Systems and methods for psychotherapy using artificial intelligence

CN117412702BActive Publication Date: 2026-10-27邹岚
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202280027564.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2021-10-13
Filing Date
2022-02-14
Publication Date
2026-10-27
Estimated Expiration
2042-02-14

Smart Images

  • Figure CN117412702B_ABST
    Figure CN117412702B_ABST
Patent Text Reader

Abstract

Systems and methods for providing psychotherapy through artificial intelligence (AI) are described herein. The AI psychotherapy system obtains patient messages from a patient through a user interface. The AI psychotherapy system then determines one or more classifiers based on the patient messages and other information about the patient. The other information about the patient includes in-hospital and out-of-hospital data. The AI psychotherapy system then selects a psychotherapy for the patient from a list of candidate psychotherapies. Subsequently, the AI psychotherapy system provides the psychotherapy to the patient in one of a text format or a speech format based on patient preferences.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The disclosed embodiments generally relate to the diagnosis and treatment of mental health disorders, and more specifically, to systems and methods for psychotherapy using artificial intelligence (AI).

[0002] background Talk therapy can be very helpful for patients suffering from major depressive disorder, anxiety disorder, and other mental health problems or disorders.

[0003] Traditional talk therapy involves face-to-face, in-person conversations, with each session typically lasting nearly an hour. However, low moods and suicidal ideation are more likely to occur at night or when the patient is alone.

[0004] summary Therefore, there is a need for systems and methods to transform face-to-face communication into text messaging and voice interaction. The digital psychotherapists described in this article are particularly useful in the evenings or when patients are alone, providing 24 / 7 companionship and targeted, effective, personalized, and customized emotional support and psychotherapy. According to some embodiments, digital psychotherapists utilize artificial intelligence and psychotherapy knowledge systems to help patients cope with psychological problems and emotional distress. These systems and methods are based on psychotherapy theories and evidence-based psychotherapy used by human psychologists worldwide.

[0005] Systems and methods for providing AI-based psychotherapy are described. These systems and methods utilize artificial intelligence and / or machine learning to perform evidence-based psychotherapy (e.g., cognitive behavioral therapy (CBT), dialectical behavioral therapy (DBT), interpersonal therapy (IPT)) and other related treatments and therapies without human intervention. The AI-based psychotherapy system obtains patient messages (sometimes referred to as user messages) from the patient through a user interface. The system then determines one or more natural language processing (NLP) classifiers, task-oriented dialogue systems, machine learning predictors, intent classifiers, recommendation engines, and named entity recognizers based on the patient messages and other information about the patient. This other information about the patient may include in-hospital and out-of-hospital data and types of cognitive distortions. Next, the system selects a psychotherapy for the patient based on a list of candidate cognitive techniques. Finally, the system delivers the psychotherapy to the patient in text or voice format, according to the patient's preferences.

[0006] This application discloses the subject matter corresponding to the following numbered clauses: (A1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, the method comprising: obtaining patient messages from a patient via a user interface; identifying emotions in user input and associating them with emotion-related terms based on the patient messages using one or more named entity recognizers; and / or identifying terms in user input as terms related to psychotherapy, mental disorders, psychological problems, and / or psychology based on the patient messages using one or more named entity recognizers.

[0007] (A2) The method described in Clause (A1) also includes the use of knowledge graphs relating to the following areas: psychology, psychotherapy, mental disorders, psychological problems and / or emotional domains.

[0008] (A3) The method according to any one of clauses (A1)-(A2), wherein the user input may be in text format or voice format, the method further comprising: converting any voice input into text input, and using the text input to identify emotions and terms in the user input.

[0009] (B1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, the method comprising: obtaining patient messages from a patient via a user interface; and, based on the patient messages, using one or more natural language processing (NLP) classifiers to identify and classify terms in user input into one or more terms in psychotherapy, emotion, mental disorder, psychological problem and / or psychology.

[0010] (B2) The method described in Clause (B1) also includes the use of knowledge graphs relating to the following areas: psychology, psychotherapy, mental disorders, psychological problems and / or emotional domains.

[0011] (B3) The method according to any one of clauses (B1)-(B2), wherein the user input may be in text format or voice format, the method further comprising: converting any voice input into text input, and using the text input to identify and classify terms in the user input.

[0012] (C1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, the method comprising: obtaining patient messages from a patient via a user interface; and, based on the patient messages, identifying intent in the user input using one or more natural language processing (NLP) intent recognizers, one or more topic classifiers, one or more sentiment analysis models and / or one or more intent classifiers.

[0013] (C2) The method described under Clause (C1) also includes the use of knowledge graphs relating to the following areas: psychology, psychotherapy, mental disorders, psychological problems and / or emotional domains.

[0014] (C3) The method according to any one of clauses (C1)-(C2), wherein the user input may be in text format or voice format, the method further comprising: converting any voice input into text input and using the text input to identify intent.

[0015] (D1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, comprising: obtaining patient messages and timestamps from the patient via a user interface; or obtaining one or more emojis and timestamps from the patient via a user interface; and predicting future psychological conditions and treatment plans using one or more machine learning prediction models based on the timestamps and patient messages or emojis.

[0016] (D2) The method according to clause (D1), wherein the user input may be in text or voice format, the method further comprising: converting any voice input into text input, and using the text input to predict future mental conditions and treatment plans.

[0017] (D3) The method according to any one of clauses (D1)-(D2), wherein predicting future mental condition and treatment plan includes: acquiring facial images and videos of the patient; and using mental condition recognition technology based on facial images and videos.

[0018] (D4) The method according to any one of clauses (D1)-(D3), wherein predicting future psychological conditions and treatment plans includes: collecting heart rate data from patients; and analyzing the heart rate data using a time series model.

