KR102359466B1 - Deep metadata based emotion analysis method and system - Google Patents

Abstract

A deep metadata-based sentiment analysis method and system are disclosed. A sentiment analysis system implemented by a computer according to an embodiment includes: a crawler module for collecting data related to postings; a data pre-processing module for pre-processing the collected data; a sentiment analysis module for obtaining a sentiment score for a sentiment word filtered based on a pre-established sentiment dictionary from the pre-processed data; and a service module for providing a service related to the sentiment analysis information of the post by using the acquired sentiment score.

Description

Deep metadata-based sentiment analysis method and system {DEEP METADATA BASED EMOTION ANALYSIS METHOD AND SYSTEM}

The description below relates to the technology for analyzing emotions.

Sentiment Analysis is a research field that analyzes people's opinions, emotions, evaluations, and attitudes in texts. In addition, sentiment analysis is based on identifying positive, negative, or neutral in a given text, and this is called polarity of the text. Such sentiment analysis is being used as an important element in business based on user emotions as individual expression of opinions is diversified through social media. In fact, it is being conducted several times to analyze subjective attitudes and emotions. However, Korean sentiment analysis often excludes elements other than Korean. For accurate sentiment analysis, it is important to interpret the entire text. In particular, emoji is an abbreviation of facial expression and a non-verbal element, reflecting the human psychological state in the text to supplement the content.

It is possible to provide a deep metadata-based sentiment analysis method and system.

It is possible to provide a method and system for obtaining a sentiment score by filtering crawled data related to a post based on a sentiment dictionary.

Using the acquired emotional information, it is possible to provide a service related to the sentiment analysis information of the post.

The computer-implemented sentiment analysis system includes: a crawler module for collecting data related to postings; a data pre-processing module for pre-processing the collected data; a sentiment analysis module for obtaining a sentiment score for a sentiment word filtered based on a pre-established sentiment dictionary from the pre-processed data; and a service module for providing a service related to the sentiment analysis information of the post by using the acquired sentiment score.

The sentiment analysis module filters sentiment words based on a pre-established Korean sentiment dictionary and emoji sentiment dictionary from the pre-processed data, and a sentiment score set in the filtered sentiment words in the filtered sentiment words can be matched to calculate the sentiment score of the post.

The emotion analysis module includes constructing an emotion dictionary including a Korean emotion dictionary and an emoji emotion dictionary for emotion analysis in the form of metadata, and the emoji emotion dictionary includes a Unicode name (Unicode Name). ), CLDR, and sentiment dictionary data including scores, and the Korean sentiment dictionary may store sentiment dictionary data including sentiment words, roots, and scores.

The sentiment analysis module may assign a weight to the score of each sentiment word included in the Korean sentiment dictionary based on the sentiment score of the post.

The sentiment analysis module compares the root of the result of analyzing the morpheme in the data of the post with the root of the sentiment word existing in the pre-constructed sentiment dictionary as the pre-processing is performed, and gives a sentiment score to the result of the analysis of the morpheme A first method of giving a sentiment score to a word of a post that matches a word of a post with a sentiment word existing in the pre-established sentiment dictionary, or a second method of giving a sentiment score to the word of the post When the emotion word and the word of the post are compared and the word of the post does not exist in the pre-established emotional dictionary, the root of the result of morpheme analysis in the data of the post and the root of the emotion word as the pre-processing is performed By comparison, an emotional score may be given by using any one of the third methods for assigning an emotional score to the result of analyzing the morpheme.

When the CLDR of the morpheme analysis result of the post and the CLDR of the emoji sentiment dictionary coincide with the pre-processing, the sentiment analysis module may assign a score to the morpheme analysis result.

The sentiment analysis module may calculate and represent the positive or negative values of the posting as a percentage by counting the number of words by distinguishing the words representing emotions in the posting for each score or each weighted score.

The sentiment analysis module may calculate a ratio of the number of sentiment words in a score range of a preset range to the total number including the sum of the emotional words existing in the posting to indicate the true or negative value of the posting.

The service module may represent the emotional change by analyzing the data of the post as a graph as the user logs into the service with the right of the right holder, and may represent emotional weather according to the positive or negative value of the post.

The service module may display statistical information including a sentiment value and emotion distribution of the post, display a post classified based on a preset category, and highlight an area corresponding to a sentiment word in the post.

