Computing device for providing depression adaptive conversation, method for providing depression adaptive conversation in computing device

Abstract

The present invention relates to a technology for providing an adaptive conversation for depressive disorders, comprising: an acquisition unit for acquiring a user's utterance; a storage unit including a command for returning a response to the utterance using a pre-trained artificial neural network and calculating the user's depression intensity score for each conversation turn of the utterance and the response; and a processing unit for processing to return a response to the utterance by executing the command and to calculate the user's depression intensity for each conversation turn and output the depression intensity score, wherein the artificial neural network may include a pre-trained artificial neural network that quantitatively analyzes the text of the utterance using a KoELECTRA (Korean efficiently learning an encoder that classifies token replacements accurately) model and calculates the depression intensity score based on the DSM-5 (diagnostic and statistical manual of mental disorders, fifth edition) criteria for the analysis.

Description

Technology Field

[0001] The present invention relates to a depression adaptive conversational system, a computing device for providing depression adaptive conversations, and an artificial neural network, and to a method for providing conversations using the same.

[0002] This invention is proposed as a result of joint industry-academia research conducted as part of an industry-academia collaboration R&ED project for a commercial service of a training data generation model and a RAG model utilizing generative AI technology. Background Technology

[0003] Most conversational systems, such as chatbots, have limitations due to the absence of standardized diagnostic criteria, that is, the inability to qualitatively evaluate a user's condition or quantitatively evaluate it based on clinical data.

[0004] In addition, most emotional support systems are developed primarily in English, so there is a lack of models or datasets suitable for Korean users, and there are limitations to real-time response to reflect dynamic user states using fixed response patterns or independent conversation generation.

[0005] In other words, existing conversational systems lack the capability to dynamically adjust conversational strategies by analyzing the user's state in real time, and there is a absence of emotional support data and models specialized for Korean speakers, as well as limitations in technology that simultaneously satisfies high-speed processing and response suitability.

[0006] The aforementioned background technology is technical information that the inventor possessed for the derivation of the present invention or acquired during the process of deriving the present invention, and it cannot be considered as publicly known technology disclosed to the general public prior to the filing of the present invention. Prior art literature

[0007] Published Patent Application No. 10-2025-0006371 (January 13, 2025) The problem to be solved

[0008] In an embodiment of the present invention, we propose an adaptive conversational environment for depressive disorders that quantitatively analyzes the user's psychological state based on the criteria of the DSM-5 (diagnostic and statistical manual of mental disorders, fifth edition) and provides customized emotional support tailored to Korean users.

[0009] In an embodiment of the present invention, we propose an adaptive conversational environment for depressive disorders that can provide emotional support and effective counseling strategies through real-time conversation with a user by calculating a depression intensity score through an artificial neural network, summarizing the conversation, and selecting (returning) a response.

[0010] In an embodiment of the present invention, we propose an adaptive conversational environment for depressive disorders that can provide a user-customized conversational experience through quantitative analysis based on the DSM-5 using Korean text data.

[0011] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems to be solved will be clearly understood by those skilled in the art to which the present invention pertains from the descriptions below. means of solving the problem

[0012] According to an embodiment of the present invention, a computing device for providing an adaptive conversation for depressive disorders may be provided, comprising: an acquisition unit for acquiring a user's utterance; a storage unit including a command for returning a response to the utterance using a pre-trained artificial neural network and calculating the user's depression intensity score for each conversation turn of the utterance and the response; and a processing unit for processing to return a response to the utterance and calculate the user's depression intensity score for each conversation turn by executing the command, wherein the artificial neural network includes a pre-trained artificial neural network that quantitatively analyzes text regarding the utterance using a KoELECTRA (Korean efficiently learning an encoder that classifies token replacements accurately) model and calculates the depression intensity score based on the DSM-5 (diagnostic and statistical manual of mental disorders, fifth edition) criteria for the analysis.

[0013] Here, the processing unit may include: a depression intensity detection unit that calculates the depression intensity score according to the DSM-5 criteria and calculates the average of the depression intensity scores according to the conversation turn; a conversation summary unit that outputs a conversation summary of the utterance and the response according to the conversation turn; and a response selection unit that determines the similarity between the utterance and a plurality of response candidates and returns the response.