[0019] (D5) The method according to any one of clauses (D1)-(D4), wherein predicting future psychological conditions and treatment plans includes: collecting electrocardiogram data of the patient; and analyzing the electrocardiogram data using a machine learning pattern recognition model.

[0020] (D6) The method according to any one of the clauses (D1)-(D5), wherein predicting future mental conditions and treatment plans includes: responding to received timestamps, new patient messages, emojis, facial images, facial videos, heart rate data, electrocardiograms and / or medical records; repeating the prediction of future mental conditions and treatment plans; and storing new predictions of future mental conditions and treatment plans.

[0021] (E1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, the method comprising: obtaining patient messages from the patient through a user interface; and identifying cognitive distortions based on the patient messages using one or more natural language processing (NLP) classifiers.

[0022] (E2) The method described under clause (E1) further includes: using a list of cognitive distortions to identify cognitive distortions.

[0023] (E3) The method according to any one of the clauses (E1)-(E2), wherein the user input may be in text format or voice format, the method further comprising: converting any voice input into text input and using the text input to identify cognitive distortions.

[0024] (F1) An intelligently controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, the method comprising: obtaining patient information from the patient through a user interface; and determining personalized psychotherapy based on the patient information using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills.

[0025] (F2) The method described under Clause (F1) also includes the use of knowledge graphs relating to the following areas: psychology, psychotherapy, mental disorders, psychological problems and / or emotional domains.

[0026] (F3) The method according to any one of the clauses (F1)-(F2), wherein the user input may be in text or voice format, the method further comprising: converting any voice input into text input and using the text input to generate personalized psychotherapy.

[0027] (F4) The method according to any one of the clauses (F1)-(F3), wherein determining individualized psychotherapy includes: using the patient’s feedback history, the feedback history of each psychotherapy and / or a list of cognitive distortions.

[0028] (F5) The method according to any one of the clauses (F1)-(F4), wherein determining personalized psychotherapy includes: after receiving feedback from the patient confirming that the psychotherapy is ineffective: obtaining a second user input from the patient; and repeating the determination of personalized psychotherapy and providing new psychotherapy for the second user input.

[0029] (G1) A smart-controlled psychotherapy method that utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention, comprising: obtaining patient messages from the patient through a user interface; and providing personalized psychotherapy based on the patient messages using (i) one or more natural language processing (NLP) task-oriented dialogue systems and (ii) a list of cognitive skills.

[0030] (G2) The method described under Clause (G1) also includes the use of knowledge graphs relating to the following areas: psychology, psychotherapy, mental disorders, psychological problems and / or emotional domains.

[0031] (G3) The method according to any one of clauses (G1)-(G2), wherein the user input may be in text or voice format, the method further comprising: converting any voice input into text input and using the text input to generate personalized psychotherapy.

[0032] (G4) The method according to any one of clauses (G1)-(G3) further includes: performing psychotherapy in a speech format based on the determination of the patient's preferred speech output.

[0033] (G5) The method according to any one of clauses (G1)-(G4), wherein providing personalized psychotherapy includes: responding upon receiving confirmation from the user that the psychotherapy is ineffective; obtaining a second user input from the patient; and repeatedly generating personalized psychotherapy and providing new psychotherapy for the second user input.

[0034] (G6) The method according to any one of clauses (G1)-(G5), wherein providing psychotherapy includes engaging in dialogue with the patient using the following models: (i) a closed-domain rule-based model, (ii) a closed-domain retrieval-based model, (iii) a closed-domain retrieval-generated hybrid model, (iv) a closed-domain generation-based model, (v) an open-domain rule-based model, (vi) an open-domain retrieval-based model, (vii) an open-domain retrieval-generated hybrid model, or (viii) an open-domain generation-based model.

[0035] (G7) The method according to any one of the clauses (G1)-(G6) further includes: collecting the patient’s medical records; and using OCR text recognition technology to read non-electronic information on the medical records.

[0036] (G8) The method according to any one of (G1)-(G7) further includes: generating a preferred speech using speech synthesis technology based on the patient's speech preferences; and using the generated preferred speech to converse with the patient.

[0037] According to some embodiments, a non-transitory computer-readable storage medium is provided. This non-transitory computer-readable storage medium stores one or more programs for execution by a computer system having one or more processors. The one or more programs include instructions for performing the methods described in any one of clauses (A1)-(G8).

[0038] Therefore, the system provides improved methods for delivering on-demand, targeted, effective, personalized, and customized emotional psychotherapy.

[0039] Brief description of the attached figures To better understand the various embodiments described, reference should be made to the following description of the embodiments in conjunction with the accompanying drawings. In all the figures, the same reference numerals refer to corresponding parts.

[0040] Figure 1 A block diagram of an artificial intelligence psychotherapy system according to some embodiments is shown.

[0041] Figure 2 Example block diagrams of task-oriented dialogue systems using open-domain or closed-domain natural language processing (NLP) models are shown, according to some embodiments.

[0042] Figure 3A and Figure 3B A flowchart is shown illustrating an example method of psychotherapy performed on a user interface by an artificial intelligence psychotherapy system, according to some embodiments.

[0043] Figure 4 The diagram shows a flowchart of an example method for psychoanalyzing patient messages using natural language processing (NLP) preprocessing methods, according to some embodiments.

[0044] Figure 5 Examples of patient in-hospital and out-of-hospital data collected and / or used by an artificial intelligence psychotherapy system in health data collection 808, according to some embodiments, are shown.

[0045] Figure 6 These are examples of patient treatment timelines that may be used by artificial intelligence psychotherapy systems, according to some embodiments.

[0046] Figure 7 A schematic diagram of a speech synthesis workflow according to some embodiments is shown.

[0047] Figures 8A-8C The flowchart shows an example process of a user interacting with an artificial intelligence psychotherapist system according to some embodiments.

[0048] Figure 9 , 10Figures 11A and 11B show examples of user interfaces for interacting with an artificial intelligence psychotherapy system, according to some embodiments.