The service module provides a service screen for individual user management of the sentiment analysis information of the post, and uses any one communication method including a comment and a note on the user's post displayed through the provided service screen. As the manager's response is performed, the result of the sentiment analysis may enable contact with the anxious user.

The crawler module crawls the address of the post as it accesses the web page, accesses the post through the address of the crawled post, crawls the body and comments of the accessed post, and stores it in each list, and the After storing each of the stored lists in the data frame, it is possible to store the list stored in the data frame as a JSON file having a post address as a file name.

The data pre-processing module removes noise data from the title and body content of the post included in the collected data, and combines words with different expression methods in the title and body content of the post from which the noise data has been removed to create the same word After performing text normalization, space is placed on the normalized text, and then morpheme analysis is performed, the morpheme-analyzed words can be stored in JSON format by distinguishing the body and title of the post.

A sentiment analysis method performed by a computer-implemented sentiment analysis system includes: collecting data related to postings; performing pre-processing of the collected data; obtaining a sentiment score for the filtered sentiment word based on a pre-established sentiment dictionary from the pre-processed data; and using the acquired sentiment score to provide a service related to sentiment analysis information of the post.

By building an emoji sentiment dictionary and a Korean sentiment dictionary, you can improve the accuracy of sentiment analysis for posts.

In addition to sentiment analysis, management and posting of services related to emotional state information can be performed integrally.

1 is a diagram for explaining the structure of a sentiment analysis system according to an embodiment.
2 is an example for explaining storing data in the sentiment dictionary in the sentiment analysis system according to an embodiment.
3 is a flowchart illustrating an operation of crawling data in a sentiment analysis system according to an embodiment.
4 to 6 are examples for explaining a method of storing a list as a JSON file in the sentiment analysis system according to an embodiment
7 is a flowchart illustrating an operation of performing data pre-processing in the sentiment analysis system according to an exemplary embodiment.
8 is an example for explaining an operation of matching sentiment words in the sentiment analysis system according to an embodiment.
9 to 14 are diagrams for explaining an operation of providing a service related to emotional state information analyzed in the emotion analysis system.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

1 is a diagram for explaining the structure of a sentiment analysis system according to an embodiment.

The sentiment analysis system 100 is for analyzing emotions, and may include a crawler module 110 , a data preprocessing module 120 , a sentiment analysis module 130 , and a service module 140 .

The crawler module 110 may crawl data of posts for sentiment analysis. A post may mean an article posted/uploaded for communication or information sharing on the Internet, such as SNS or a community. In this case, the crawler module 110 may crawl data not only for text data but also for video data or audio data using a function such as Speech to Text (SST). The crawler module 110 may record crawled data together with metadata surrounding the corresponding time point. For example, the crawler module 110 may record social indicators such as site information, category information, comments, and sharing. Alternatively, the crawler module 110 may record related advertisement exposure information (eg, URL-based, information is also stored if necessary).

The data pre-processing module 120 may pre-process the data in a form that enables sentiment analysis according to the purpose of the text data. Specifically, the data preprocessing module 120 may perform category classification, refinement and normalization, morphological analysis, and the like.

The sentiment analysis module 130 may input the preprocessed processed data to filter the sentiment words, and may calculate the total sentiment score by matching the sentiment score set in advance. The sentiment analysis module 130 may give weights to the scores of individual sentiment words included in the Korean sentiment dictionary based on the sentiment scores of the entire post. In this case, weight information can be managed based on specific criteria such as category, period, and site, and a separate weight can be set through manager intervention. When a system is configured using ML/DL, a weight parameter, etc. may correspond. The sentiment analysis module 130 may be repeated more than 'N (N is a natural number)' times based on the time point, and if there is a word that is not registered in the dictionary, it is determined as a new word, and the determined new word is reported to the manager to quickly It can provide functions that can be added to the system.

For example, the sentiment analysis module 130 may set a sentiment vocabulary (word) in order to filter sentiment words from the preprocessed processed data. The sentiment analysis module 130 may construct the Korean sentiment dictionary and the emoji sentiment dictionary in the form of deep metadata in which deep learning and metadata are combined. In this case, the detailed metadata information may be configured in a form that can be combined with weight information. For example, detailed metadata information may include basic information such as the site, category, and crawling time of the posting, frequency and position vector values with specific surrounding words may be recorded, and ML/DL may be used. In this case, weight information and neural network weight parameters necessary for configuring a corresponding network may be included. Also, in the case of data extracted from the STT, the sentiment analysis module 130 may use a negative pitch value as weight information.