[0014] In addition, the depression intensity detection unit can calculate the mean through the regression layer of the artificial neural network and train the artificial neural network using the root-mean-square deviation (RMSE) loss function.

[0015] In addition, the conversation turns for the conversation summary output from the conversation summary section can be based on 4 turns.

[0016] In addition, the above conversation summary can be used as the context for the following conversation.

[0017] Additionally, the response selection unit may include: a poly-encoder that first sorts the rankings of the plurality of response candidates to filter initial response candidates by determining the similarity; a cross-encoder that secondarily sorts the rankings of the plurality of response candidates based on the rankings from the first sorting to generate a context score; an epitome re-ranker that generates an emotional score by adjusting the conversation direction based on the average; and a score aggregation unit that aggregates the context score and the emotional score to set a final response ranking and returns the response based on the final response ranking.

[0018] In addition, the above-mentioned Epitome Reranker can adjust the direction of the conversation by dynamically switching between exploration questions and emotional responses.

[0019] According to an embodiment of the present invention, a method for providing an adaptive conversation for depressive disorder in a computing device comprises: activating a conversational bot within the computing device and acquiring a user's utterance; returning a response from the utterance using a pre-trained artificial neural network; calculating a depression intensity score of the user for each conversation turn of the utterance and the response using the artificial neural network; and proceeding the conversation turn for a pre-set number of turns and outputting a conversation summary for the utterance and the response according to the pre-set turns; wherein the artificial neural network includes a pre-trained artificial neural network that quantitatively analyzes the text of the utterance using a KoELECTRA model and calculates the depression intensity score according to DSM-5 standards for the analysis.

[0020] Here, the returning step may include a step of determining the similarity between the utterance and a plurality of response candidates and returning the response.

[0021] Additionally, the above-mentioned calculating step may include the step of calculating the depression intensity score according to the DSM-5 criteria and calculating the average of the depression intensity scores according to the conversation turn.

[0022] In addition, the above-mentioned calculating step may include the step of calculating the mean through the regression layer of the artificial neural network and training the artificial neural network using an RMSE loss function.

[0023] In addition, the above-mentioned pre-set turns may include 4 turns.

[0024] In addition, the above conversation summary can be used as the context for the following conversation.

[0025] Additionally, the method may further include: a step of first sorting the rankings of the plurality of response candidates to filter initial response candidates by determining similarity after performing the step of outputting the conversation summary; a step of secondarily sorting the rankings of the plurality of response candidates based on the rankings from the first sorting to generate a context score; a step of adjusting the conversation direction based on the average to generate an empathy score; and a step of combining the context score and the empathy score to set a final response ranking and returning the response based on the final response ranking.

[0026] Additionally, the step of generating the empathy score may include a step of adjusting the direction of the conversation by dynamically switching between inquiry-type questions and empathy-type responses.

[0027] According to an embodiment of the present invention, a computer-readable recording medium storing a computer program, wherein the computer program includes instructions for a processor to perform a method for providing a conversational adaptive to depressive disorder of a computing device, and the method comprises: a step of activating a conversational bot within the computing device and acquiring a user’s utterance; a step of returning a response from the utterance using a pre-trained artificial neural network; a step of calculating a depression intensity score of the user for each conversational turn of the utterance and the response using the artificial neural network; and a step of proceeding the conversational turn for a pre-set number of turns and outputting a conversational summary of the utterance and the response according to the pre-set turns; wherein the artificial neural network includes a pre-trained artificial neural network that quantitatively analyzes text regarding the utterance using a KoELECTRA model and calculates the depression intensity score according to DSM-5 standards for the analysis.