[0049] In all the accompanying drawings, the same reference numerals refer to the same or corresponding parts in each drawing.

[0050] Detailed description Embodiments will now be described in detail, with examples shown in the accompanying drawings. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the various described implementations. In other instances, well-known methods, processes, components, circuits, and networks are not described in detail to avoid unnecessarily obscuring aspects of the implementation.

[0051] It should also be understood that although the terms first, second, etc., are used in some cases to describe various elements herein, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first electronic device may be referred to as a second electronic device, and similarly, a second electronic device may be referred to as a first electronic device, without departing from the scope of the various implementations described. Both the first electronic device and the second electronic device are electronic devices, but they are not the same electronic device.

[0052] The terminology used in the description of the various embodiments herein is for the purpose of describing particular embodiments only and is not intended to be limiting. In the description of the various described embodiments and the appended claims, the singular forms “a,” “an,” and “the” are used to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and covers any and all possible combinations of one or more of the associated listed items. It should also be understood that when the terms “comprising” and / or “including” are used in this specification, they specify the presence of the stated features, integrals, steps, operations, elements, and / or components. However, the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0053] As used herein, the word "if" can be interpreted, depending on the context, as meaning "when," "depending on," "in response to determine," "in response to detect," or "according to determination." Similarly, the phrase "if determined" or "if [the condition or event] is detected" can optionally be interpreted as meaning "when determined," "in response to determine," "when [the condition or event] is detected," "in response to the detection of [the condition or event]," or "according to the determination of the detection of [the condition or event]," depending on the context.

[0054] Patients who cannot afford the high costs of counseling with a human psychotherapist can use the techniques and systems described herein. Such embodiments are particularly useful for those who need but cannot access human psychotherapists or other therapeutic resources in a timely manner. Systems and methods based on the techniques described herein provide conversational therapy, evidence-based psychotherapy, and other related treatments and therapies. The system is capable of providing professional, effective, personalized, low-cost, readily available, continuous, systematic, timely, and side-effect-free solutions. Such systems can be used to provide solutions for general psychological problems, severe psychological problems, neurotic psychological problems, and mental disorders based on text or voice, through artificial intelligence (AI) technology, in multiple languages ​​(e.g., English, Chinese, and other Chinese dialects, including Mandarin, Cantonese, Shanghainese, Hakka, etc.) and different scripts (e.g., Chinese characters, such as Simplified or Traditional Chinese). The system and methods can be offered in markets where English, Chinese, and / or other Chinese dialects are the primary languages. The software can be installed on various devices, including but not limited to electronic mobile devices, tablets, smartphones, personal computers, web-based devices or platforms, wearable devices, and other devices that use electronic interfaces for human-computer interaction. This system and method can be used to provide digital psychotherapy, counseling and first aid methods during mood disorders, and for general psychological problems, severe psychological problems, neurotic psychological problems and mental disorders (e.g., depression, anxiety), based on speech and text, such as Chinese and / or different Chinese dialects.

[0055] According to some embodiments, the system is capable of executing data processing methods and techniques, and includes endpoints, interfaces, inputs and outputs (including but not limited to content, format, and methods) of models (e.g., machine learning models, prediction models, neural network weights), connection and communication methods (including but not limited to various network communication interfaces and related methods, data storage management system interfaces and related processes), programs and their order (including but not limited to data processing, transformation, compression, extraction, generation, analysis, and other data and algorithm-related programs and their order, programs and their order of each module in an end-to-end model, and software programs and their order between front-end, data processing, and back-end modules), content (including but not limited to data content, instruction content, stored content, generated content, and content shared by intermediate modules), and processing and storage (including but not limited to content, interface, mode, format, and method).

[0056] Figure 1A block diagram illustrating an example of an artificial intelligence (AI) psychotherapy system 100 according to some embodiments is shown. According to some embodiments, in the AI ​​psychotherapy system 100, a user interface 102 is connected to a server 104. In some embodiments, the server 104 checks user authentication 106, and upon authorization (e.g., a "200" status response received upon successful user authentication), user authentication 106 sends data to a queue generator 108.

[0057] According to some embodiments, the AI ​​psychotherapy system 100 can realize a human-computer interactive digital psychotherapist (e.g., a robot), which may be equipped with voice interaction and text dialogue capabilities (e.g., Mandarin and other Chinese dialects). In some embodiments, the AI ​​psychotherapy system 800 can communicate with patients on an electronic device 132 with a screen, and achieve various functions through natural language processing technology, such as emotion-related counseling, psychotherapy, relief of symptoms of mental disorders, and tracking and predicting treatment effects. Through deep learning or machine learning algorithms and statistical techniques, the system and methods can continuously learn and evolve to understand the patient's personal, mental, and biological characteristics, including daily behaviors and mental changes under the combined influence of internal physical indicators and external environmental factors. In this way, the system can learn continuously.

[0058] As described in this paper, Natural Language Processing (NLP) and Computer Vision are cross-domain subfields spanning AI and human languages ​​(e.g., text and speech), image processing and computational intelligence, and machine learning. Due to the significant differences in nature and grammar between Chinese and alphabetic languages ​​(e.g., English), NLP analysis and processes differ considerably, increasing the complexity of such systems. In some embodiments, these systems and methods utilize AI techniques, deep learning algorithms, machine learning algorithms, and statistical techniques for processing.