' 표시는 CLDR에서 Loudly Crying Face로 변환될 수 있다. 감성 사전 데이터는 JSON 파일 형태로 저장될 수 있다. 예를 들면, 한국어 감성 사전은 단어, 어근, 점수, 이모지 감성 사전은 유니코드 이름(Unicode Name), CLDR, 점수 순으로 사전을 구성할 수 있다. 예를 들면, 형태소 분석기는 한국어를 지원하는 'MeCab'을 이용할 수 있다. 한국어 감성 사전은 감성 어휘, 어근, polarity 순으로, 이모지 감성 랭킹은 유니코드(Unicode), CLDR Short Name, polarity 순으로 사전을 구성할 수 있다. 실시예에서는, 분석된 감성 어휘 및 이모지를 감성 요소라고 하며, polarity를 감성 지수라고 기재하기로 한다. 일례로, 소셜 미디어에 작성된 게시글 등을 수집하고 형태소 분석을 통해 감성 요소가 추출될 수 있다. 한국어 감성 사전과 이모지 감성 사전을 비교하여 추출된 감성 요소를 필터링하고, 각 사전과 일치하는 필터링된 감성 요소는 필터링된 감성 요소에 해당하는 값과 그에 따른 점수를 반환하여 게시글의 긍정 또는 부정 점수를 나타낼 수 있다. The constructed sentiment dictionary can be used as a basis for sentiment analysis. The Korean emotional dictionary can be composed of emotional words that represent universal basic emotional expressions of humans, and the deep learning technique can be used to categorize the emotions of the standard Korean dictionary. The emoji sentiment dictionary is indicated by utilizing the emoji sentiment ranking, and the sentiment dictionary may be formed with the most frequently used emoji among data sets for a specific service (eg, tweet). The Korean sentiment dictionary and the emoji sentiment dictionary may be composed of different criteria, and the emoji sentiment dictionary of the Korean sentiment dictionary and the emoji sentiment dictionary composed of these different criteria may be normalized and used. Referring to FIG. 2 , an example of saving a JSON file of the Korean sentiment dictionary and an example of saving a JSON file of the emoji sentiment dictionary are shown. The emoji emotion dictionary construction method can be stored in advance by converting it to a CLDR format rather than an emoji format using an emoji module. In this case, CLDR may mean locale information provided by the Unicode Consortium in XML format. For example, emoticons'

' mark can be converted from CLDR to Loudly Crying Face. The sentiment dictionary data may be stored in a JSON file format. For example, the Korean sentiment dictionary may configure the dictionary in the order of word, root, score, and the emoji sentiment dictionary, Unicode name, CLDR, and score. For example, the morpheme analyzer may use 'MeCab' that supports Korean. The Korean emotional dictionary can be composed in the order of emotional vocabulary, root, and polarity, and the emoji emotional ranking in the order of Unicode, CLDR Short Name, and polarity. In the embodiment, the analyzed emotional vocabulary and emoji are referred to as emotional elements, and polarity is described as emotional index. For example, an emotional element may be extracted through collection of posts written on social media and the like and morphological analysis. By comparing the Korean sentiment dictionary and the emoji sentiment dictionary, the extracted emotional elements are filtered, and the filtered emotional element matching each dictionary returns the value corresponding to the filtered emotional element and the corresponding score to give the positive or negative score of the post. can represent

The sentiment analysis module 130 may add the word and score to the sentiment dictionary when, upon sentiment analysis, if there is an emotional word among the words that are not filtered, it is reported to a separate manager and approved as a neologism word. At this time, if it is determined as a neologism even before the intervention of the user (administrator), it may be used as a weight.

Specifically, referring to FIG. 8 , an operation of matching emotional words will be described. For example, in the case of Hangul, the emotion analysis module may compare only the root of the result of morpheme analysis from the post with the root of the emotional word, and may give a score to the root of the result of the morpheme analysis (Method 1). As another example, the sentiment analysis module may give a score to the word of the post when the word included in the sentiment dictionary completely matches the word of the post (Method 2). As another example, after applying method 2, the sentiment analysis module may apply method 1 when the corresponding word does not exist in the dictionary. In the case of an emoji, when the CLDR of the morpheme analysis result from the posting matches the CLDR of the emoji sentiment dictionary, the emotion analysis module 130 may give a score to the morpheme analysis result.