[0028] According to an embodiment of the present invention, a computer program stored on a computer-readable recording medium may be provided, wherein the computer program includes instructions for a processor to perform a method for providing a conversational adaptive to depressive disorder of a computing device, and the method comprises: a step of activating a conversational bot within the computing device and acquiring a user’s utterance; a step of returning a response from the utterance using a pre-trained artificial neural network; a step of calculating a depression intensity score of the user for each conversational turn of the utterance and the response using the artificial neural network; and a step of proceeding the conversational turn for a pre-set number of turns and outputting a conversational summary of the utterance and the response according to the pre-set turns; wherein the artificial neural network may be provided as a computer program stored on a recording medium that includes a pre-trained artificial neural network that quantitatively analyzes text regarding the utterance using a KoELECTRA model and calculates the depression intensity score according to DSM-5 standards for the analysis. Effects of the invention

[0029] According to an embodiment of the present invention, accurate and scientific evaluation results are provided by quantitatively analyzing the user's condition by applying clinical criteria based on the DSM-5, and emotional support optimized for domestic users can be provided by utilizing Korean datasets and models. Furthermore, according to an embodiment of the present invention, real-time response is possible by generating high-speed responses tailored to the user's condition through a multi-encoder-based response selection system, and psychological stability and emotional connection can be strengthened by effectively reflecting empathy in conversations with the user. Additionally, according to an embodiment of the present invention, efficient resource utilization is possible even in environments with limited computing resources by implementing operation through a lightweight model design. Brief explanation of the drawing

[0030] FIG. 1 is a block diagram illustrating the function of a computing device for providing an adaptive conversation for depressive disorder according to an embodiment of the present invention. FIG. 2 is a block diagram for specifically explaining the function of the processing unit in a computing device for providing an adaptive conversation for depressive disorder of FIG. 1. Figure 3 is a block diagram for specifically explaining the function of the depression intensity detection unit in the processing unit of Figure 2. Figure 4 is a block diagram for specifically explaining the function of the conversation summary unit in the processing unit of Figure 2. FIG. 5 is a block diagram for specifically explaining the function of the response selection unit in the processing unit of FIG. 2. FIG. 6 is a block diagram for explaining the functions of a computing device for providing an adaptive conversation for depressive disorder according to an embodiment of the present invention, and is a diagram for exemplarily explaining the overall functions of a depression intensity detection unit, a conversation summary unit, and a response selection unit included in the processing unit of FIG. 2. FIG. 7 is a flowchart illustrating, in an exemplary manner, a method for providing an adaptive conversation for depressive disorder in a computing device according to an embodiment of the present invention. Specific details for implementing the invention

[0031] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various forms. These embodiments are provided merely 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, and the scope of the present invention is defined only by the claims.

[0032] In describing the embodiments of the present invention, specific descriptions of known functions or configurations will be omitted unless actually necessary for describing the embodiments of the present invention. Furthermore, the terms described below are defined in consideration of the functions in the embodiments of the present invention, and these may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification.

[0033] Existing conversational systems lack the capability to dynamically adjust conversational strategies by analyzing the user's state in real time, and there is a absence of emotional support data and models specialized for Korean speakers. Furthermore, there are limitations in technology that simultaneously satisfies high-speed processing and response suitability.

[0034] Accordingly, in an embodiment of the present invention, we propose a system that provides real-time emotional support by integrating quantitative analysis based on DSM-5 standards, the use of a Korean dataset, and a multi-encoder-based response selection structure.

[0035] Specifically, in an embodiment of the present invention, the invention aims to provide quantitative analysis of user status and depression intensity assessment results based on DSM-5 criteria, provide emotional support specialized for Korean users, develop technology that dynamically adjusts conversation strategies in real time according to user status, and provide a response selection algorithm equipped with high-speed processing and high contextual appropriateness.

[0036] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0037] FIG. 1 is a block diagram illustrating the function of a computing device (100) for providing an adaptive conversation for depressive disorder according to an embodiment of the present invention.

[0038] A computing device (100) is an electronic device possessed by a user who needs to verify a user's psychological state, such as a depressive disorder. In an embodiment of the present invention, the electronic device may include at least one of, for example, a smartphone, an e-book reader, a tablet personal computer, a laptop personal computer, a netbook computer, a desktop computer, a server computer, and a wearable device (for example, a smart watch, smart glasses, a head-mounted device (HMD), electronic clothing, an electronic bracelet, an electronic necklace, a smart mirror, etc.). In another embodiment, the electronic device may include at least one of network security equipment, navigation equipment, marine electronic equipment (for example, a marine navigation device, a gyrocompass, etc.), avionics, a vehicle head unit, a point of sales (POS) of a store, and an Internet of Things device. In addition, the electronic device may be equipped with an application for executing various operations (speech input, data processing, response output, etc.) to provide an adaptive conversation for depressive disorder according to an embodiment of the present invention. The application may include, for example, a smartphone application, a PC (personal computer) application, a set-top box (STB) application, a web application, an instant application, etc., and is not limited to a specific application.

