Document analysis system for analyzing document by using artificial intelligence to identify keyword and emotional state of document creator and identify relationships among group members

Abstract

The present invention relates to a document analysis system for analyzing a document by using artificial intelligence to identify a keyword and an emotional state of a document creator and identify relationships among group members, wherein the system may analyze, using artificial intelligence, documents created by group members in everyday activities so as to extract a main keyword and a main emotion of a document creator, and analyze, using the extracted main keyword and main emotion, members who are associated with each other in a group or as a pair, and a member who cannot get along with other members and is isolated, among the group members so as to intuitively provide same. According to the present invention, since daily documents created by members of a group are analyzed, it is possible to realistically analyze the events and emotions of the members unlike intentional psychological surveys, and it is possible to effectively identify relationships among members within a group and whether isolation exists therein, whereby quick and effective counseling for mentally distressed members can be provided.

Description

A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and to identify relationships among group members.

[0001] The present invention relates to a document analysis system that analyzes documents using artificial intelligence to identify relationships among group members, and more specifically, to a document analysis system that analyzes documents written by group members to identify emotions and events, and based on this, analyzes how group members are connected and who are isolated members, and provides documents that are analyzed using artificial intelligence to identify the emotional state and keywords of the document writer and identify relationships among group members.

[0002] Modern people face serious psychological challenges due to a variety of factors, including social pressures, economic instability, rapid technological change, and interpersonal relationships. Problems such as anxiety, depression, and stress are becoming increasingly common.

[0003] Anxiety and depression are among the most common mental health problems today, affecting millions of people worldwide. As awareness of the importance of mental health grows, the need for diagnosis for mental health conditions is also increasing.

[0004] A variety of factors, including the influence of social media, economic pressures, work-life imbalances, and interpersonal relationships between students and employees, play a crucial role in exacerbating mental health issues. Understanding these factors is essential for diagnosing and treating mental health conditions.

[0005] Accurate diagnosis and understanding of an individual's psychological state are crucial for the effective treatment and management of mental health issues. Diagnosis allows for the development of a personalized treatment plan tailored to each individual's unique needs, ensuring more effective and targeted treatment.

[0006] For students, identifying and treating psychological conditions is crucial for their overall well-being and academic success. A student's mental health is crucial for their emotional, physical, and psychological well-being. Addressing mental health issues can significantly impact students' learning, participation, and academic performance.

[0007] For employees, mental health significantly impacts productivity, job satisfaction, and the overall workplace environment. Implementing mental health programs in the workplace can help identify and address issues early, ultimately reducing absenteeism, increasing productivity, and fostering a healthier work environment.

[0008] Understanding the psychological state of group members, including students and working professionals, and identifying and treating problems early are essential for individual happiness, academic success, and professional productivity.

[0009] There are various methods for understanding the psychological state of group members. Among them, the following are prior patents related to understanding the psychological state using documents written by group members and prior patents related to understanding the relationship between members.

[0010] Patent Publication No. 10-2024-0050710 (Method for providing a conversation-based emotion diary, computer device, and computer program) automatically records emotions according to the conversation context in a conversation scenario with a virtual character to create an emotion diary, thereby identifying the user's daily emotions and providing various reports on the emotion diary, such as changes in emotions or distribution of emotions over a unit period, thereby enabling the user to easily and conveniently check their emotional state.

[0011] Patent No. 10-2642589 (Personalized Content Recommendation System through Diary Analysis) analyzes daily record data such as text and voice voluntarily written by the user based on machine learning to identify the user's emotions and situations, and recommends necessary content to the user based on this, thereby enabling accurate recommendation of content preferred by the user, thereby increasing the convenience and satisfaction of the user's content selection, and recommending content tailored to the user's ever-changing emotions and interests.

[0012] Publication Patent No. 10-2023-0030693 (Mind Management Maintenance System) analyzes the user's mental health or mental health more accurately through autonomous records such as a diary to accurately analyze the user's daily appearance, classifies the user, displays the results of the user's diagnosis and classification as a visualization display means, and provides customized content that suits the characteristics of the user.

[0013] Patent Publication No. 10-2023-0067888 (Peer Relationship Diagnosis Server and Peer Relationship Diagnosis Method Using the Same) comprises the steps of transmitting a selected psychological diagnosis problem to a student terminal and receiving the result data of the psychological diagnosis problem transmitted from the student terminal in response thereto, the step of generating a peer relationship chart of students who are members of a psychological diagnosis class based on the result data of the psychological diagnosis problem, the step of analyzing the generated peer relationship chart to calculate a risk index for each student, and the step of generating a psychological diagnosis report for teachers that includes the risk index.

[0014] Services have been developed to analyze various types of individual emotions, create relationship maps between members, and calculate individual risk indices. However, these services only analyze individual emotions and provide personalized services. Furthermore, to understand relationships between members, separate psychological assessment questions are sent to the user, resulting in a psychological assessment. However, these methods have limitations, making it impossible to identify natural relationships between members based on their everyday behavior.