The emotion analysis module 130 may calculate the emotion score in various ways and store it in a form that can be displayed on a web page. For example, in the case of college student emotion monitoring, positive/negative scores for each post, emotional scores for each parking, emotional scores for each category, and the like can be derived. In this case, various methods may exist for the sentiment analysis module to calculate and display the positive/negative score (numerical value) as a percentage. For example, the sentiment analysis module may count the number of words by discriminating words representing emotions in one post for each score or for each weighted score. For example, the number of words divided into very anxious (-2 to -1), anxious (-1 to 0), average (0), good (0 to +1), and very good (+1 to +2). can be counted. As another example, the sentiment analysis module may derive a ratio of each emotion within one post. For example, the ratio of the number of emotional words in each score range to the total number that means the sum of the emotional words excluding the word with 0 points (words that do not have emotions) may be calculated (the number of emotional words in each score/total number of emotional words) Count). As another example, the sentiment analysis module may determine that a value obtained by adding very good and good is positive, and a value obtained by adding very anxiety and anxiety is negative. As another example, the sentiment analysis module may reclassify the emotion of the post into a preset step (eg, step 5) through the difference between positive and negative. For example, very anxious (-50% to -25%), anxious (-25% to -5%), moderate (-5% to +5%), good (5% to 25%), very good ( 25% to 50%).

The service module 140 may provide a service screen (user interface) for providing the analyzed emotional state information. The service module 140 may provide a login screen, a dashboard screen, a statistics screen, a category screen, and a setting screen. The service module 140 may provide access to related data when logging in with administrator authority through the login screen. The service module 140 may express a monthly statistical graph of emotional change of the analyzed data through the dashboard screen. The service module 140 may express statistical graphs for each emotional change period (eg, monthly, weekly, daily), emotional values, and emotional distribution through the statistical screen. The service module 140 may express the categorized data through the category screen and express the detailed emotional score. The service module 140 may provide to set user setting information through a setting screen.

Specifically, with reference to FIGS. 9 to 14 , it is a diagram for explaining an operation of providing a service related to the analyzed emotional state information. The service may be performed by the service module, and the service screens of FIGS. 9 to 13 may be provided through the service module. 9 shows a login screen, and when logging in with an administrator account, it is possible to access data analyzed by emotion. At this time, as the ID and password of the administrator account are input, the service module can be accessed. In addition, the administrator account may be set in advance, and may become an administrator account as access rights are changed to an administrator later.

Referring to FIG. 10 , a dashboard screen is shown. After the emotion is analyzed, the dashboard screen may be expressed as emotional weather according to the positive or negative values of all posts (1). The emotional weather criterion may be calculated for all posts through the same process as step 5 of post sentiment analysis. After the emotion is analyzed, the dashboard screen may be expressed as emotional weather according to the positive or negative values of posts in the category (2). The emotional weather criterion may be calculated for postings within a category through the same process as in step 5 of post sentiment analysis. On the dashboard screen, the change in the emotional score may be expressed as a bar graph for each date in which the positive or negative values of all postings are preset for each date (eg, by a week, by a month, etc.) (3). A calendar is displayed in the FullCalendar library on the dashboard screen, and the academic calendar for two semesters (from March to February of the next year) can be displayed as a list (4).

11 and 12 , a statistical screen is shown. On the statistics screen, positive or negative values of all posts may be expressed as a line graph with respect to the change in emotional score for a preset period (eg, by one week) for each date (5). On the statistics screen, the number of posts corresponding to each emotional state among emotional posts for a specific period (eg, one week) may be expressed in a radial graph and numerical values (6). On the statistics screen, positive or negative values for all sentiment posts for a specific period (eg, one week) may be expressed as a donut chart (7). On the statistics screen, the number of posts belonging to a category among emotional posts for a specific period (eg, one week) may be expressed.

