Emotion analysis-based recommendation system

Abstract

The present invention provides a book recommendation system based on personal emotion analysis, comprising: a database storing self-expression indicators representing personal emotions; an interface server that receives the self-expression indicators from the personal via a network, stores them in the database, generates a recommended book code, and transmits it to the personal terminal; and a personal terminal that receives the self-expression indicators, transmits them to the interface server via the network, and receives the recommended book code. The interface server comprises: an emotion analysis module that analyzes the self-expression indicators to calculate an emotional state; a keyword generation module that generates keywords corresponding to the emotional state; a book content analysis module that analyzes publicly available book information; a mapping module that calculates a correlation from keywords corresponding to the emotional state and the analysis results of the book content analysis module; and an output module that displays a recommended book code based on the correlation calculated by the mapping module. The invention also provides a computer program and a computer-readable recording medium for executing the same.

Description

Technology Field

[0001] The present invention relates to a book recommendation system, and more specifically, to a sentiment analysis-based book recommendation system. Background Technology

[0002] With the recent advancement of information and communication technology, emotion-based recommendation technology that analyzes an individual's tendencies and mood to provide personalized content is being actively researched. In particular, content-based recommendation or collaborative filtering technologies, which analyze users' search history or purchasing patterns to recommend books of similar genres, are widely used in online bookstores, e-book platforms, and audiobook services.

[0003] However, these conventional recommendation methods have limitations in that they fail to reflect the user's current emotional state or psychological changes. For instance, even if a young student user usually reads books for entertainment, they may prefer comforting essays when feeling depressed; yet, existing recommendation systems fail to account for such changes in emotional state and perform recommendations primarily based on past behavioral data.

[0004] Under these circumstances, user emotion recognition technology has evolved into various forms. For instance, technologies are being proposed that include facial expression recognition via cameras, Natural Language Processing (NLP)-based emotion classification analyzing the linguistic features of text or speech, and techniques that estimate emotional states by interpreting non-verbal expressions such as colors, emoticons, and avatar facial expressions.

[0005] However, conventional sentiment analysis technologies were mostly limited to making temporary emotional judgments or relied on static information such as user preferences, search history, and age group. These methods, however, failed to reflect the user's current psychological and emotional state, resulting in limitations in personalized recommendations. In particular, in areas where the consistency between linguistic context and emotional meaning is critical, such as book recommendations, there was a problem in quantitatively calculating the correlation between emotions and content meanings. Furthermore, existing emotion-based systems merely dichotomized users' emotional states into positive or negative categories or relied on emotion dictionaries, failing to classify detailed nuances or complex emotions in detail. Additionally, emotion-based recommendation services in public environments via kiosks or mobile devices are still underdeveloped.

[0006] [Prior Art]

[0007] - Korean Published Patent No. 2023-0128596 (Published September 5, 2023)

[0008] - Korean Published Patent No. 2023-0128878 (Published September 5, 2023)

[0009] - Korean Registered Patent No. 2232957 (Registered on March 22, 2021) The problem to be solved

[0010] The present invention aims to provide a system capable of more sophisticated and personalized book recommendations by classifying a user's emotional state in detail through an artificial intelligence model and calculating the degree of association by mapping keywords corresponding to the emotion with the linguistic characteristics of book content in vector units, as well as a computer program and a computer-readable recording medium for executing the same. means of solving the problem

[0011] To solve the above problem, the present invention provides a book recommendation system based on personal emotion analysis, comprising: a database storing a self-expression indicator representing a personal emotion; an interface server that receives the self-expression indicator from the personal via a network, stores it in the database, generates a recommended book code, and transmits it to the personal terminal; and a personal terminal that receives the self-expression indicator, transmits it to the interface server via the network, and receives the recommended book code. The interface server comprises: an emotion analysis module that analyzes the self-expression indicator to calculate an emotional state; a keyword generation module that generates a keyword corresponding to the emotional state; a book content analysis module that analyzes publicly available book information; a mapping module that calculates a correlation from the keyword corresponding to the emotional state and the analysis result of the book content analysis module; and an output module that displays a recommended book code based on the correlation calculated by the mapping module.