[0015] The present invention was invented to improve the above problems, and to provide a document analysis system that analyzes documents written by group members in their daily lives using artificial intelligence to extract the main emotions and main keywords of the document writers, and uses the extracted main emotions and main keywords to analyze which of the group members are related to each other as a group or close friends, and which members are isolated and unable to get along with other members, and to provide an intuitive system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writers and to identify the relationships between group members.

[0016] The present invention is not a method that uses psychological surveys or psychological diagnosis questions, but rather analyzes documents that naturally record daily life, such as diaries commonly written by members of a group, so that realistic analysis of the emotions and daily events of members is possible, and based on this, whether the member is happy or depressed and how he or she gets along with other members of the group is possible. The present invention provides a document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and to identify the relationships between members of the group.

[0017] In order to achieve the above-described purpose, a document analysis system according to the present invention, which analyzes a document using artificial intelligence to identify the emotional state and keywords of a document writer and to identify relationships between group members, includes an emotion extraction unit that analyzes the context of a document written by a group member to extract a set number of main emotions of the document writer, a keyword extraction unit that analyzes the context of the document to extract a set number of main keywords, and a member relationship analysis unit that identifies relationships between group members using a plurality of extracted main emotions and main keywords and analyzes and provides connections and isolation between group members.

[0018] The above emotion extraction unit,

[0019] It is composed of an artificial intelligence emotion extraction module that analyzes multiple emotions listed in the emotion list using artificial intelligence and extracts the main emotions of the document writer that appear in the context of the document from among the multiple emotions listed in the emotion list, and an emotion scoring module that comprehensively calculates the emotion score of the document writer based on the emotion score for each emotion extracted by the artificial intelligence emotion extraction module.

[0020] Specifically, the above emotion scoring module,

[0021] The artificial intelligence emotion extraction module divides each emotion extracted into positive, neutral, and negative, and calculates the document writer's emotion score based on the emotion classification score table that assigns a score to each emotion.

[0022] The above member relationship analysis department,

[0023] Based on the documents of the same date of the members, it is determined whether the main keywords are similar among the members (1), and based on the documents of the same date of the members, it is determined whether at least one emotion type matches among multiple emotions and the difference in the overall emotion score is below a certain level (2). Members who satisfy both of these (1, 2) for keywords and emotions are determined to be related to each other.

[0024] Specifically, the member relationship analysis unit,

[0025] If the number of days that members satisfy both keywords and emotions (1, 2) exceeds a certain level in the recent period, the members are judged to be related as a group or buddy. If the members are not related to each other and their negative emotions exceed a certain level in the recent period, the members are judged to be isolated.

[0026]

[0027] If we look specifically at the above member relationship analysis section, it is structured as follows.

[0028] The above member relationship analysis department,

[0029] The above keyword extraction unit includes an event sharing judgment module that judges similarity between multiple keywords extracted from documents of group members on the same date and determines that an event is shared between the corresponding members if the similarity is above a certain level, an emotion sharing judgment module that judges that an emotion is shared between the corresponding members if at least one emotion type matches among multiple emotions extracted from documents of group members on the same date by the artificial intelligence emotion extraction module and a difference in the comprehensive emotion score of the same-date documents of the corresponding members calculated by the emotion scoring module is below a certain level, and a member relationship judgment module that judges that in the case where three or more members share an event and emotion together and the difference exceeds a certain level over a recent period of time as a result of the judgment of the event sharing judgment module and the emotion sharing judgment module, the corresponding members are judged to be in the same group if the difference exceeds a certain level over a recent period of time, the corresponding members are judged to be close friends if the corresponding members do not belong to a group or close friends and the corresponding members are judged to be isolated if the negative emotion exceeds a certain level over a recent period of time.

[0030] And the above member relationship analysis department,

[0031] The method further includes an outlier judgment module that analyzes the comprehensive emotional scores of members by date for a recent period of time and determines as outliers members who have exceeded a certain number of dates corresponding to values ​​that deviate from the average or median by a certain level, and the member relationship judgment module determines as risk members members who do not belong to a group or a close friend, whose negative emotions exceed a certain level for a recent period of time, and who are determined as outliers by the outlier judgment module.

[0032] In addition, the above member relationship analysis department,

[0033] The member relationship judgment module further includes a cheering phrase output module that analyzes the connection or isolation between group members of each member judged by the member relationship judgment module and the keywords and emotions that serve as the basis for judging the connection or isolation of each member using artificial intelligence, and generates cheering phrases that reflect the connection or isolation and the keywords and emotions and provides them to each member.