When a day is selected from among data (eg, one week data) for a preset period on the statistics screen, the emotional score value of the selected day of the week may be expressed ( 9 ). Alternatively, the radial graph, the donut chart, and the emotional score of the day when all the numerical values are selected on the statistics screen may be expressed. Posts can be categorized in the stats screen. As an example, posts may be statically categorized. For example, posts can be classified by setting candidates for a category in advance. As another example, posts may be dynamically categorized. For example, content mentioned above a preset standard among analyzed data may be set as a category. In this way, you can check the topic of the day at a glance, and at the same time you can know the emotions and interests of the students. Also, if my post falls under multiple categories, it can be marked according to the policy. For example, it is possible to indicate the category in the form of a label to allow duplication of categories, or to indicate the category in the form of a dictionary to express individual categories.

Referring to FIG. 13 , a category screen is shown. Posts classified by category may be displayed on the category screen. For example, a positive or negative value may be calculated and displayed as a percentage. A positive or negative value calculated as a percentage may be expressed as emotional weather (10). Referring to FIG. 14 , it may be provided so that key words can be identified at a glance through the emotion word highlighting function on the category screen ( 11 ). In this case, by the sentiment analysis module, the word of the post that matches the 'word' of which 'exist' is 1 may be highlighted. When highlighting, it can be provided to enable quick analysis by using the color temperature difference or color difference according to the degree of positive or negative. In addition, it may be possible to turn on/off the highlighting function on the category screen. In addition, sentiment words can be listed as keywords on the category screen. On the category screen, the emotion level of the post calculated by the sentiment analysis module may be displayed in the form of a gauge (12). It may be provided so that individual management of each user can be performed by an administrator for sentiment analysis information of a post on the category screen. On the category screen, the counselor can directly provide a communication method such as a comment or a note to enable contact with the author who seems anxious about the result of the sentiment analysis (13). For example, the manager (counselor) may directly input a response message to the user's post by using sentiment analysis information analyzed for the user's post on the category screen.

Additionally, if the emotion analysis result shows a certain periodicity pattern, the service module may predict the emotion of the post creator and proceed with a program that can respond in advance. In addition, the service module can notify the risk by sending an e-mail to the administrator when there are 3 or more posts in a negative emotional state per day. In addition, if the negative emotional state persists for more than 3 days or more than 70% of the total number of posts are negative, the service module can send SMS to the counselor to proactively notify the risk.

3 is a flowchart illustrating an operation of crawling data in a sentiment analysis system according to an embodiment. The crawling operation may be performed by a crawler module of the sentiment analysis system.

The crawler module may crawl (320) the post address as the web page is accessed (310). The crawler module can crawl the post address existing on each page for each bulletin board. For example, the crawler module can store posting addresses in the form of a list (address[]). The crawler module may access 330 the post based on the crawled post address. In other words, the crawler module can access each post through a list in which addresses are stored.

The crawler module may crawl (340) the content of the posting. Specifically, the crawler module may crawl the body of the post and the comments and store them in each list. For example, the crawler module may crawl the title, body content, writing time, number of recommendations, number of comments, author of comments, comment content, and comment time of the body of the post. In this case, the crawler module may periodically crawl based on a predetermined time. As an example, the crawler module may periodically crawl the body and comments of a post every time (for example, 4 hours) determined to be appropriate compared to the number of posts written in a day as a result of analyzing posts made for a preset period. . As another example, the crawler module may crawl when more than a certain number of posts are written. For example, as a result of analyzing data for a preset period, the body and comments of the post are based on the posts generated on average at each interval of time (eg, 4 hours) that is judged appropriate compared to the number of posts made in a day. can be crawled.

The crawler module may store (350) the stored list as a data frame (D) as the body and comments of the post are crawled. The crawler module can crawl and store the contents included in the body of the post and the list stored by crawling the comments in the data frame. Specifically, the crawler module crawls and stores the content included in the title, body content, writing time, number of recommendations, and number of comments of the post, and the crawled list of comment authors, comment content, and comment time. Each frame (D1, D2) can be stored separately. This is to separate the content of the main text and the comment part.

The crawler module may store 360 the list as a JSON file. 4 to 6 , it is an example for explaining a method of storing a list as a JSON file. The crawler module can save the post address as a JSON file with the file name. For example, the crawler module can generate one JSON file for each post. One JSON file: D1 (body, D2 (comment)) can all be saved in one JSON file. If D1 is saved as a single JSON file, D2 is stored for each address (e.g., post address). It can be created as a JOSN file.

7 is a flowchart illustrating an operation of performing data pre-processing in the sentiment analysis system according to an exemplary embodiment. The data pre-processing operation may be performed by the data pre-processing module of the sentiment analysis system.