[0012] In addition, the self-expression indicator includes an avatar expression that changes according to the state in which one of two or more icon shapes preset in the personal terminal is input in correspondence with the degree of the individual's psychology or mood, a color that changes according to the state in which one of two or more colors preset in the personal terminal is input, and a score input among scores preset in the personal terminal. This provides an emotion analysis-based book recommendation system.

[0013] In addition, the above-mentioned sentiment analysis module provides a sentiment analysis-based book recommendation system characterized by classifying at least one sentiment state among a plurality of sentiment classes generated through an artificial intelligence model.

[0014] In addition, the keyword generation module extracts basic keywords corresponding to the classified emotional states by referring to an emotion dictionary, and expands the basic keywords into one or more detailed keywords using a language model, thereby providing an emotion analysis-based book recommendation system.

[0015] In addition, the book content analysis module provides a sentiment analysis-based book recommendation system characterized by separating the disclosed book information into sentence units and analyzing morphemes, converting each sentence into an embedding vector using a Natural Language Processing (NLP) model, and generating a book feature vector by integrating the embedding vectors and storing it in the database.

[0016] In addition, the above mapping module provides a sentiment analysis-based book recommendation system characterized by converting keywords corresponding to the sentiment state into sentiment keyword vectors using a Natural Language Processing (NLP) model and calculating the degree of association between the sentiment keyword vectors and the book feature vectors using an association analysis model.

[0017] Additionally, the interface server further includes a self-expression indicator statistics generation module, and the sentiment analysis module predicts future self-expression indicators from the generated self-expression indicator statistics using a time series model or an artificial intelligence model, and analyzes the future self-expression indicators to calculate an emotional state, thereby providing a sentiment analysis-based book recommendation system.

[0018] To solve the above additional problem, the present invention provides a computer program stored in a medium for executing the system.

[0019] To solve the above additional problem, the present invention provides a computer-readable recording medium that records a computer program for executing the system. Effects of the invention

[0020] According to the present invention, by analyzing various forms of self-expression indicators such as a user's facial expression, color selection, and emotion score, an emotion analysis-based book recommendation system capable of precise emotion-based personalized recommendations can be provided.

[0021] In addition, by calculating the degree of association between emotion keyword vectors and book feature vectors using an association analysis model, it is possible to provide an emotion analysis-based book recommendation system that implements matching at the level of emotion semantics.

[0022] In addition, by utilizing language models to expand detailed concepts of emotion and generate refined keywords, it is possible to provide an emotion analysis-based book recommendation system that richly reflects emotional context.

[0023] In addition, through time-series analysis of user emotion data, it is possible to provide an emotion analysis-based book recommendation system that predicts future emotional changes and proactively recommends books that correspond to them. Brief explanation of the drawing

[0024] FIG. 1 is a simplified diagram showing the configuration of an emotion analysis-based book recommendation system according to one embodiment of the present invention. FIGS. 2 to 4 are drawings showing a self-representation indicator input screen displayed on a user terminal in one embodiment of the present invention. FIG. 5 is a diagram showing a screen displaying the result of inputting a self-expression index in one embodiment of the present invention. FIG. 6 is a diagram showing a self-expression indicator checkerboard displayed on an administrator terminal in one embodiment of the present invention. FIG. 7 is a diagram showing a screen displaying a recommended book code (QR code) output by an output module in one embodiment of the present invention. FIG. 8 is a diagram showing a screen in which statistical indicators are displayed on a user terminal in one embodiment of the present invention. FIG. 9 is a diagram showing a screen displaying a list of recommended books by predicting future emotional changes in one embodiment of the present invention. Specific details for implementing the invention

[0025] Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In describing the present invention, detailed descriptions of related prior art are omitted if it is determined that such detailed descriptions may obscure the essence of the invention. In the drawings, parts unrelated to the description have been omitted to clearly explain the present invention, and similar parts throughout the specification have been assigned similar reference numerals. Furthermore, throughout the specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0026] FIG. 1 is a simplified diagram showing the configuration of an emotion analysis-based book recommendation system according to one embodiment of the present invention.