[0039] As illustrated in FIG. 1, a computing device (100) for providing an adaptive conversation for depressive disorder may include an acquisition unit (110), a processing unit (120), and a storage unit (130).

[0040] The acquisition unit (110) can acquire a user's utterance. At this time, the utterance may include, for example, the user's voice, text, etc., and the acquisition unit (110) may be configured with a communication device, an input device, etc. to acquire such user utterance. Here, the input device may provide a UI (user interface) environment for selecting and inputting the user's voice or text, and the UI environment may include input means such as, for example, a key pad, a touch pad, etc. Additionally, the communication device is equipment that connects the computing device (100) and a remote computing device (such as a mobile communication terminal), and such a communication device may include, for example, a short-range communication device such as Bluetooth, Zigbee, Wi-Fi, or UWB (ultra-wide band), a broadband communication device such as the Internet or a mobile communication network, and is not limited to a specific communication device.

[0041] The processing unit (120) processes to acquire a user's utterance through the acquisition unit (110) by executing a command within the storage unit (130), processes to output a response to the acquired user utterance using an artificial neural network within the storage unit (130), and can calculate the user's depression intensity score for each turn of the conversation consisting of utterance-response pairs. Here, an output device may be connected to the processing unit (120) to output a response to the user utterance to the outside, and the output device may provide a UI environment for outputting the like. Such an output device may include display means such as, for example, an LCD (liquid crystal display) or an OLED (organic light emitting diodes), and is not limited to a specific display means. Such a processing unit (120) may include, for example, a microprocessor-based processing device, and the specific configuration and function description of the processing unit (120) will be described in detail in FIGS. 2 to 7.

[0042] The storage unit (130) may include a command to return a response to a user utterance obtained through the acquisition unit (110) using a pre-trained artificial neural network, and to calculate the user's depression intensity score for each conversation turn of the user utterance and response. Any command within the storage unit (130) may be stored in the form of an application, a program, etc., and any stored command may be selected and executed by the processing unit (120). Such a storage unit (130) may include a recording medium such as memory (e.g., RAM, read-only memory), a local disk (local disc) connected to a network (not shown), or storage, and there is no need to be limited to a specific recording medium in implementing an embodiment of the present invention.

[0043] FIG. 2 is a block diagram for specifically explaining the function of the processing unit (120) in the computing device (100) for providing an adaptive conversation for depressive disorder of FIG. 1.

[0044] As illustrated in FIG. 2, the processing unit (120) may include a depression intensity detection unit (122), a conversation summary unit (124), and a response selection unit (126).

[0045] The depression intensity detection unit (122) can apply the DSM-5 (diagnostic and statistical manual of mental disorders, fifth edition) criteria when calculating the depression intensity score and can calculate the average value of the depression intensity score according to the conversation turn. According to an embodiment of the present invention, this depression intensity detection unit (122) can analyze the user's utterance (voice or text) using a KoELECTRA-based regression model and can process at least one of the processes of input data preprocessing, tokenization, model prediction, and result interpretation. Here, ELECTRA can proceed with learning using a replaced token detection method, that is, by looking at the token from the generator and determining whether the discriminator is a 'real' token or a 'fake' token, and KoELECTRA can proceed with learning using, for example, 14GB of Korean text.

[0046] FIG. 3 is a block diagram for specifically explaining the function of the depression intensity detection unit (122) in the processing unit (120) of FIG. 2.

[0047] As illustrated in FIG. 3, the depression intensity detection unit (122) may include a depression intensity calculation unit (122-1), calculate the average value of the depression intensity score through the regression layer of the artificial neural network, and train the artificial neural network using the root-mean-square deviation (RMSE) loss function.