[0059] Psychotherapy Process According to certain embodiments, Figures 8A-8CA flowchart illustrating an example flow 800 of a user interaction with an AI psychotherapist system 100 (sometimes referred to as an AI psychotherapy system) is shown. In some embodiments, the AI ​​psychotherapy system 100 determines whether the user is registered and / or logged in at 802. If the user is a new patient at 806, the AI ​​psychotherapy system 100 may prompt the patient to submit input, as shown in health data collection at 808. In some embodiments, after the patient successfully logs in at 804, the AI ​​psychotherapy system 100 may use biometric information collection and biometric verification at 814 to verify the patient's identity and analyze the biometric information. In some embodiments, additional biometric information collection and biometric analysis at 812 are performed. The biometric identifier 812 may include (i) a health record identifier, (ii) a heart rate time series identifier, and (iii) an electrocardiogram pattern identifier. Once the patient is successfully verified, the AI ​​psychotherapy system 100 may collect information about the patient at the patient's current mood and / or emotion collector at 816. According to some embodiments, the output of the psychologically relevant emotion classifier 822 (sometimes referred to as the psychologically relevant emotion classifier 350) can be the input of the emotion and / or mood tracker and predictor 824 (sometimes referred to as the emotion and / or mood tracker and predictor 352).

[0060] Features and Functions: Patient Messages According to some embodiments, patients can input the artificial intelligence psychotherapist system 100 through user interface 102 from one or more terminals such as computers, mobile phones, websites, smartwatches or other electronic platforms by entering text messages or speaking voice messages, such as "I feel so lonely" or "I am feeling so lonely now".

[0061] For example, according to some embodiments, the system can begin a conversation with the patient with "How are you feeling today?". The dialogue below is an example of a conversation between a digital psychotherapist (T) and a patient (U). T: How are you feeling today? U: Hmm... I didn't sleep well last night, I couldn't fall asleep and kept waking up.

[0062] T: Are you still worried about your business trip tomorrow? U: Ugh...yes, it's so annoying...I probably won't be able to sleep tonight.

[0063] T: Let me check your outpatient data.

[0064] T: It seems you're currently taking a prescription benzodiazepine; it helps you sleep. Try taking it 30 minutes before bed. Putting some lavender essential oil in your bedroom can also help you feel calm.

[0065] Data collection and storage system for health data and records inside and outside the hospital According to some embodiments, the AI-powered psychotherapy system 100 may request the patient's permission to collect and store the patient's in-hospital data and records, as well as out-of-hospital data (including physical and mental biometrics). Some embodiments may display a "microphone" button for collecting the patient's voice messages. In some embodiments, the device's camera (e.g., the front-facing camera of an iPhone) may collect images and videos of the patient's face. Other information, such as text messages, conversation content, self-assessments of mental health, heart rate, etc., may also be collected and stored by the health data collection 808. Figure 5 An example design of a data storage management system according to certain embodiments is shown, and Figure 6 Examples of health records for different time periods are described according to certain embodiments.

[0066] Features and Functions: Depression Detection Based on Patient Facial Images and Videos According to some embodiments, in the collection of biometric information and biometric verification 814, the input is a real-time facial image or video captured by the camera in the patient's (user_ID) mobile phone, or an uploaded facial image and record with a timestamp.

[0067] The depression detector API is based on neural networks (such as neural networks, long short-term memory, and deep belief networks) and statistical modeling (such as analysis of variance).

[0068] It can use facial expressions, eye movements, gestures, and human figures as features to identify a patient's positive / negative / neurotic emotions or fine-grained sentiments (such as depression, sadness, surprise, etc.). The output is a sentiment classification label with a significance value (p-value) and confidence score, as well as a fine-grained sentiment classification label with the corresponding user (user_ID), data type, quantity (video duration or image count), and inference timestamp. According to some embodiments, sample data might look like this: (1) user_ID:98803311, device_ID:001, real-time image, 1, severe depression -95% (p value <0.05), January 2, 2022, 13:02; (2) user_ID: 00313423, device_ID: 001, uploaded video, 10 seconds, bipolar disorder -60% (p value < 0.1), November 15, 2021, 11:58.

[0069] Figure 3A and Figure 3B A flowchart is shown of an example method 300 in which an artificial intelligence psychotherapy system 100 performs psychotherapy on a user interface, according to certain embodiments.

[0070] Features and functions: Heart rate measurement system The link between heart rate and mental disorders (such as major depressive disorder or bipolar disorder) has been established. Multiple studies have shown that depressed patients typically experience higher heart rates at night, while a normal person's heart rate should decrease without any medication. According to some embodiments, the AI-powered psychotherapy system 100 can use the API of a heart rate monitor to measure a patient's immediate and long-term continuous heart rate (e.g., during sleep), via touch (e.g., through a digital watch's API) or without physical contact. Biometric information collection and biometric verification 814 collects the patient's heart rate data (including heart rate measurements) through device 132. A biometric identifier 812 identifies and analyzes the heart rate data using a heart rate time-series recognition model. Its input is the user (user_ID) monitoring their heart rate using the device (device_ID) for a specific time period (e.g., 30 seconds or 6 hours during sleep) and the timestamp measurement. The output is the user's (user_ID) heart rate count per minute using the device (device_ID), along with the measurement timestamp and monitoring duration. For example: (1) user_ID: 98803311, device_ID: 001, heart rate per minute: 58, monitoring duration: 30 seconds, 01 / 01 / 2022 03:14; (2) user_ID:00313423, device_ID:002, heart rate per minute: 79, monitoring duration: 5 hours and 32 minutes, start timestamp: December 27, 2021, 23:20.

[0071] In some embodiments, device 132 may be a mobile phone, Apple Watch, or other digital device. Data may be stored in a database, such as... Figure 1 As shown. Then, a time-series-based deep neural network model analyzes the heart rate data to monitor recovery progress and symptom development.

[0072] Features and Functions: Electrocardiogram (ECG) Tracker and Pattern Recognition System Some patients with mental disorders may experience chest pain. According to certain embodiments, in biometric acquisition and biometric verification 814, the AI-powered psychotherapy system 100 can use an ECG tracker API to collect and monitor the patient's heart activity and rhythm, then store and display a wave-like pattern over a specific duration. The input is a user (user_ID) using device 132 (device_ID) to measure heart rhythm and activity with a timestamp over a certain duration (e.g., 15 minutes). This device 132 can be a mobile phone, Apple Watch, or other digital device. This machine-readable data is stored in... Figure 1The database shown.