[0027] As illustrated in FIG. 1, the sentiment analysis-based book recommendation system (100) according to the present invention basically includes an interface server (110), a database (120), and a user terminal (130). Below, each component will be examined in more detail.

[0028] The interface server (110) is connected to the network (150) and configured so that a user can access it from a remote location via the network (150). Thus, the user can input a self-expression indicator through the network (150) using a terminal (130) on which a website provided by the interface server (110) is displayed, as exemplified in FIGS. 2 to 4, and the interface server (110) stores the self-expression indicator received from the user terminal (130) in the database (120), and the self-expression indicator can be stored in the database (120) as an item of the self-expression indicator DB (121).

[0029] In the present invention, the self-expression indicator is an indicator representing the degree of a user's psychology or mood and can be displayed in various ways. However, in the present invention, particularly to induce interest and frequent accessibility to electronic devices for young students, as illustrated in FIG. 2, the expression of an avatar (132) can be displayed to change as an icon shape (131) having various types of expressions pre-set corresponding to the degree of psychology or mood is selected (touched) (Fig. 2a) on the screen of a user terminal (130), as illustrated in FIG. 3. Alternatively, as illustrated in FIG. 3, a color that can express the current psychology or mood as a color can be selected (touched) (Fig. 3a) from a color wheel (133) displaying various pre-set colors on the screen of the user terminal (130), so that, for example, the selected color appears as the background color of the avatar (132), as illustrated in FIG. 4. Furthermore, as illustrated in FIG. 4, a score that can express the current psychology or mood can be displayed on a pre-set scoreboard or score gauge (134) on the screen of the user terminal (130). It can also be displayed by selecting (touching). Figure 5 shows a screen displaying the result in which a sad-faced avatar, a blue color, and a score of 3 points are entered as self-expression indicators.

[0030] In this case, as a preferred example, an icon shape (131) having various preset types of expressions may be displayed around the expression area of ​​an avatar (132) displayed on a user terminal (130), so that the expression of the avatar (132) changes according to the user's selection (touch) of a specific icon shape (131). Additionally, in a similar manner, if a color wheel (133) with various colors arranged around the expression area of ​​the avatar (132) is displayed, a scoreboard or score gauge (134) may be displayed in the next step according to the user's selection (touch) of a specific color.

[0031] In the present invention, the network (150) may generally refer to the Internet network, but other network networks, such as LAN (Local Area Network), WAN (Wide Area Network), PSTN (Public Switched Telephone Network), PSDN (Public Switched Data Network), cable TV network, wireless communication network, etc., are also possible. The user can access the interface server (110) through a user terminal (130), such as a computer or mobile device (smartphone or tablet PC, etc.), and input and transmit self-representation indicators (using a save button on the terminal screen, etc.) to store them in the database (120).

[0032] Additionally, the interface provided by the interface server (110) to the user for receiving self-expression indicators may generally be a website or application through a web browser installed on a computer or mobile device, but is not limited thereto and is not restricted in form as long as it can receive self-expression indicators from the user and transmit them to the interface server (110) through the network (150).

[0033] The interface server (110), which receives and inputs self-expression indicators from the user terminal (130), can update the database (120) periodically or in real time, and through this process, it can store the self-expression indicators in the database (120) or modify them according to external requests, etc.