[0048] This depression intensity calculation unit (122-1) can quantitatively analyze user input text using the KoELECTRA model to classify the user's current condition based on the nine major symptoms of DSM-5.

[0049] The nine major symptoms according to the DSM-5 criteria are as follows.

[0050] 1) I feel depressed almost all day long.

[0051] 2) Interest in almost everything or nearly everything, or in most daily activities, is significantly reduced.

[0052] 3) There is a phenomenon of weight loss or gain, or a decrease or increase in appetite.

[0053] 4) Insomnia or hypersomnia.

[0054] 5) Feel psychomotor agitation or retardation.

[0055] 6) Feel easily fatigued or experience a loss of energy.

[0056] 7) Feel a sense of worthlessness in life and inappropriate guilt.

[0057] 8) Feeling a decrease in thinking ability, memory, or concentration, or feeling indecisive.

[0058] 9) Have had recurring thoughts of death or suicidal thoughts, attempts, or plans.

[0059] Here, if symptoms 1) and 2) are essential and a total of 5 or more symptoms persist for 2 weeks or more, it can be diagnosed as depressive disorder.

[0060] This depression intensity calculation unit (122-1) processes input data by padding or truncating it into 128 tokens, and each input can be structured into [CLS] and [SEP] tokens. The output of the model is converted into a final depression intensity score through a regression layer, and the model can be trained using an RMSE loss function.

[0061] Referring again to FIG. 2, the conversation summary unit (124) can output a conversation summary for a conversation turn consisting of utterance-response pairs.

[0062] FIG. 4 is a block diagram for specifically explaining the function of the conversation summary unit (124) in the processing unit (120) of FIG. 2.

[0063] As illustrated in FIG. 4, the conversation summary unit (124) may include a conversation turn progression unit (124-1), and the conversation can be summarized through the conversation turn progression unit (124-) to maintain context and solve the problem of the model's input length being limited. The summarized conversation can be stored in the storage unit (130) and can be used for subsequent conversations. At this time, the conversation turns for the conversation summary output from the conversation turn progression unit (124-1) can be conducted based on, for example, 4 turns (4 utterance-response pairs).

[0064] Referring again to FIG. 2, the response selection unit (126) can determine the similarity between the utterance and a plurality of response candidates and return a response.

[0065] FIG. 5 is a block diagram for specifically explaining the function of the response selection unit (126) in the processing unit (120) of FIG. 2.

[0066] As illustrated in FIG. 5, the response selection unit (126) may include a poly encoder (126-1), a cross encoder (126-2), an epitome reranker (126-3), and a score aggregation unit (126-4).

[0067] The poly encoder (126-1) can determine the similarity between the user's utterance and multiple response candidates and sort the rankings of multiple response candidates in order to filter the initial response candidate.

[0068] The cross encoder (126-2) can generate a context score by secondarily sorting the rankings of multiple response candidates based on the rankings from the first sorting of the poly encoder (126-1).

[0069] The epitom reranker (126-3) can generate an emotional score by adjusting the direction of conversation based on the average value of the depression intensity score according to the initial response candidates filtered by the poly encoder (126-1) and the conversation turn. At this time, the epitom reranker (126-3) can adjust the direction of conversation by dynamically switching between exploration questions and emotional reactions.

[0070] Epitome reranker (126-3) can include a regression classification model trained based on the epitome framework, and the epitome framework proposes a conversation direction as exemplified in [Table 1] below.

[0071]

[0072] In an embodiment of the present invention, two conversation directions among the conversation directions of [Table 1]—empathetic responses and exploratory questions—can be dynamically adjusted through the epitome reranker (126-3). For example, the epitome reranker (126-3) can label the conversation direction according to whether the response candidate group corresponds to an empathetic response or an exploratory question.

[0073] The score aggregation unit (126-4) can aggregate the context score of the cross encoder (126-2) and the empathy score of the epitome reranker (126-3) to set the final response ranking, and can return a response based on the set final response ranking.

[0074] Therefore, the present invention enables real-time adjustment of conversation strategies based on the user's average depression intensity.