[0073] In some embodiments, the AI-powered psychotherapy system 100 is equipped with an electrocardiogram (ECG) pattern recognition API in its biometric identifier 812. This ECG pattern recognizer detects specific signals associated with mental illnesses such as major depressive disorder and bipolar disorder. In some embodiments, the input may be a measured ECG, a scanned or uploaded photograph, or a stored routine ECG record (which may be up to 15 minutes long, depending on the embodiment), with a user ID, upload timestamp, and the initial measurement timestamp of the ECG. The ECG pattern recognizer, based on deep neural networks and statistical modeling, outputs visual detection results (e.g., positive, negative, location) of specific visual signals on the digital ECG image related to mental disorders, including but not limited to: • Low-frequency heart rate variability (LF-HRV) • High baseline respiratory sinus arrhythmia (RSA) QT interval prolongation • Increased QT dispersion T-wave repolarization • QRS wave (fQRS) splitting The output may be the recognition result (e.g., classification label) for each of the above visual signals, accompanied by a confidence score, probability, or significance value (p-value). For example, the processed timestamps and the initial measurement timestamps of the electrocardiogram are shown below: user_ID [e.g., user_ID: 98803311, LF-HRV-RSA – 98% (p < 0.001), fQRS – 34% (p > 0.05), QTIP – 10% (p > 0.1), IQTD – 76% (p < 0.01), TWR – 10% (p > 0.05), processing timestamp 23:1201 / 01 / 2022, initial timestamp 9:45 12 / 12 / 2021].

[0074] According to some embodiments of the artificial intelligence psychotherapy system 100, the output may be the input of an emotion and / or mood tracker and predictor 824.

[0075] Features and Functions: Patient Medical File Processor In some embodiments, patients can submit their medical records, laboratory test reports and results, and other information to the AI-powered psychotherapy system 100 as part of health data collection 808 via a reserved "data upload" button and interface. With the patient's permission, the AI-powered psychotherapy system 100 can use Optical Character Recognition (OCR) APIs to automatically collect and store biometric information (such as facial images and videos, heart rate) to establish a comprehensive mental health data collection. In some embodiments, the system and methods use OCR text recognition technology to read non-electronic health records, such as handwritten notes, scans, photographs, and other formats uploaded by the patient. Input can be user-identified documents, including text uploaded to the system, such as a two-dimensional photograph of a doctor's notes and test results taken by the patient's mobile phone camera with a timestamp. OCR technology is a deep neural network-based method that converts paper documents or images into a machine-understandable electronic version with associated user (user_ID), device ID, document type, content, institution name, doctor's name, timestamp of the record, any comments entered by the patient, and upload time, and stores it in the system's database.

[0076] The above example data can be found in the AI-powered psychotherapist system 100, as shown below: (1) user_ID: 98803311, device_ID: 002, upload timestamp 16:05 05 / 04 / 2021, record timestamp 13:02 02 / 17 / 2021, type: prescription drug, content: [sertraline 50mg (half a tablet daily for 2 weeks, 1 tablet daily)], patient comment: headache started on day 5, stopped taking the medication on my own after 1 month without informing the doctor; (2) user_ID: 00313423, device_ID: 003, upload timestamp 19:45 09 / 13 / 2021, record timestamp 10:51 08 / 15 / 2021, type: diagnosis report, content: [symptoms: depressive mood for 3 weeks, weight loss, insomnia, decreased attention, diagnosis result - major depressive disorder F33.1], patient comments: none.

[0077] The data storage provided by Health Data Collection 808 may help accurately track and predict each patient's treatment response, mental health, and other conditions. In some embodiments, the data and methods used to collect and store the data are secure and comply with relevant laws and regulations regarding data privacy, security, and protection.

[0078] Features and Functions: Health Record Recognition Device According to some embodiments, the AI-powered psychotherapy system 100 can use OCR text recognition technology in the health record reader 812 to read non-electronic health records, such as handwritten notes, scanned and photographed records, and other formats uploaded by the patient. The patient uploads data through a pre-defined "Upload Data" interface. The input to the health record reader is a non-electronic version of the health record (e.g., image format). The output is an electronic format, such as text, that can be read, edited, and processed by electronic devices (e.g., computers, smartphones).

[0079] • Alibaba Cloud OCR API, Huawei Cloud OCR API, Baidu OCR API, Swift OCR API, Tianruo OCR, QQ Screenshot Recognition, ABBY Fine Reader, Text Grabber, CamScanner, OnlineOCR, OneNote, XunJie PDF Converter, BaiMiao Software, Lightning Character Recognition, Wonder Share PDF Converter, Cajviewer, Google Cloud Vision API, NewOCR, i2OCR, Convertio, OCR .space, Amazon AWS Textract • Gated Recurrent Unit (GRU) models and their variants; Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM); Convolutional Neural Network (CNN) models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN); Recurrent Neural Network (RNN) models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF); Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN); RNN-based VAEs; RNN-based LSTMs; Semantic-conditioned LSTMs; Fast R-CNN, RRPN, TextBoes); DMP Net models and their variants; CTPN models and their variants; Seg Link models and their variants; FOTS framework variants; encoder / decoder model variants; MAML models and their variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants known in the art can be used to achieve similar or equivalent purposes to health record recognizers. • Bayesian probabilistic models and their variants, logistic regression models and their variants, decision tree models and their variants, support vector machine models and their variants, random forest models and their variants, ensemble models and their variants (including but not limited to bagging models, boosting models, stacking models, XGBoost models, Bayesian optimal classification models, Bayesian averaging models, Bayesian weighted average models), K-nearest neighbor models and their variants, and K-means models and their variants. • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to the health record recognizer 812. • Other specific and non-specific interfaces, software programs, cloud services, APIs, SDKs, etc., capable of performing similar or equivalent health record identification purposes. • Other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants can be used to achieve a target model and its variants similar to or equivalent to the health record recognizer 812. • Other specific artificial intelligence methods and their variants, non-specific artificial intelligence methods and their variants, or combinations of multiple artificial intelligence methods and their variants can be used to achieve models and variants that serve similar or equivalent purposes to the health record recognizer 812. According to some embodiments, the output of the health record recognizer 812 is the input of the emotion and / or mood tracker and prediction system 824.