[0034]

[0035] Meanwhile, the above member relationship analysis department,

[0036] The above keyword extraction unit may include a centrality display module that determines the similarity between multiple keywords extracted from documents of members on the same date, connects members with a similarity level above a certain level with a relationship line, and displays the relationship line in a manner that indicates that it is connected with a positive or negative emotion when at least one emotion type matches among multiple emotions based on documents of the same date of the members connected by the relationship line and the difference in the comprehensive emotion score is below a certain level, and displays the relationship line in a manner that indicates that it is connected with an unrelated emotion when there is no emotion type matching among multiple emotions based on documents of the same date of the members connected by the relationship line and the difference in the comprehensive emotion score is above a certain level.

[0037] The above central display module is,

[0038] Multiple relationship lines can be displayed between members by additionally displaying the above relationship lines for each document.

[0039] The above member relationship analysis department,

[0040] It may further include an individual median calculation module that calculates the median for each member by summing the scores corresponding to the number and type of relationship lines connected to each member.

[0041] And the above member relationship analysis department,

[0042] A lifestyle judgment module may further be included that determines a member as an insider if the median of the members is higher than a certain level, determines a member as an outsider if the median of the members is lower than a certain level, and determines a member as a risky member if there are many lines connected by negative emotions in the relationship line.

[0043] In addition, the above member relationship analysis department,

[0044] A median change observation module can be further included to determine whether the median of a member is increasing or decreasing through daily observation, and if the median is decreasing, to determine that the member's situation is becoming isolated and to notify the member and the member's manager of the situation.

[0045] The present invention utilizes a method that extracts emotions and keywords by analyzing daily documents commonly written by group members, rather than through direct psychological surveys. This allows for a precise analysis of each member's daily emotions. Furthermore, keywords within these documents can be used to accurately analyze the events of the day. Using these analyzed emotions and events, it is easy to understand how members are connected to one another. Furthermore, it is easy to identify which members are isolated from the group.

[0046] Managers at work or teachers at school can use analysis reports to monitor how members relate to each other and check the mental health of each member based on their daily emotional changes.

[0047] And it is possible to immediately identify which members are not getting along with other members of the group and are in a dangerous state, so that by caring for and taking interest in those members, the mental health of the members can be improved.

[0048] Figure 1 is a configuration diagram of a document analysis system that analyzes a document according to the present invention using artificial intelligence to identify the emotional state and keywords of the document writer and to identify relationships between group members.

[0049] Figure 2 is an example diagram that visually shows the connection and isolation between group members using main emotions and main keywords.

[0050] Figure 3 is an example of an emotion classification score table showing the emotion score for each emotion in the emotion list.

[0051] Figure 4 is an example of a centrality indicator that represents the connection and emotions between group members as relationship lines.

[0052] The advantages and features of the present invention and the method for achieving them will become clear with reference to the embodiments described in detail below together with the attached drawings.

[0053] However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms.

[0054] The embodiments in this specification are provided to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention.

[0055] And the present invention is defined only by the scope of the claims.

[0056] Accordingly, in some embodiments, well-known components, well-known operations, and well-known techniques are not specifically described to avoid obscuring the present invention.

[0057] Additionally, like reference numerals throughout the specification refer to like elements, and the terms used (referred to) herein are for the purpose of describing embodiments and are not intended to limit the present invention.

[0058] In this specification, the singular includes the plural unless specifically stated otherwise in the phrase, and the reference to an element or action as “including (or comprising)” does not exclude the presence or addition of one or more other elements or actions.

[0059] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in the sense commonly understood by a person of ordinary skill in the art to which the present invention belongs.

[0060] Also, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless they are defined otherwise.

[0061] Hereinafter, a preferred embodiment of the present invention will be described with reference to the attached drawings.

[0062] Referring to FIGS. 1 to 4, a document analysis system according to the present invention, which analyzes documents using artificial intelligence to identify the emotional state and keywords of document writers and to identify relationships between group members, comprises an emotion extraction unit (100), a keyword extraction unit (200), and a member relationship analysis unit (300). Documents written by group members and uploaded from a PC or smartphone are stored in a document database (400). The emotion extraction unit (100) and the keyword extraction unit (200) analyze documents of group members stored in the document database (400).

[0063] The emotion extraction unit (100) analyzes the context of a document written by a group member and extracts a set number of key emotions of the document writer. The number of key emotions extracted can be set by the system.

[0064] Group members include students from the same school, employees from the same company, etc., and any group that usually lives together as a group can be included.

[0065] Documents include diaries written by students and journals written by office workers.

[0066] The keyword extraction unit (200) analyzes the context of the document and extracts a set number of main keywords.

[0067] The member relationship analysis unit (300) uses the extracted multiple main emotions and main keywords to identify the relationships between group members and analyzes and provides the connection relationship and isolation between group members.

[0068] Referring to Figure 2, this is an example of a member relationship analysis unit (300) analyzing and providing the connection and isolation between students in a class by using the main emotions and main keywords of students in the same class.