[0034] Referring to FIG. 1, the present invention may further include an administrator terminal (140). The administrator terminal requests and receives self-expression indicators from an interface server (110) via a network (150) to display or output the self-expression indicators. In an environment where users form groups, such as students within a school, the self-expression indicators transmitted by each user through the interface server (110) are received from the self-expression indicator DB (121), thereby enabling the viewing of self-expression indicators for all users belonging to the group. For example, when the administrator terminal (140) requests self-expression indicators from the interface server (110), as shown in FIG. 6, the content of the self-expression indicators for the user who transmitted the self-expression indicators can be checked on the self-expression indicator checkerboard (141), which is pre-displayed on the screen of the administrator terminal (140) by dividing it into user names, etc. The administrator terminal (140), like the user terminal (130), can access the interface server (110) via a computer or mobile device (such as a smartphone or tablet PC) and view the self-representation indicators by receiving them (using automatic updates or request buttons on the terminal screen).

[0035] From the screen of the administrator terminal (140) that receives the self-expression indicator, an administrator, such as a homeroom teacher managing students in a specific class at a school, can check the individual student's psychology or mood at a glance from the avatar (132) expression. In addition, the student's psychology or mood can be indirectly checked from the color of the avatar.

[0036] In the present invention, the database (120) is connected to the interface server (110) and stores the self-expression indicator received from the user terminal (130) through the interface server (110) in the self-expression indicator DB (121). Thus, when a request for a self-expression indicator is received from the administrator terminal (140), the database (120) transmits the self-expression indicator to the administrator terminal (140) as a response to the request.

[0037] Prior to the input of the user's self-expression indicator and the verification of the administrator's self-expression indicator, means for registration and authentication for access to the user's interface environment are considered. To this end, in the present invention, a user registration and authentication module (112) is provided in the interface server (110) to receive personal information of each user from the user terminal (130), and the user terminal (130) transmits the received information through the network (150). The user registration and authentication module (112) of the interface server (110) can receive the user's personal information through the network (150) and store it in the user information DB (122) of the database (120). Subsequently, the user registration and authentication module (112) requests authentication processing from the user information DB (122) to the interface server (110), and the interface server (110) transmits the authentication status to the user registration and authentication module (112) to allow access to the user's interface environment.

[0038] In the present invention, the interface server (110) includes an emotion analysis module (111), a keyword generation module (112), a book content analysis module (113), a mapping module (114), and an output module (115), and may further include a self-expression indicator statistics generation module (116).

[0039] The above emotion analysis module (111) is a module that analyzes the self-expression indicator to produce an emotional state, and the analysis of the input self-expression indicator can be performed through an artificial intelligence model such as a deep learning model (LSTM (Long Short-Term Memory)), CNN (Convolutional Neural Network)), Transformer (BERT, GPT, etc.), or machine learning model, and if multiple emotion classes are produced through the artificial intelligence model, one or more of them can be classified as an emotional state.

[0040] For example, if a user inputs an avatar with a sad (crying) expression, a blue color, and a score of 1 point as a self-expression indicator, the AI ​​model can output as follows, and among them, “Sadness” with the highest probability (0.84) is selected and can be classified as the current emotional state.

[0041] {"Joy": 0.02, "Sadness": 0.84, "Anger": 0.07, "Fear": 0.03, "Disgust": 0.01, "Neutral": 0.03}

[0042] The keyword generation module (112) is a module that generates keywords corresponding to the emotional state classified by the emotion analysis module (111), and can extract basic keywords corresponding to the classified emotional state by referring to an emotion dictionary, and expand the basic keywords into one or more detailed keywords using a language model (such as BERT, KoBERT, GPT family lightweight models, etc.).

[0043] For example, the above-mentioned emotion dictionary serves to set basic emotion keywords, and if the current emotional state is classified as “Sadness” through the emotion analysis module (111) as described above, the emotion dictionary can extract “Sadness” as the basic emotion keyword, which is determined to be the most suitable from a category containing similar emotion keywords such as “depression,” “anger,” and “frustration.”