[0080] Features and functions: Patient's current emotions / feelings collection system Type 1: According to some embodiments, the AI ​​psychotherapist system 100 provides different emotional descriptions in the user interface 102, ranging from extremely positive emotions to extremely negative emotions (e.g., regret, sadness, happiness, depression, etc.).

[0081] Users can choose the most appropriate word to describe their current emotion. The selected word will be sent to the sentiment classifier 822.

[0082] The input is a timestamped emoji selected by the user (user_ID) from the emoji list. The output is a numeric text label associated with the selected emoji, such as "happy," stored in the database corresponding to the user (user_ID), with a timestamp. For example, user_ID: 98803311, sentiment: happy, 02 / 01 / 2022 07:50 The output of the first type of sentiment classifier 822 is a possible input to the sentiment and / or emotion tracker and predictor 824, which will be described below.

[0083] Type 2: According to some embodiments, the system and method provide a placeholder on the user interface; the patient can enter phrases or sentences to describe their emotions. In some embodiments, these words or sentences can be indirect expressions of emotions, such as "I had insomnia all night" or "I'm very tired," or emotional descriptions, such as "I'm very depressed right now," or sentences related to suicide (attempt) or death, such as "When will I be able to leave this world?" (See above appendix) Figures 3A-3B and Figure 4 In some embodiments, the input is sent as patient message 818 to NLP processing method 400 and other machine learning and NLP models. Sentences and / or words may pass through sentiment and psychological named entity recognizer 820 (sometimes referred to as sentiment and psychological named entity recognizer 314) to output psychotherapy and psychological as well as mental disorders and psychological problems named entities 318 (sometimes referred to as psychotherapy, psychological, mental disorders and / or psychological problems named entities 418), and may be sent to psychologically related sentiment classifier 822. According to some embodiments, the output may be as follows: The SMS user (user_ID), message_ID, placeholder (e.g., "I am very frustrated"), and timestamp. For example: 60354067, 00098723, I am frustrated, 11 / 28 / 2021 19:20.

[0084] The output of the second type of sentiment classifier 822 is a possible input to the sentiment and psychological named entity recognizer 820.

[0085] Features and functions: Emotion and / or mood tracker and prediction system In some embodiments, the emotion and / or mood tracker and predictor 824 is a processor that runs machine learning, deep learning, or statistical algorithms for classification and prediction tasks. The emotion and / or mood tracker 824 serves as a long-term emotion tracker. In some embodiments, the emotion and / or mood tracker 824 analyzes and collects emotion-triggered events (events that cause a sharp shift in the patient's mood to positive or negative) and predicts specific or unspecific events or behaviors that may cause mood changes. The system and method may use Transformer-based algorithms and variations thereof to track and predict the patient's emotional events in a timely manner.

[0086] The processor acquires the output of the emotion and psychological named entity recognizer 820, namely the patient's historical emotion recognition on a specific or non-specific timeline, such as the following: “3 / 3 / 2020 9:32PM Sadness”, “3 / 4 / 12:32 PM Frustration”, and the emotional triggering event: arguing with mother.

[0087] In some embodiments, the output is a patient record with a timeline, emotion, and cause, as well as any specific or non-specific regularity that leads to mood changes, such as "Several times, your mood improved after eating ice cream." The output is a prediction of treatment progress with a timeline and / or duration (e.g., after / each), mood trends, triggering events (i.e., specific or non-specific regularities that lead to mood changes), and predictions of treatment progress. Treatment progress refers to the current stage / episode of the corresponding mental disorder in the patient, depending on the symptoms (e.g., mild, moderate, severe, peak, inactive). Treatment progress determines how long treatment will last and future treatment schedules. For example, (1) user_ID: 7822034, mood trend: improving, triggering event: ice cream, timeline: late stage, stage: mild; (2) user_ID: 98803311, mood trend: negative, triggering event: wife, timeline: end of month, stage: peak; (3) user_ID: 00283118, mood trend: depressed, triggering event: cloudy / rainy, timeline: winter, stage: moderate.

[0088] In some embodiments, the emotion and / or mood tracker 824 may generate patient psychotherapy records as follows and store them in the data storage management system 122: [Patient ID, Emotional Expression, Emotional Analysis (if applicable), Emotional Type – Before and After (if applicable), Cause, Cognitive Distortion Classification, Psychotherapy, Patient Feedback] Figure 9 An example of a user interface 102 according to some embodiments is shown. In this example, the user interface includes various areas for chat, therapy and emotions, and a target tracker. Figure 10 Another example of a user interface 102 according to some embodiments is shown. This example displays a mood and goal tracker, as well as an overview over a certain period of time (e.g., multiple months).