[0069]

[0070] The emotion extraction unit (100) includes an artificial intelligence emotion extraction module (110) and an emotion scoring module (120).

[0071] The artificial intelligence emotion extraction module (110) analyzes multiple emotions listed in the emotion list using artificial intelligence and extracts the main emotion of the document writer that appears in the context of the document from among the multiple emotions listed in the emotion list.

[0072] Below is an example of a list of 45 emotions.

[0073] [List of emotions]

[0074] 'Warmth, love, hate, anger, irritation, tension, relief, calmness, happiness, joy, sadness, compassion, regret, remorse, shame, anxiety, timidity, embarrassment, fear, surprise, disgust, antipathy, generosity, ignorance, loneliness, solitude, longing, depression, boredom, hope, passion, excitement, giving up, disappointment, frustration, admiration, spring, hope, satisfaction, pride, pleasure, gratitude, wish, worry, fatigue'

[0075]

[0076] Referring to Figure 1, the artificial intelligence emotion extraction module (110) analyzes the written document written by group member A today using artificial intelligence and extracts the main emotions of group member A that appear in the context of the document from among the 45 emotions listed in the emotion list above. Similarly, the context of the document is analyzed for group members B and C to extract the main emotions of the document writer.

[0077] The artificial intelligence emotion extraction module (110) analyzes the context of documents written by group members, such as diaries, to extract the author's main emotions. It uses natural language processing (NLP) to identify the context and analyze sentence structure to extract emotions. A more detailed look at natural language processing (NLP) is as follows.

[0078] First, preprocessing is performed to organize the document text and remove unnecessary information. Important features are extracted from the text and input into a sentiment analysis model, using word embeddings. The sentiment analysis model uses machine learning to classify the document's sentiment. Representative algorithms include Naive Bayes and Support Vector Machines (SVM).

[0079] When the sentiment analysis model classifies the sentiment of a document using machine learning, it extracts the sentiment from among the 45 sentiments listed in the sentiment list above.

[0080] The emotion scoring module (120) comprehensively calculates the emotion score of the document writer based on the emotion score for each emotion extracted by the artificial intelligence emotion extraction module (110).

[0081] The emotion scoring module (120) divides each emotion extracted by the artificial intelligence emotion extraction module (110) into positive, neutral, and negative, and calculates the emotion score of the document writer based on an emotion classification score table that assigns a score to each emotion.

[0082] Referring to Figure 3, the above list of 45 emotions was divided into positive (extroversion / introversion), neutral, and negative (extroversion / introversion), and each emotion was scored. "Happiness" received the highest score of 9, while "Surprise" and "Amazement" received 0, and "Depression" received the lowest score of -9.

[0083]

[0084] For example, let's say that if a document written by group member A is analyzed by artificial intelligence and three of the 45 emotions listed in the emotion list above are extracted as the main emotions of group member A that appear in the context, such as 'excitement', 'joy', and 'proud' are extracted. Group member A's emotion score is 'excitement' = 7, 'joy' = 7, 'proud' = 3, for a total of 17 points.

[0085] Let's say that by analyzing the document written by group member B using artificial intelligence, three of the main emotions of group member B that appear in the context are extracted from the 45 emotions listed in the emotion list above, such as 'joy', 'excitement', and 'satisfaction'. Group member B's emotion score is 'joy' = 8, 'excitement' = 7, 'satisfaction' = 6, for a total of 21 points.

[0086] Let's say that by analyzing the document written by group member C using artificial intelligence, three of the main emotions of group member C that appear in the context are extracted from the 45 emotions listed in the above emotion list, such as 'annoyance', 'hate', and 'loneliness'. Group member C's emotion score is 'annoyance' = -5, 'hate' = -6, 'loneliness' = -7, for a total of -18 points.

[0087]

[0088] The keyword extraction unit (200) uses artificial intelligence to extract multiple main keywords from documents written by group members based on TF-IDF (Term Frequency - Inverse Document Frequency). The number of main keywords extracted can be changed and configured in the system.

[0089] For example, extract 5 main keywords from a document written by group member A, extract 5 main keywords from a document written by group member B, and extract 5 main keywords from a document written by group member C.

[0090] TF-IDF is a representative method for extracting important words from documents. This method calculates the importance of each word by combining the word frequency (TF) and the inverse document frequency (IDF). The keyword extraction unit (200) uses TF-IDF to extract five key keywords from each document and uses these to determine the similarity between documents. The five keywords here are just examples.

[0091] In addition to the TF-IDF method, you can also use the BERT embedding to extract five main keywords from each document.

[0092]

[0093] The member relationship analysis unit (300) determines (1) whether key words are similar among group members based on documents dated on the same day. For example, it determines (1) whether key words are similar among group members A, B, and C based on diaries written on the same day. The purpose is to determine whether group members share events by determining whether keywords are similar. Simply put, similar keywords indicate that they likely spent time together that day.

