired emotion analysis result to the broadcasting entity of the device in real time;
including,
The step (b) is
Recognize and extract slang words, abbreviations, and abbreviated expressions in the comments as morphemes, extract morphemes from all of the plurality of comments,
nouns extracted from the morphemes;
slang, abbreviation, and abbreviated expression extracted from the morphemes; and
Inputting at least one of the extracted morphemes from among the morphemes, verbs, adjectives, special characters, uppercase letters, lowercase letters, and emoticons into a quantitative analysis model.
Real-time comment judgment method using high-speed analysis artificial intelligence including.
delete delete delete The method of claim 1,
The step (d) is
A real-time comment judgment method using ultra-high-speed analysis artificial intelligence that compares the output value of the emotion analysis model with one or more predetermined reference values to classify and display the type of emotion of the comment.
The method of claim 1,
The step (d) is
Real-time comments using super-high-speed analysis artificial intelligence that visualizes and displays positive, neutral, and negative trends by time by quantifying the satisfaction of the quantification analysis model results as sentence-type comments that are input in real time at regular intervals from the output value of the model. judging method.
The method of claim 6,
The step (d) is
Real-time comment judgment method using high-speed analysis artificial intelligence to quantify and display the real-time comments as positive / negative or out of 5 points.
The method of claim 1,
The step (d) is
A real-time comment judgment method using high-speed analysis artificial intelligence that calculates a probability corresponding to each type of emotion of the comment and calculates an emotion score of the comment based on the calculated probability.

Images