[0044] In addition, the above language model is a model that further concretizes emotions by understanding the context of language. When a basic emotion keyword such as "Sadness" is extracted from the emotion dictionary, the language model can differentiate emotions more finely by understanding the context, for example, by referring to the input self-expression indicators and expanding the keyword "Sadness" into detailed emotion keywords such as "consolation," "sympathy," and "depression."

[0045] The above book content analysis module (113) is a module that analyzes publicly available book information, mainly book information published on the internet, and plays the role of generating book feature vectors by embedding the text, introduction, metadata (structural information describing the book) of the book, etc., using Natural Language Processing (NLP) techniques. Specifically, it may include a process of separating the publicly available book information into sentence units and analyzing morphemes, converting each sentence into an embedding vector using a natural language processing model, and integrating the embedding vectors to generate book feature vectors and storing them in the database (120).

[0046] Models such as TF-IDF, Word2Vec, BERT, Sentence-BERT (SBERT), and KoBERT may be used as the NLP techniques mentioned above. Here, TF-IDF (Term Frequency-Inverse Document Frequency) is a statistical-based vectorization technique that calculates the importance of each word using the word's "in-document frequency (TF)" and "in-document rarity (IDF)," and represents the document as a "word importance vector." Additionally, Word2Vec is a neural network-based word embedding model that represents words as dense vectors of fixed length, and is a model trained so that words with similar meanings are located close together in the vector space. Furthermore, BERT (Bidirectional Encoder Representations from Transformers) is a Transformer-based pre-trained language model that reads entire sentences bidirectionally to reflect context, and is a model trained using Masked Language Modeling, Next Sentence Prediction, etc. In addition, the above-mentioned Sentence-BERT (SBERT) is a sentence embedding extension model of BERT, which is a model trained to represent the meaning of an entire sentence as a vector by modifying BERT into a Siamese structure (a neural network structure used to learn similarity or relationships between two inputs). Furthermore, KoBERT is a Korean-specific BERT model, a Korean corpus-based BERT model released by SKT, which has the same BERT structure but is specialized for understanding Korean context by being trained on Korean data.

[0047] The above book content analysis model (113) first separates the disclosed book information into sentence units and then analyzes the morphemes. The morphological analysis can be performed by, for example, by using KoNLPy (Korean Natural Language Processing in Python) such as Okt (Open Korean Text), KOmoran (Korean Morphological Analyzer), and Mecab (Yet Another Part-of-Speech and Morphological Analyzer) to tag parts of speech and extract keywords centered on nouns and adjectives. Afterward, preprocessing steps such as stop word removal, lowercase conversion, number / symbol normalization, and removal of duplicate spaces can be performed.

[0048] Next, the analyzed morphemes are converted into high-dimensional embedding vectors using a selected model among the above NLP models, and the embedding vectors can be optimized for cosine similarity calculation by performing L2 normalization. The embedding vectors can quantify the meaning of words, sentences, documents, etc., and represent them in a way that a computer can read, for example, as shown below; and if words are placed in a vector space as shown below, semantic similarity can be calculated by calculating the distance between the two vectors.

[0049] “Depressed” : [0.23, -0.55, 1.02, … ]

[0050] “Lonely” : [0.20, -0.53, 1.01, … ]

[0051] Next, the above embedding vectors are integrated to generate book feature vectors and stored in the database (120). The integration of embedding vectors can be performed by integrating the embedding vectors for sentences using methods such as the average, weighted sum, or IDF weighted average, and a book feature vector can be generated through such integration of embedding vectors. For example, text-based embedding vectors can be calculated by cutting the main text and introduction into paragraph units as follows;

[0052] book_text = intro + summary

[0053] book_vec = model.encode(book_text)

[0054] Metadata (genre, keywords, author, etc.) can be converted into text and embedded together or integrated using a weighted average to generate a weighted combined book feature vector (combined_vec) containing metadata as shown below.