[0075] FIG. 6 is a block diagram for explaining the functions of a computing device (100) for providing a conversational adaptation for depressive disorder according to an embodiment of the present invention, and is a diagram for exemplarily explaining the overall functions of a depression intensity detection unit (122), a conversation summary unit (124), and a response selection unit (126) included in the processing unit (120) of FIG. 2.

[0076] As illustrated in FIG. 6, the depression intensity detection unit (122) includes a depression intensity calculation unit (122-1), the conversation summary unit (124) includes a conversation turn progression unit (124-1), and the response selection unit (126) may include a poly encoder (126-1), a cross encoder (126-2), an epitome reranker (126-3), and a score aggregation unit (126-4).

[0077] Here, the poly encoder (126-1) receives multiple response candidates from the database (102) and determines the similarity between the provided multiple response candidates and the user utterance obtained from the acquisition unit (110) to sort the rankings of the multiple response candidates in the first order. The database (102) connected to the poly encoder (126-1) can be implemented in accordance with the purpose of the present invention using a relational database management system (RDBMS) such as Oracle, Informix, Sybase, DB2, for example, or an object-oriented database management system (OODBMS) such as Gemston, Orion, O2, etc., and can have appropriate fields to achieve its functions.

[0078] The cross encoder (126-2) performs a second sort of the rankings of multiple response candidates based on the rankings from the first sort of the poly encoder (126-1), and the epitome reranker (126-3) adjusts the direction of conversation based on the average value of the depression intensity scores according to the initial response candidates and the conversation turn.

[0079] The score aggregation unit (126-4) aggregates the context score of the cross encoder (126-2) and the empathy score of the epitome reranker (126-3) to determine the final response ranking.

[0080] A computing device (100) including such a depression intensity detection unit (122), a conversation summary unit (124), and a response selection unit (126) may include a process of processing user input, analyzing depression intensity, summarizing conversation, and selecting an optimal response.

[0081] That is, the input text is quantitatively evaluated by the KoELECTRA-based depression intensity detection unit (122). The conversation summary unit (124) summarizes the context, and the response selection unit (126) generates a response reflecting the empathy score using a poly encoder (126-1) and a cross encoder (126-2). This process takes place in real time, and the epitome reranker (126-3) can dynamically adjust the conversation strategy according to the user's state.

[0082] FIG. 7 is a flowchart illustrating, in an exemplary manner, a method for providing a depressive disorder adaptive conversation of a computing device (100) according to an embodiment of the present invention.

[0083] As illustrated in FIG. 7, the computing device (100) can activate a conversational bot, for example, a chatbot, and obtain a user's utterance (S100, S102).

[0084] When a user's utterance is obtained, the computing device (100) can return a response corresponding to the utterance through the response selection unit (126) (S104).

[0085] Afterwards, the computing device (100) can calculate a depression intensity score for each conversation turn consisting of a pair of speech-response using a depression intensity detection unit (122) (S106).

[0086] Afterward, the computing device (100) determines whether a preset number of conversation turns (n) have been conducted (S108), and if it is determined that a preset number of conversation turns (n) have been conducted, the computing device (100) can output a conversation summary for the utterance-response pairs according to the turn using the conversation summary unit (124) (S110). Here, the preset number of conversation turns can be conducted based on, for example, 4 turns (4 utterance-response pairs).

[0087] At this time, in step (S110), the computing device (100) can calculate the average value of the depression intensity score according to 4 turns using the depression intensity detection unit (122).

[0088] (Experimental Example)

[0089] In this invention, experiments were conducted using Korean-based datasets such as BWS data, DSM-5 labeling data, and the AI ​​Hub emotion conversation corpus, and the following experimental results were obtained.

[0090] - Depression intensity detection unit: RMSE 0.9452, R² 0.9123

[0091] - Conversation Summary: ROUGE-1 0.478, ROUGE-2 0.243

[0092] - Response selector: hits@1 0.8792, MRR 0.9532

[0093] Through this invention, it is expected that it can be utilized in the development of mental health counseling and digital therapy platforms, Korean-based emotion analysis and counseling chatbots, etc.