[0094] We've discussed how to extract key keywords above. To determine the similarity between the key keywords extracted from each document authored by A, B, and C, each keyword is first vectorized. Vectorization methods such as Word2Vec and BERT embedding are used. A high similarity score indicates that the keywords in the two documents are similar. Generally, a cosine similarity of 0.7 or higher indicates high similarity.

[0095] The member relationship analysis unit (300) determines (2) whether at least one emotion type matches among multiple emotions based on members' documents of the same date and whether the difference in the overall emotion score is below a certain level. Thus, members who satisfy both of these two criteria (1, 2) regarding keywords and emotions are considered to be related. If the difference in the overall emotion score is 5 points or less, the score difference is considered to be below a certain level. Of course, this value can be changed.

[0096] For example, group member A's emotional scores above are 'excited' = 7, 'happy' = 7, 'proud' = 3, a total of 17 points; group member B's emotional scores are 'joy' = 8, 'excited' = 7, 'satisfied' = 6, a total of 21 points; and group member C's emotional scores are 'annoyed' = -5, 'hate' = -6, 'lonely' = -7, a total of -18 points.

[0097] Group members A and B share the same emotion, "excited," among the three, and their overall emotional scores differ by 4 points. In other words, group members A and B share the same emotion, but their overall emotional scores differ by less than a certain level.

[0098] In summary, if the cosine similarity of the main keywords based on the documents of the same date of group members A and B indicates a high similarity, that is, the similarity is higher than the standard value, and at least one emotion type matches and the difference in the overall emotion score is below a certain level, then both of the above (1, 2) are satisfied, so group members A and group members B are judged to be related to each other.

[0099]

[0100] In the above example, group members were considered related if they had similar keywords extracted from documents written by group members on the same date, had at least one emotion type consistent among multiple emotions, and had a difference in their overall sentiment scores below a certain level. In other words, if these two conditions (1, 2) were met, the group members were considered related. However, determining whether group members are related as a group or as close friends cannot be determined based on a single day. Therefore, observation of a recent period is necessary.

[0101] The member relationship analysis unit (300) determines that the group members are related as a group or close friends if the number of days on which the group members satisfy two (1, 2) of keywords and emotions over a recent period exceeds a predetermined level, and determines that the group members are isolated if they are not related to other members and their negative emotions over a recent period exceed a predetermined level.

[0102] For example, if group members A and B satisfy both keywords and emotions (1, 2) more than 8 days in the past 20 days, which is 40% of the time, then group members A and B are considered to be best friends.

[0103] And if group member A, group member B, and group member C satisfy two (1, 2) of the keywords and emotions more than 8 days, which is 40% of the number of days in the last 20 days, group member A, group member B, and group member C are judged to be related to each other.

[0104] If group member C is not connected to other members and has experienced negative emotions exceeding a certain level over the past 20 days, group member C is considered isolated. Figure 3 shows the emotional categories of positive, neutral, and negative. Three main emotions are extracted daily for the past 20 days to calculate a daily emotional score. If negative emotions, corresponding to negative values, exceed 50% (10 days), group member C is considered isolated and unable to socialize with other members.

[0105]

[0106] The specific modules of the member relationship analysis unit (300) that performs the above are as follows.

[0107] The member relationship analysis unit (300) includes an event sharing judgment module (310), an emotion sharing judgment module (320), a member relationship judgment module (330), and an outlier judgment module (340).

[0108] The incident sharing judgment module (310) determines the similarity between multiple keywords extracted from documents of the same date of group members by the keyword extraction unit (200), and if the similarity is above a certain level, it determines that the incident is shared between the members.

[0109] The emotion sharing judgment module (320) determines that emotions are shared between the members if at least one emotion type matches among multiple emotions extracted from documents of the same date by the artificial intelligence emotion extraction module (110) and if the difference in the comprehensive emotion scores of the documents of the same date by the emotion scoring module (120) is below a certain level.

[0110] The member relationship judgment module (330) determines that if three or more members share events and emotions together and exceed a predetermined level over a recent period of time based on the judgment results of the event sharing judgment module (310) and the emotion sharing judgment module (320), the member is judged to be in the same group, and if two members share events and emotions and exceed a predetermined level over a recent period of time, the member is judged to be a close friend. In addition, the member relationship judgment module (330) determines that if the member does not belong to a group or a close friend and the negative emotion exceeds a predetermined level over a recent period of time, the member is judged to be isolated.

[0111] Figure 2 is a visual example of how this method can be used to determine whether a group is a group 1, a group 2, a close friend, or an isolated member. In Figure 2, Jacob is isolated.

[0112] If a group member is isolated, it means that they are in a dangerous situation, and teachers at school or managers at work should immediately begin counseling them and work to improve their situation.

[0113]

[0114] The outlier judgment module (340) analyzes the comprehensive emotional scores of group members by date over a recent period of time, and judges members who have a certain number of dates corresponding to values ​​that deviate from the average or median by a certain level or more as outliers.