[0055] meta_text = "Genre: Novel, Keywords: Comfort, Growth, Family Love"

[0056] combined_vec = 0.8*model.encode(book_text) + 0.2*model.encode(meta_text)

[0057] The generated book feature vector is stored in the database (120) and can be indexed through the construction of vector indexes such as FAISS (Facebook AI Similarity Search), Annoy (Approximate Nearest Neighbors Oh Yeah), and HNSW (Hierarchical Navigable Small World) for a large-scale search environment, and, for example, the quality of the similarity search can be checked using a sample sentiment keyword vector.

[0058] The mapping module (114) is a module that calculates the correlation from keywords corresponding to the emotional state and the analysis results of the book content analysis module (113). It can convert the keywords corresponding to the emotional state into emotional keyword vectors and calculate the correlation between the converted emotional keyword vectors and the book feature vectors using a statistical method, a rule-based method, a machine learning-based method, or a deep learning-based method.

[0059] The method of converting keywords corresponding to the above emotional states into emotional keyword vectors can be performed using an NLP model. For example, if the keywords corresponding to the emotional states are ["sadness", "comfort", "emptiness"], they can be converted into emotional keyword vectors as follows using NLP models such as Word2Vec, FastText, GloVe, KoBERT, Sentence-BERT, and GPT embedding.

[0060] “Sadness” : [0.21, -0.32, 0.58, 0.77, … ]

[0061] “Comfort” : [0.23, -0.29, 0.60, 0.74, … ]

[0062] “Void” : [0.05, -0.80, 0.33, 0.90, … ]

[0063] When the keywords corresponding to the above emotional state are converted into emotional keyword vectors, the relevance score between the emotional keyword vectors and the above book feature vectors is calculated using an association analysis model.

[0064] Among the techniques used as association analysis models for calculating the above association, statistical techniques such as TF-IDF (Term Frequency-Inverse Document Frequency) and BM25 (Best Matching 25) can be applied, and vector similarity-based methods such as Cosine Similarity and Sentence Embedding Similarity can be used. In addition, as a method using neural network techniques, a deep learning matching network or a transformer-based matching model can be used to learn the non-linear relationship between sentiment keyword vectors and book feature vectors and to calculate contextual and semantic associations.

[0065] The mapping module (114) can calculate a final ranking score by integrating a plurality of similarity indicators calculated using a weighted combination method, and, for example, can be expressed as in Equation 1 below.

[0066] [Mathematical Formula 1]

[0067] Score = α·cosine(emotion_vec, book_vec) + β·BM25(emotion_keywords, book_text) + γ·context_fit(user_profile, book_meta)

[0068] In Equation 1, cosine(emotion_vec, book_vec) represents the cosine similarity between the emotion keyword vector and the book feature vector, BM25(emotion_keywords, book_text) represents the BM25-based score between the emotion keywords and the book text, context_fit(user_profile, book_meta) represents the fit between the user profile and the book metadata, and α, β, and γ are weight parameters that can be adjusted by experimental learning.

[0069] In addition, the sentiment analysis-based book recommendation system according to the present invention can improve personalization suitability by reflecting various context information, such as the user's time, gender, weather, face image, free text, and chatbot log, as well as avatar facial expressions, colors, and scores. These personalization factors can be reflected as context_fit, which is the third term of Equation 1 above, or as a separate correction parameter.

[0070] In addition, the mapping module (114) can gradually improve ranking accuracy by applying a machine learning-based ranking technique called Learning To Rank (LTR) to learn the actual user's selection pattern from multiple pairs of sentiment keyword vectors and book feature vectors.

[0071] The output module (115) is a module that displays recommended book codes based on the correlation calculated by the mapping module (114). For example, the result of calculating the correlation may be sorted for the top N books and output in the form of a recommendation list. This list may include the book title, author, cover image, emotion keyword tag, emotion suitability score, etc., to intuitively display the book most suitable for the user's emotional state. It may be generated as various types of code, such as a QR code (see FIG. 7) containing this information, and transmitted to a user terminal (130).