[0094] According to the embodiments of the present invention described above, accurate and scientific evaluation results are provided by quantitatively analyzing the user's condition by applying clinical criteria based on the DSM-5, and emotional support optimized for domestic users can be provided by utilizing Korean datasets and models. Furthermore, according to the embodiments of the present invention, real-time response is possible by generating high-speed responses tailored to the user's condition through a multi-encoder-based response selection system, and psychological stability and emotional connection can be strengthened by effectively reflecting empathy in conversations with the user. Additionally, according to the embodiments of the present invention, efficient resource utilization is possible even in environments with limited computing resources by implementing operation through a lightweight model design.

[0095] Meanwhile, combinations of each block of the attached block diagram and each step of the flowchart may be executed by computer program instructions. Since these computer program instructions can be loaded into the processor of a general-purpose computer, a specialized computer, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create a means to perform the functions described in each block of the block diagram.

[0096] Since these computer program instructions may be stored in a computer-available or computer-readable recording medium (or memory), etc., which can be directed toward a computer or other programmable data processing equipment to implement a function in a specific way, the instructions stored in the computer-available or computer-readable recording medium (or memory) may also be used to produce a manufactured item containing instruction means that perform the function described in each block of the block diagram.

[0097] And, since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that perform a series of operation steps on a computer or other programmable data processing equipment to create a process executed by a computer and perform the computer or other programmable data processing equipment may also provide steps for executing the functions described in each block of the block diagram.

[0098] Additionally, each block may represent a module, segment, or part of code containing at least one executable instruction for executing a specified logical function(s). Also, it should be noted that in some alternative embodiments, the functions mentioned in the blocks may occur out of order. For example, two blocks shown in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order according to the corresponding function. Explanation of the symbols

[0099] 100: Computing device for providing adaptive dialogue for depressive disorder 110: Acquisition Department 120: Processing unit 122: Depression Intensity Detector 122-1: Depression Intensity Calculator 124: Conversation Summary 124-1: Dialogue Turn Progression 126: Response selection part 126-1: Poly Encoder 126-2: Cross Encoder 126-3: Epitome Reranker 126-4: Score Compilation Section 130: Storage section