[0089] In some embodiments, similar functions to generating emotion and / or sentiment trackers and prediction systems 824 can be implemented through one or more of the following algorithms and / or models: • Markov models and their variants, Gated Recurrent Unit (GRU) models and their variants, Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM), Convolutional Neural Network (CNN) models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN), Recurrent Neural Network (RNN) models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF), Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN), Encoder / Decoder model variants, MAML models and their variants • Specific and non-specific time series methods, tools, and their variants that can achieve similar or equivalent purposes, including but not limited to linear regression, cyclic factor methods, ARMA-based models, AR IMA-based models, Facebookprophet, tslearn, tsfresh, TPA-LSTM, LSTNet, and STH. • Model series and variants based on R AI nbow, model series and variants based on Apex, models based on A3C and A2C, updated versions and variants, model series and variants based on Q-learning, policy optimization methods and variants, World Models model series and variants, i2A model series and variants, MBMF model series and variants, MBVE model and variants, AlphaZero model series and variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants can be used to achieve similar or equivalent purposes for emotion and / or sentiment trackers and prediction systems. • Other specific reinforcement learning models and their variants, non-specific reinforcement learning models and their variants, or combinations of multiple reinforcement learning models and their variants can be used to achieve models and variants that serve similar or equivalent purposes to emotion and / or sentiment trackers and prediction systems. • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to emotion and / or sentiment trackers and prediction systems 824. • Models and their variants that serve similar or equivalent purposes to emotion and / or sentiment trackers and prediction systems can be achieved using other specific artificial intelligence methods and their variants, non-specific artificial intelligence methods and their variants, or combinations of multiple artificial intelligence methods and their variants. • Other specific statistical models and their variants, non-specific statistical models and their variants, or combinations of multiple statistical models and their variants can be used to achieve similar or equivalent purposes to sentiment and / or emotion trackers and prediction systems. In some embodiments, the system and method train model parameters from scratch using self-collected data and public datasets. To achieve a similar or equivalent purpose to training model parameters, pre-trained models, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods, including but not limited to those described above, may be used.

[0090] Features and Functions: Emotional and Psychological Named Entity Recognizer System According to some embodiments, the emotion and psychological terminology entity recognizer 820 in the AI-powered psychotherapy system 100 derives the core emotion of a given sentence or paragraph based on a series of transformer models and their variants. In some embodiments, a patient message 818 may contain a few sentences or words of text written by the patient (user_ID) about their emotions, with a timestamp. This text message is received by a patient current emotion / mood collector 816. The text can be an indirect or direct expression of emotion, or a longer text (such as a diary entry), or a statement about thoughts of suicide and death. The output is the emotion words extracted from the patient message, such as "excitement," "self-esteem," or "suicide."

[0091] The output is a machine-readable sentiment term entity (e.g., “depressed”) and a corresponding user (user_ID), message_ID, and timestamp. For example: (1) user_ID: 98803311, message_ID: 00000103, depressed, November 28, 2021, 19:20; (2) user_ID: 36036844, insomnia, October 31, 2021, 04:11.

[0092] In some embodiments, functionality similar to the emotion and psychology terminology entity recognizer 820 can be achieved through one or more of the following algorithms and / or models: • Markov models and their variants, Gated Recurrent Unit (GRU) models and their variants, based on Transformer models and their variants (including but not limited to Transformer, GPT, BERT, XLM), Convolutional Neural Network (CNN) models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN), Recurrent Neural Network (RNN) models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTMSeq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF), attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN), other specific or non-specific transfer learning methods and their variants, encoder / decoder-based models and their variants, and MAML-based models and their variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants can be used to achieve similar or equivalent purposes to the Emotion and Psychology Term Entity Recognizer 820. • Bayesian probabilistic models and their variants, logistic regression models and their variants, decision tree models and their variants, support vector machine models and their variants, random forest models and their variants, ensemble models and their variants (including but not limited to bagging models, boosting models, stacking models, XGBoost models, Bayesian optimal classifier models, Bayesian averaging models, and Bayesian weighted average models). • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to the Emotion and Psychology Term Entity Recognizer 820. • Other specific statistical models and their variants, non-specific statistical models and their variants, or combinations of multiple statistical models and their variants can be used to achieve similar or equivalent purposes to the Emotion and Psychology Term Entity Recognizer 820. • Other specific artificial intelligence methods and their variants, non-specific artificial intelligence methods and their variants, or combinations of multiple artificial intelligence methods and their variants can be used to achieve models and variants that serve similar or equivalent purposes to the Emotion and Psychology Term Entity Recognizer 820. • Other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants, can be used to achieve models and variants that serve similar or equivalent purposes to the Emotion and Psychology Term Entity Recognizer 820.

[0093] In some embodiments, the AI-powered psychotherapy system 100 trains model parameters from scratch using its own collected data and public datasets. To achieve a similarity or equivalence to the trained model parameters, the AI-powered psychotherapy system 800 may use pre-trained models, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods, including but not limited to the models described above.

[0094] In some embodiments, the inputs or outputs of the AI-powered psychotherapy system 100 may include suicide (i.e., attempted suicide) and / or other related content. In such cases, the system and method may (a) trigger an alert, or (b) quickly provide the user with relevant suicide prevention manuals or emergency calls. In some embodiments, the AI-powered psychotherapy system 100 may contact emergency contacts and / or initiate other emergency procedures.

[0095] Features and Functions: Psychology-Related Emotion Classifier System According to some embodiments, the psycho-related sentiment classifier 822 can be a processor configured to perform sentence-level and document-level psycho-related sentiment classification tasks by running machine learning, deep learning, or statistical algorithms, and to perform fine-grained classification based on the number of categories of key sentiment features and specific mental disorders. Inputs might look like this: user_ID, message_ID, timestamp, received from (1) the emotion / feeling collector 816 and (2) the psychotherapy (e.g., “writing a letter”, “defense lawyer”) patient input of sentence type or article type.

[0096] In some embodiments, the AI-based psychotherapy system 100 can use a series of Transformer-based models and their variants to classify emotions under different mental disorders. For example, the psychologically relevant emotion classifier 822 can perform fine-grained classification of depressive moods and similar emotions of different types and degrees in major depressive disorder. Patients with depressive disorders typically experience a range of common depressive moods, such as "frustration," "numbness," and "despair," which are categorized according to their intensity. The output is as follows: User ID, message ID, timestamp, and multiple fine-grained sentiment tags. For example, user ID: 98803311, message ID: 00000023, 02 / 16 / 2021 00:35, despair, suicidal ideation, pain.