The data pre-processing module may import raw data (Raw Data Import) 710 . The data preprocessing module can preprocess only the post title and body content, excluding additional information such as post ID and post creation time. The data pre-processing module may refine 720 the data. The data pre-processing module may remove noise data (eg, words that do not represent sentiment from the collected data). For example, '$, #, /, +, =. &, @' can be removed, and symbols '(, ), :, ;, ^, *, <, >' that can represent emoticons are not removed.

The data pre-processing module may normalize ( 730 ) the text as the noise data is removed. Normalization is the process of generating the same word by integrating words with different expression methods. In the data preprocessing module, when 'ㅋ, ㅎ, ㅠ', etc. repeatedly appear more than a certain number of times, it can be normalized into the form of the same score by unifying it into 'ㅋ', ''heh'', ''ㅠ', etc.

The data pre-processing module may execute the spacing module ( 740 ). If the spacing in the data is not done properly, an error may occur during morphological analysis. The data preprocessing module can prevent errors by using the spacing module. For example, you can use Chatspace (Korean space model), KoSpacing and RAWS. KoSpacing is a CNN-RNN structure of the Korean automatic spacing package for accurate results before morpheme analysis, and RAWS constructs a basic model by borrowing character-level embedding techniques. In the case of the Mecab morpheme, if an emoticon is repeated multiple times, it is recognized as one, so a space is placed before and after the emoticons to separate them.

The data preprocessing module may analyze 750 the morphemes in the data. The data pre-processing module can run the Mecab stemming analyzer on the spaced data. As a result of morphological analysis of data, if the part-of-speech is 'SY (symbol)', it can be converted into CLDR using the emoji module and stored.

The data pre-processing module may store ( 760 ) the analysis result. The data preprocessing module can store the stemmed words in JSON format by distinguishing titles and posts. The data preprocessing module performs morphological analysis with elements such as word name, part-of-speech tag, semantic classification, support presence, and reading when executing Mecab. can be saved

The device described above may be implemented as a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, devices 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 array (FPGA). , a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose or special purpose computers. The processing device may execute 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 the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

Software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or apparatus, to be interpreted by or to provide instructions or data to the processing device. may be embodied in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, although the embodiments have been described with reference to the limited embodiments and drawings, various modifications and variations are possible from the above description by those skilled in the art. For example, the described techniques are performed in an order different from the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (14)