Claims

Claim 1 A computing device for providing an adaptive conversation for depressive disorders, comprising: an acquisition unit for acquiring a user's utterance; a storage unit including instructions for returning a response to the utterance using a pre-trained artificial neural network and calculating the user's depression intensity score for each conversation turn of the utterance and the response; and a processing unit for processing to return a response to the utterance and calculate the user's depression intensity score for each conversation turn by executing the instructions; wherein the artificial neural network includes a pre-trained artificial neural network that quantitatively analyzes the text of the utterance using a KoELECTRA (Korean efficiently learning an encoder that classifies token replacements accurately) model and calculates the depression intensity score based on the DSM-5 (diagnostic and statistical manual of mental disorders, fifth edition) criteria for the analysis. Claim 2 A computing device for providing a conversation adaptive to depressive disorder, wherein the processing unit comprises: a depression intensity detection unit that calculates the depression intensity score according to the DSM-5 criteria and calculates the average of the depression intensity scores according to the conversation turn; a conversation summary unit that outputs a conversation summary for the utterance and the response according to the conversation turn; and a response selection unit that determines the similarity between the utterance and a plurality of response candidates and returns the response. Claim 3 In claim 2, the depression intensity detection unit calculates the mean through the regression layer of the artificial neural network and trains the artificial neural network using a root-mean-square deviation (RMSE) loss function, a computing device for providing a conversational adaptation for depression. Claim 4 In claim 2, the conversation turns for the conversation summary output from the conversation summary unit are based on 4 turns, and the computing device for providing a conversation adaptive for depressive disorder. Claim 5 In claim 4, the above conversation summary is utilized as the context of the next conversation. A computing device for providing an adaptive conversation for depressive disorder. Claim 6 In claim 2, the response selection unit comprises: a poly-encoder that first sorts the ranks of the plurality of response candidates to filter initial response candidates by determining the similarity; a cross-encoder that secondarily sorts the ranks of the plurality of response candidates based on the ranks from the first sorting to generate a context score; an epitome re-ranker that generates an emotional score by adjusting the conversation direction based on the average; and a score aggregation unit that aggregates the context score and the emotional score to set a final response rank and returns the response based on the final response rank; a computing device for providing an adaptive conversation for depressive disorder. Claim 7 In claim 6, the epitome reranker is a computing device for providing an adaptive conversation for depressive disorder that dynamically switches between exploration and emotional reactions to adjust the direction of the conversation. Claim 8 A method for providing an adaptive conversation for depressive disorder in a computing device, comprising: activating a conversational bot within the computing device and acquiring a user's utterance; returning a response from the utterance using a pre-trained artificial neural network; calculating a depression intensity score of the user for each conversation turn of the utterance and the response using the artificial neural network; and proceeding the conversation turn for a pre-set number of turns and outputting a conversation summary for the utterance and the response according to the pre-set turns; wherein the artificial neural network includes a pre-trained artificial neural network that quantitatively analyzes the text of the utterance using a KoELECTRA model and calculates the depression intensity score according to DSM-5 standards for the analysis. Claim 9 In claim 8, the returning step comprises the step of determining the similarity between the utterance and a plurality of response candidates and returning the response, a method for providing an adaptive conversation for depression in a computing device. Claim 10 In claim 9, the calculating step comprises the step of calculating the depression intensity score according to the DSM-5 criteria and calculating the average of the depression intensity scores according to the conversation turn, a method for providing an adaptive conversation for depressive disorders on a computing device. Claim 11 In claim 10, the calculating step comprises the step of calculating the mean through the regression layer of the artificial neural network and training the artificial neural network using an RMSE loss function. A method for providing an adaptive conversation for depressive disorder on a computing device. Claim 12 In claim 8, the above-mentioned preset turn is a method for providing an adaptive conversation for depressive disorder in a computing device comprising 4 turns. Claim 13 In claim 8, the above conversation summary is utilized as the context of the next conversation. A method for providing an adaptive conversation for depressive disorders on a computing device. Claim 14 A method for providing an adaptive conversation for depressive disorder in a computing device according to claim 10, further comprising: a step of first sorting the rankings of the plurality of response candidates to filter initial response candidates by determining similarity after performing the step of outputting the conversation summary; a step of secondarily sorting the rankings of the plurality of response candidates based on the rankings from the first sorting to generate a context score; a step of adjusting the conversation direction based on the average to generate an empathy score; and a step of combining the context score and the empathy score to set a final response ranking and returning the response based on the final response ranking. Claim 15 In claim 14, the step of generating the empathy score comprises the step of dynamically switching between inquiry-type questions and empathy-type responses to adjust the direction of the conversation, thereby providing a method for providing an adaptive conversation for depressive disorder on a computing device. Claim 16 A computer-readable recording medium storing a computer program, wherein the computer program includes instructions for a processor to perform a method for providing a conversational adaptive for depressive disorder of a computing device, and the method comprises: a step of activating a conversational bot within the computing device and acquiring a user's utterance; a step of returning a response from the utterance using a pre-trained artificial neural network; a step of calculating a depression intensity score of the user for each conversational turn of the utterance and the response using the artificial neural network; and a step of proceeding the conversational turn for a pre-set number of turns and outputting a conversational summary of the utterance and the response according to the pre-set turns; wherein the artificial neural network includes an artificial neural network pre-trained to quantitatively analyze text regarding the utterance using a KoELECTRA model and to calculate the depression intensity score according to DSM-5 standards for the analysis. Claim 17 A computer program stored on a computer-readable recording medium, wherein the computer program includes instructions for a processor to perform a method for providing a conversational adaptive for depressive disorder of a computing device, and the method comprises: a step of activating a conversational bot within the computing device and acquiring a user's utterance; a step of returning a response from the utterance using a pre-trained artificial neural network; a step of calculating a depression intensity score of the user for each conversational turn of the utterance and the response using the artificial neural network; and a step of advancing the conversational turn by a pre-set number of turns and outputting a conversational summary of the utterance and the response according to the pre-set turns; wherein the artificial neural network is a computer program stored on a recording medium that includes a pre-trained artificial neural network that quantitatively analyzes text regarding the utterance using a KoELECTRA model and calculates the depression intensity score according to DSM-5 standards for the analysis.

Images