[0072] Additionally, when a user scans a QR code issued to a terminal (130) with a kiosk camera, for example, the kiosk linkage system can decode the information contained in the QR code to restore the user session and output recommended books corresponding to the emotional state, and the kiosk terminal can output recommended book information through a printing device or an audio device, or recite the contents of the book or play a summary as voice when earphones or a headset are connected.

[0073] Through the structure of linking terminals (130) such as smartphones and kiosks, the emotion analysis-based book recommendation system according to the present invention can be immediately utilized in various public environments such as educational environments, public libraries, counseling centers, and schools.

[0074] Meanwhile, the interface server (110) further includes a self-expression indicator statistics generation module (116), and the emotion analysis module (111) can predict future self-expression indicators from the generated self-expression indicator statistics using a time series model or an artificial intelligence model, and analyze the future self-expression indicators to calculate an emotional state.

[0075] The above self-expression indicator statistics generation module (116) can generate statistics on the changes in all self-expression indicators input and transmitted from the user terminal (130) and store them in the self-expression indicator statistics DB (123) of the database (120).

[0076] The self-expression indicator statistics generation module (116) can generate various statistical indicators, such as an emotion change graph that can display the history of changes in avatar expressions, the history of changes in scores, and the history of changes in colors over a certain period, for example, as illustrated in FIG. 8. FIG. 8 illustrates the history of changes over 7 days.

[0077] Here, referring further to Figure 8, the emotional changes from today (Friday the 24th) to tomorrow (Saturday the 25th) and the day after tomorrow (Sunday the 26th) are predicted and shown as self-expression indicators.

[0078] That is, the emotion analysis module (111) can predict the future emotional state by analyzing the user's past emotional input data in a time series. The prediction of the emotional state utilizes various forms of emotion-related data, such as the history of changes in avatar facial expressions, scores, and colors entered in the past, as training data. At this time, the emotion data is organized in a time series format for each user and stored in a database (120), and the model can predict the future emotional state based on this time series data.

[0079] In the case of time series model-based forecasting, deep learning-based time series models such as LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and Transformer-based sequence models can be applied during the time series analysis phase. By learning patterns of emotional change over time, these models can predict the emotional state of the next time point (e.g., tomorrow's or the day after tomorrow's session) from the emotional input values ​​of previous time points. For instance, the model [takes] an input time series vector (X t = [emoji t , color t , score t Emotional score at the next point in time (Y t+1 It can produce ) and emotion classes (sadness, joy, anxiety, etc.).

[0080] In addition to time series models, traditional machine learning techniques such as Random Forest and Gradient Boosting can also be used to classify emotion classes or predict emotion scores using regression.

[0081] A specific example of an operation for predicting a future emotional state in the above-mentioned emotion analysis module (111) is shown as follows:

[0082] On Monday, a student user entered a sad-faced avatar (emoji), the color blue, and an emotion score of 3 points via smartphone.

[0083] The sentiment analysis-based book recommendation system analyzes sentiment patterns (avatar facial expressions, colors, scores, etc.) over the past week and merges an LSTM time series model and a Random Forest model to predict the sentiment state at the next time point (Tuesday).

[0084] Prediction result: Probability of the next day's emotion class "Depressed" 0.82, primary emotion color "Blue", emotion score 2 points or less.

[0085] Accordingly, the emotion analysis-based book recommendation system (100) recommends books corresponding to emotion keywords “comfort,” “empathy,” and “recovery” in a high ranking (see FIG. 9), and the selected books can be played as audio through a recitation function or printed immediately through a kiosk printing function.

[0086] A method for supporting the management of group members according to one embodiment of the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the computer-readable recording medium may be those specifically designed and configured for the present invention, or those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

[0087] Preferred embodiments of the present invention have been described in detail above with reference to the drawings. The description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without changing the technical concept or essential features of the present invention.