[0115] For example, when using the mean and standard deviation, the individual comprehensive emotional scores of group members are analyzed by date for the past 20 days, the average and standard deviation of the comprehensive emotional scores of all group members by date are calculated, and if the number of days corresponding to values ​​that deviate from the average ±standard deviation exceeds 50% (10 days), the member is judged as an outlier.

[0116] As another example, when using the median and interquartile range (IQR), the individual comprehensive emotional scores of group members are analyzed by date for the past 20 days, and the median and IQR values ​​of the comprehensive emotional scores of all group members by date are calculated. If the number of days where the value is less than IQR x 1.5 in the first quartile or greater than IQR x 1.5 in the third quartile exceeds 10 days (50%), the member is judged as an outlier.

[0117] The member relationship judgment module (330) determines that a member who does not belong to a group or a close friend, whose negative emotions have exceeded a certain level over a recent period of time, and whose outlier judgment module (340) determines as an outlier is a member at risk.

[0118]

[0119] The member relationship analysis unit (300) includes a cheering message output module (350).

[0120] The cheering phrase output module (350) analyzes the connection or isolation between group members of each member as determined by the member relationship judgment module (330) and the keywords and emotions that serve as the basis for determining the connection or isolation of each member using artificial intelligence, and generates a cheering phrase reflecting the connection or isolation and the keywords and emotions, and provides it to each member. In the present invention, analyzing using artificial intelligence includes not only using a self-developed artificial intelligence engine, but also using an LLM (Large Language Model) such as ChatGPT.

[0121] When providing a cheering phrase to Richard in 'Group 2' in Figure 2, if Richard and David have similar keywords of 'board game' and 'cube' and share the same emotions of 'fun' and 'passion', the following cheering phrase can be provided.

[0122] [Example of a message of encouragement for Richard]

[0123] "Richard, you and David had a lot of fun playing board games and playing cubes yesterday. How was your day? Board games and cubes are great. You're also working hard at your academy these days. When you get tired of studying, try practicing cubes with David. That'll make school fun!"

[0124]

[0125] When providing a cheering phrase to the 'isolated' Jacob in Figure 2, the following cheering phrase can be provided.

[0126] [Example of a message of encouragement for Jacob]

[0127] "Jacob! Yesterday was really tough, but are you feeling okay today? The other friends don't mind hanging out with you. Tomorrow, try saying hello to your friends first. Then they'll welcome you back!"

[0128]

[0129] So far, we have examined whether group members are judged to be in a group or close friends if their keywords are similar (event sharing), at least one type of emotion matches, and the difference in their overall emotional scores is below a certain level (emotion sharing). In addition, if they do not belong to a group or close friends and their negative emotions have exceeded a certain level over a recent period of time, they are considered isolated.

[0130]

[0131] From here on, we'll discuss centrality analysis. Centrality analysis is a method of social network analysis (SNA) that assesses the importance of a particular group member within a network. Centrality analysis allows us to gain a deeper understanding of the relationships between group members and gain a three-dimensional understanding of the structure of groups, such as bullies, close friends, and cliques.

[0132] The member relationship analysis unit (300) includes a centrality display module (360), an individual median calculation module (370), a lifestyle judgment module (380), and a median change observation module (390).

[0133] The centrality display module (360) determines the similarity between multiple keywords extracted from members' documents of the same date by the keyword extraction unit (200) and connects members with a similarity level above a certain level with a relationship line.

[0134] Centrality analysis connects members with similar keywords, even if their emotions are not identical or similar. In other words, sharing an event connects members with a relationship line, even if their emotions differ.

[0135] The centrality display module (360) displays the relationship line as being connected by a positive or negative emotion if at least one emotion type matches among multiple emotions based on documents of the same date of members connected by a relationship line and the difference in the overall emotion score is below a certain level.

[0136] In addition, the centrality display module (360) indicates that the relationship line is connected to an unrelated emotion if there is no matching emotion type among multiple emotions based on documents of the same date among members connected by a relationship line and the difference in the overall emotion score is above a certain level. Here, an unrelated emotion means including an opposite emotion.

[0137] Referring to Figure 4, it shows the results of a centrality analysis on one day. Thick lines indicate positive emotional connections, thin lines indicate unrelated emotional connections, and dotted lines indicate negative emotional connections.

[0138] In the example in Figure 4, a thick line (positive emotion) was assigned 3 points, a thin line (unrelated emotion) was assigned 2 points, and a dotted line (negative emotion) was assigned 1 point. The median score for each student was shown. Michael scored the highest at 15 points, while Jacob scored the lowest at 2 points.

[0139] Figure 4 analyzes the centrality of a single day, revealing that one person can be connected to multiple others. This provides a different perspective from the group, close friend, and isolation relationships depicted in Figure 2. While Figure 2 only showed Jacob's isolation, Figure 4 reveals who Jacob is negatively emotionally connected to. Using the results of Figures 2 and 4, school teachers and workplace managers can effectively provide psychological counseling and guidance to isolated members.