In the emotional analysis system implemented by a computer,
a crawler module that collects data related to posts;
a data pre-processing module for pre-processing the collected data;
a sentiment analysis module for obtaining a sentiment score for a sentiment word filtered based on a pre-established sentiment dictionary from the pre-processed data; and
A service module that provides a service related to the sentiment analysis information of the post by using the acquired sentiment score
including,
The pre-built emotional dictionary is,
A sentiment dictionary including a Korean sentiment dictionary for sentiment analysis and an emoji sentiment dictionary was built in the form of deep metadata combined with deep learning and metadata.
The emotion analysis module,
From the pre-processed data, sentiment words are filtered based on a pre-established Korean sentiment dictionary and emoji sentiment dictionary, and sentiment scores set in the filtered sentiment words are matched to the filtered sentiment words to match the sentiment of the post. Calculate the score, but compare the root of the result of analyzing the morpheme in the data of the post on which the pre-processing has been performed with the root of the emotional word existing in the pre-built Korean emotional dictionary and emoji emotional dictionary. A first method of giving an emotional score, a second method of giving an emotional score to a word of a matching post when the emotional word existing in the pre-established Korean emotional dictionary and emoji emotional dictionary matches the word of the post; or When the word of the post does not exist in the pre-established Korean sentiment dictionary and emoji sentiment dictionary, as the word of the post is compared with the emotional word contained in the pre-established Korean sentiment dictionary and emoji sentiment dictionary, the above In the post data on which preprocessing has been performed, the root of the result of the morpheme analysis is compared with the root of the emotional word, and the emotion score is given by using any one of the third methods of giving the emotion score to the result of the morpheme analysis. and, when the CLDR of the morpheme analysis result of the post on which the preprocessing has been performed matches the CLDR of the emoji sentiment dictionary, giving a score to the morpheme analysis result,
The service module is
Providing a service screen for individual user management of emotional state information including the emotional value and emotional distribution of the post, and sentiment analysis information of the post
Sentiment Analysis System. delete According to claim 1,
The emoji emotion dictionary stores emotion dictionary data including a Unicode name (Unicode Name), CLDR, and score,
The Korean sentiment dictionary is to store sentiment dictionary data including sentiment words, roots, and scores.
Sentiment analysis system, characterized in that. According to claim 1,
The emotion analysis module,
Based on the sentiment score of the post, weight is given to the score of each sentiment word included in the Korean sentiment dictionary.
Sentiment analysis system, characterized in that. delete delete According to claim 1,
The emotion analysis module,
By counting the number of words by distinguishing the words expressing emotion in the post for each score or for each weighted score, the positive or negative value of the post is calculated and displayed as a percentage.
Sentiment analysis system, characterized in that. According to claim 1,
The emotion analysis module,
By calculating the ratio of the number of emotional words in a preset range to the total number including the sum of the emotional words existing in the posting,
Sentiment analysis system, characterized in that. According to claim 1,
The service module is
As the user is logged into the service with the right of the right holder, the emotional change by analyzing the data of the post is shown in a graph, and the emotional weather is shown according to the positive or negative value of the post.
Sentiment analysis system, characterized in that. According to claim 1,
The service module is
Displaying statistical information including the emotional value and emotional distribution of the post, displaying the post classified based on a preset category, and highlighting the area corresponding to the emotional word in the post
Sentiment analysis system, characterized in that. According to claim 1,
The service module is
As the manager's response is performed using any one communication method including a comment and a note to the user's post displayed through the provided service screen, it is possible to contact the user who is anxious about the emotional analysis result.
Sentiment analysis system, characterized in that. According to claim 1,
The crawler module is
As the web page is accessed, the address of the post is crawled, the post is accessed through the address of the crawled post, and the body and comments of the accessed post are crawled and stored in each list, and each of the stored lists After saving in the data frame, the list stored in the data frame is saved as a JSON file with the post address as a file name.
Sentiment analysis system, characterized in that. According to claim 1,
The data pre-processing module,
Remove noise data from the title and body content of the post included in the collected data, and perform text normalization to generate the same word by integrating words with different expression methods in the title and body content of the post from which the noise data has been removed And, after executing the space in the normalized text, the morpheme analysis is performed to distinguish the morpheme-analyzed words from the body of the post and store them in JSON format.
Sentiment analysis system, characterized in that. In the sentiment analysis method performed by a computer-implemented sentiment analysis system,
collecting data related to postings;
performing pre-processing of the collected data;
obtaining a sentiment score for the filtered sentiment word based on a pre-established sentiment dictionary from the pre-processed data; and
providing a service related to the sentiment analysis information of the post by using the acquired sentiment score;
including,
The pre-built emotional dictionary is,
A sentiment dictionary including a Korean sentiment dictionary for sentiment analysis and an emoji sentiment dictionary was built in the form of deep metadata combined with deep learning and metadata.
The obtaining step is
From the pre-processed data, sentiment words are filtered based on a pre-established Korean sentiment dictionary and emoji sentiment dictionary, and sentiment scores set in the filtered sentiment words are matched to the filtered sentiment words to match the sentiment of the post. Calculate the score, but compare the root of the result of analyzing the morpheme in the data of the post on which the pre-processing has been performed with the root of the emotional word existing in the pre-built Korean emotional dictionary and emoji emotional dictionary. A first method of giving an emotional score, a second method of giving an emotional score to a word of a matching post when the emotional word existing in the pre-established Korean emotional dictionary and emoji emotional dictionary matches the word of the post; or When the word of the post does not exist in the pre-established Korean sentiment dictionary and emoji sentiment dictionary, as the word of the post is compared with the emotional word contained in the pre-established Korean sentiment dictionary and emoji sentiment dictionary, the above In the post data on which preprocessing has been performed, the root of the result of the morpheme analysis is compared with the root of the emotional word, and the emotion score is given by using any one of the third methods of giving the emotion score to the result of the morpheme analysis. and if the CLDR of the morpheme analysis result of the post on which the preprocessing has been performed matches the CLDR of the emoji sentiment dictionary, giving a score to the morpheme analysis result,
The step of providing the service comprises:
Providing a service screen for individual user management of emotional state information including the emotional value and emotional distribution of the post, and sentiment analysis information of the post
Sentiment analysis method comprising.

Images