[0088] Accordingly, the scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning, scope, and equivalent concepts of the claims should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0089] 100: Sentiment Analysis-Based Book Recommendation System 110: Interface Server 111: Sentiment Analysis Module 112: Keyword Generation Module 113: Book Content Analysis Module 114: Mapping Module 115: Output Module 116 : Self-representation indicator statistics generation module 120 : Database 121 : Self-expression indicator DB 122 : User information DB 123 : Self-expression indicator statistics DB 130 : User terminal 131 : Icon Shape 132 : Avatar 133: Color Wheel 134: Score Gauge 140: Administrator Terminal 141: Self-Representation Indicator Checklist 150: Network

Claims

Claim 1 A sentiment analysis-based book recommendation system comprising: a database storing self-expression indicators representing the emotions of a user; an interface server that receives the self-expression indicators from the user via a network, stores them in the database, generates a recommended book code, and transmits it to the user's terminal; and a user terminal that receives the self-expression indicators, transmits them to the interface server via the network, and receives the recommended book code; wherein the interface server comprises: a sentiment analysis module that analyzes the self-expression indicators to calculate an emotional state; a keyword generation module that generates keywords corresponding to the emotional state; a book content analysis module that analyzes publicly available book information; and a mapping module that calculates a correlation from keywords corresponding to the emotional state and the analysis results of the book content analysis module. A sentiment analysis-based book recommendation system comprising: an output module that displays a recommended book code based on the correlation calculated by the mapping module; wherein the sentiment analysis module classifies at least one sentiment state among a plurality of sentiment classes calculated through an artificial intelligence model; the keyword generation module extracts a basic keyword corresponding to the classified sentiment state by referring to a sentiment dictionary and expands the basic keyword into one or more detailed keywords using a language model; the book content analysis module separates the disclosed book information into sentence units and analyzes morphemes, converts each sentence into an embedding vector using a Natural Language Processing (NLP) model, integrates the embedding vectors to generate a book feature vector and stores it in the database; and the mapping module converts a keyword corresponding to the sentiment state into a sentiment keyword vector using a Natural Language Processing (NLP) model and calculates the correlation between the sentiment keyword vector and the book feature vector using a correlation analysis model.An emotion analysis-based book recommendation system characterized by the above mapping module integrating multiple calculated similarity indicators using a weighted combination method to calculate a final ranking score according to the following Equation 1: [Equation 1] Score = α·cosine(emotion_vec, book_vec) + β·BM25(emotion_keywords, book_text) + γ·context_fit(user_profile, book_meta) In Equation 1, cosine(emotion_vec, book_vec) represents the cosine similarity between the emotion keyword vector and the book feature vector, BM25(emotion_keywords, book_text) represents the BM25-based score between emotion keywords and book text, context_fit(user_profile, book_meta) represents the degree of fit with the user profile and book metadata, and α, β, and γ are weight parameters to be adjusted by experimental learning Possible. Claim 2 An emotion analysis-based book recommendation system according to claim 1, wherein the self-expression indicator includes an avatar expression that changes according to the state in which one of two or more icon shapes preset in the user terminal is input in correspondence with the degree of the user's psychology or mood, a color that changes according to the state in which one of two or more colors preset in the user terminal is input, and a score input among scores preset in the user terminal. Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 delete Claim 7 An emotion analysis-based book recommendation system according to claim 1, wherein the interface server further includes a self-expression indicator statistics generation module, and the emotion analysis module predicts future self-expression indicators from the generated self-expression indicator statistics using a time series model or an artificial intelligence model, and analyzes the future self-expression indicators to calculate an emotional state. Claim 8 A computer program stored on a medium for executing a system according to any one of paragraphs 1, 2 and 7. Claim 9 A computer-readable recording medium having a computer program for executing a system according to any one of paragraphs 1, 2 and 7.

Images