[0140]

[0141] The centrality display module (360) can display multiple relationship lines between members by additionally displaying relationship lines for each document. That is, Figure 4 displays relationship lines by analyzing only documents written by group members today, but additional relationship lines can be displayed by analyzing documents daily.

[0142] The individual median calculation module (370) calculates the median for each member by synthesizing the scores corresponding to the number and type of relationship lines connected to each member.

[0143] Figure 4 shows the numerical results of today's centrality analysis. The median of each member can be calculated and displayed cumulatively each day. Today's median was highest for Michael at 15 points, while lowest for Jacob at 2 points.

[0144]

[0145] The lifestyle judgment module (380) determines a member as an insider if the median of the members is higher than a certain level, determines a member as an outsider if the median of the members is lower than a certain level, and determines a member as a risk member if there are many lines connected by negative emotions in the relationship line.

[0146] For example, if a member's median daily score is 10 or higher, they are considered insiders, while those below 5 are considered outsiders. If two or more lines of relationship are connected by negative emotions, they are considered at-risk.

[0147] In Figure 4, James, Richard, and Michael are considered insiders with 10, 10, and 15 points, respectively. John and Jacob are considered outsiders with 4 and 2 points, respectively. Furthermore, Jacob is considered a risky member because he has two or more negative emotion-related lines. A member with many negative emotion-related lines is likely to be a victim of bullying or school violence.

[0148]

[0149] The median change observation module (390) determines whether the median of a member is increasing or decreasing through daily observation, and if the median is decreasing, it determines that the member's situation is isolated and notifies the member and the member's manager of the situation.

[0150] The median of a member can be calculated daily and displayed cumulatively. By observing the cumulative median over the past month, you can determine the changes in each member's median. If the median is maintained above a certain level or is increasing, that member is popular. However, if a member's median is decreasing or remaining below a certain level over the past month, this indicates an isolated and dangerous situation, so notify the member and the manager.

[0151]

[0152] The present invention analyzes documents written by members of a group using artificial intelligence to determine whether keywords and emotions are similar among members, thereby confirming whether members share events and emotions with each other, thereby making it easy to determine whether members are connected as a group or close friends, or in an isolated situation. In addition, it applies a method of analyzing daily records rather than intentional psychological surveys, enabling realistic analysis of members' events and emotions.

[0153] In addition, by applying different relationship lines according to the type of emotion between connected members and displaying them visually, and by calculating and displaying the centrality value for each member cumulatively, it is possible to intuitively identify who the popular members are and who the isolated members are, enabling quick and effective consultation.

[0154] Although the present invention has been described in detail with reference to preferred embodiments, it should be understood that the embodiments described above are exemplary in all respects and not restrictive, as those skilled in the art can implement the present invention in other specific forms without changing the technical idea or essential characteristics thereof.

[0155] In addition, the scope of the present invention is defined by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

[0156]

[0157] 100: Emotion Extraction Unit

[0158] 110: Artificial Intelligence Emotion Extraction Module

[0159] 120: Emotion Scoring Module

[0160] 200: Keyword Extraction Unit

[0161] 300: Member Relationship Analysis Department

[0162] 310: Case Sharing Judgment Module

[0163] 320: Emotional Sharing Judgment Module

[0164] 330: Member Relationship Determination Module

[0165] 340: Outlier Determination Module

[0166] 350: Cheering message output module

[0167] 360: Centrality display module

[0168] 370: Individual Median Calculation Module

[0169] 380: Lifestyle Determination Module

[0170] 390: Median Change Observation Module

[0171] 400: Document Database

Claims

1. A sentiment extraction unit that analyzes the context of documents written by group members and extracts a set number of main sentiments of the document writers; A keyword extraction unit that analyzes the context of a document and extracts a set number of main keywords; and A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and identify the relationships between group members, including a member relationship analysis unit that uses multiple extracted major emotions and major keywords to identify relationships between group members and analyzes and provides the connection relationship and isolation between group members.

2. In claim 1, The above emotion extraction unit, An artificial intelligence emotion extraction module that analyzes multiple emotions listed in the emotion list using artificial intelligence and extracts the main emotion of the document writer that appears in the context of the document from among the multiple emotions listed in the emotion list; and A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of document writers and to identify relationships between group members, characterized by including an emotional scoring module that comprehensively calculates the emotional score of document writers based on the emotional scores for each emotion extracted by the artificial intelligence emotional extraction module.

3. In claim 2, The above emotional scoring module is, A document analysis system that uses artificial intelligence to analyze documents, which divides each emotion extracted by the artificial intelligence emotion extraction module into positive, neutral, and negative, and calculates the emotional score of the document writer based on the emotional classification score table that assigns a score to each emotion, thereby identifying the emotional state and keywords of the document writer and identifying the relationships between group members.

4. In claim 1, The above member relationship analysis department, A document analysis system that uses artificial intelligence to analyze documents, determines whether the main keywords are similar among members based on documents of the same date of the members (1), determines whether at least one type of emotion matches among multiple emotions based on documents of the same date of the members and whether the difference in the overall emotion score is below a certain level (2), and determines that members who satisfy both of these (1, 2) for keywords and emotions are related to each other, and identifies the emotional state and keywords of the document writer and the relationship between group members.

5. In claim 4, The above member relationship analysis department, A document analysis system that uses artificial intelligence to analyze documents to identify the emotional state and keywords of the document writer and to identify relationships between group members, characterized in that if the number of days on which members satisfy two (1, 2) of keywords and emotions exceeds a certain level in the recent period, the members are judged to be related as a group or close friends, and if the members are not related to each other and the negative emotions exceed a certain level in the recent period, the members are judged to be isolated.

6. In claim 2, The above member relationship analysis department, An event sharing judgment module that determines that an event is shared between the members when the keyword extraction unit determines the similarity between multiple keywords extracted from documents of the same date of the group members and the similarity is above a certain level; An emotion sharing judgment module that determines that emotions are shared between members if at least one emotion type matches among multiple emotions extracted from documents of the same date by the artificial intelligence emotion extraction module and the difference in the comprehensive emotion scores of documents of the same date of the members calculated by the emotion scoring module is below a certain level; and A document analysis system that analyzes a document using artificial intelligence to identify the emotional state and keywords of the document writer and to identify the relationships between group members, characterized by including a member relationship determination module that determines that if three or more members share events and emotions together and exceeds a predetermined level over a recent period of time as a result of the determination of the event sharing determination module and the emotion sharing determination module, the members are determined to be in the same group, if two members share events and emotions and exceeds a predetermined level over a recent period of time, the members are determined to be close friends, and if the members do not belong to a group or close friends and their negative emotions exceed a predetermined level over a recent period of time, the members are determined to be isolated.

7. In claim 6, The above member relationship analysis department, It further includes an outlier judgment module that analyzes the comprehensive emotional scores of members by date for a recent period of time and judges members who have a certain number of dates corresponding to values ​​that deviate from the average or median by a certain level as outliers. The above member relationship judgment module determines that a member who does not belong to a group or a close friend, has had negative emotions exceeding a certain level for a recent period of time, and is judged as an outlier by the outlier judgment module is a document analysis system that uses artificial intelligence to analyze documents to identify the emotional state and keywords of the document writer and to identify the relationships between group members.

8. In claim 6, The above member relationship analysis department, A document analysis system that analyzes documents using artificial intelligence to determine the emotional state and keywords of the document writer and to determine the relationships between group members, characterized by further including a support message output module that analyzes, using artificial intelligence, the keywords and emotions that serve as the basis for determining the connection or isolation of each member among group members as determined by the member relationship judgment module, and generates support messages that reflect the connection or isolation and the keywords and emotions and provides them to each member.

9. In claim 2, The above member relationship analysis department, The keyword extraction unit determines the similarity between multiple keywords extracted from members' documents of the same date, and connects members with a similarity level above a certain level with a relationship line. Based on the documents of the members of the same date connected by the relationship line, if at least one emotion type matches among multiple emotions and the difference in the overall emotion score is below a certain level, the relationship line is displayed in a way that indicates that it is connected with a positive or negative emotion. A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of document writers and to identify relationships between group members, characterized in that it includes a centrality display module that indicates that the relationship line is connected by an unrelated emotion when there is no matching emotion type among multiple emotions based on documents of the same date of members connected by a relationship line and the difference in the comprehensive emotion score is above a certain level.

10. In claim 9, The above central display module is, A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and to identify relationships between group members, characterized by additionally displaying the above relationship lines for each document and allowing multiple relationship lines to be displayed between members.

11. In claim 9, The above member relationship analysis department, A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and to identify relationships between group members, further comprising an individual median calculation module that calculates a median for each member by synthesizing scores corresponding to the number and type of relationship lines connected to each member.

12. In claim 11, The above member relationship analysis department, A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of the document writer and to identify relationships between group members, characterized by further including a lifestyle judgment module that determines a member as an insider if the median of the members is higher than a certain level, determines a member as an outsider if the median of the members is lower than a certain level, and determines a member as a risk member if there are many lines of negative emotions connected to the relationship line.

13. In claim 11, The above member relationship analysis department, A document analysis system that analyzes documents using artificial intelligence to identify the emotional state and keywords of document writers and to identify relationships between group members, characterized by further including a median change observation module that determines whether the median of a member is increasing or decreasing through daily observation, and if the median is decreasing, determines that the situation of the member in question is becoming isolated and notifies the member in question and the member's manager of the situation.

Images