[0097] According to some embodiments, the output of the psychologically relevant emotion classifier 822 may be the input of the emotion and / or mood tracker and predictor 824.

[0098] In some embodiments, to achieve similar functionality using a psychology-related emotion classifier 822, the following algorithm can be used: This includes: Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM); Convolutional Neural Network (CNN)-based models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN); Recurrent Neural Network (RNN)-based models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF); Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN); other specific or non-specific transfer learning methods and their variants; encoder / decoder-based models and their variants; and MAML-based models and their variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants can be used to achieve similar or equivalent purposes to the psychology-related sentiment classifier 822. • Bayesian probabilistic models and their variants, logistic regression models and their variants, decision tree models and their variants, support vector machine models and their variants, random forest models and their variants, ensemble models and their variants (including but not limited to bagging models, boosting models, stacking models, XGBoost models, Bayesian optimal classifier models, Bayesian averaging models, and Bayesian weighted average models). • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to the psychology-related sentiment classifier 822. • Other specific artificial intelligence models and their variants, non-specific artificial intelligence models and their variants, or combinations of multiple artificial intelligence models and their variants can be used to achieve similar or equivalent purposes to the psychology-related sentiment classifier 822. • Other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants, can be used to achieve models and variants that serve similar or equivalent purposes to the psychology-related sentiment classifier 822. In some embodiments, the AI-powered psychotherapy system 100 can train model parameters from scratch on self-collected data and public datasets. To achieve similar or equivalent training of model parameters, methods include, but are not limited to, using pre-trained models of the aforementioned models, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods.

[0099] Characteristics and functions: Mining intentions in psychology and psychotherapy In some embodiments, in order to achieve the mining of psychological and psychotherapeutic intentions (i.e., classification of psychological and psychotherapeutic intentions), patient messages 302 (sometimes referred to as patient messages 818) are input into a psychological and psychotherapeutic intention classifier and / or an NLP intention recognizer and / or a topic classifier and / or a sentiment analysis model 312 (sometimes referred to as a psychological and psychotherapeutic intention classifier and / or an NLP intention recognizer and / or a topic classifier and / or a sentiment analysis model 870), and then a classification—psychological and psychotherapeutic intention 322—is generated, which can be used as input to a cognitive reconstruction generator 348.

[0100] As shown in Figure 8, in some embodiments, intent mining is performed using a psychological and psychotherapy intent classifier and / or an NLP intent recognizer and / or a topic classifier and / or a sentiment analysis model 312 to uncover the behaviors and goals behind patient messages, such as suicide (attempt), seeking medical treatment, seeking help, needing calm, etc. These behaviors and goals are categorized into different degrees of positive and negative types, as well as whether there is a life-threatening situation. The inputs are patient messages, intent history, user profiles, dictionaries and templates from different domains, etc., and the output is psychological and psychotherapy intent 320.

[0101] For example, input: user_ID: 45678023, message_ID: 00000040, “All my classmates are jealous of me! Because I am the best student council president ever!”, intention history: [Seeking attention], personality: histrionic personality, mental disorder type: bipolar disorder. Output is intention, user_ID, message_ID, intent_ID, intention tag. For example, user_ID: 45678023, message_ID: 00000040, intent_ID: 00000002, intention tag: [Seeking attention, narcissism, self-elevation, lack of humility, self-focus].

[0102] In some embodiments, the AI-based psychotherapy system 100 may use a transformer-based algorithm model and its variants for intention mining. According to some embodiments, to achieve similarity or equivalence to intention classifiers and / or NLP intention identifiers and / or topic classifiers and / or sentiment analysis models 312 used in psychology and psychotherapy, the following algorithms and models may also be used: This includes: Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM); Convolutional Neural Network (CNN)-based models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN); Recurrent Neural Network (RNN)-based models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF); Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN); other specific or non-specific transfer learning methods and their variants; FastText models and their variants; Encoder / decoder-based models and their variants; and MAML-based models and their variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants can be used to achieve similar or equivalent purposes to intention mining in psychology and psychotherapy. • Bayesian probabilistic models and their variants, logistic regression models and their variants, decision tree models and their variants, support vector machine models and their variants, random forest models and their variants, ensemble models and their variants (including but not limited to bagging models, boosting models, stacking models, XGBoost models, Bayesian optimal classifier models, Bayesian averaging models, Bayesian weighted average models), K-nearest neighbor models and their variants, K-means models and their variants • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to intention mining in psychology and psychotherapy. • Models and their variations that can achieve similar or equivalent purposes related to intention mining in psychology and psychotherapy, using other specific statistical models and their variations, non-specific statistical models and their variations, or combinations of multiple statistical models and their variations. • Other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants, can be used to achieve models and variants that serve similar or equivalent purposes to intention mining in psychology and psychotherapy. • Other specific artificial intelligence models and their variants, non-specific artificial intelligence models and their variants, or combinations of multiple artificial intelligence models and their variants can be used to achieve similar or equivalent purposes to intention mining in psychology and psychotherapy. In some embodiments, the AI-powered psychotherapy system 800 can train model parameters from scratch on its own collected data and public datasets. To achieve similar or equivalent training of model parameters, methods include, but are not limited to, using pre-trained models, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods. If the input or output contains suicidal (intent) or related content, the system and method will initiate an alert, or quickly push relevant suicide intervention, or make an emergency call, or contact emergency contacts, or other emergency procedures.

[0103] Features and Functions: Automatic Cognitive Distortion Classifier System According to some embodiments, the artificial intelligence psychotherapy system 100 can determine the type of cognitive distortion present in the patient's input using an automatic cognitive distortion classifier 826. This cognitive distortion classifier 826 can be implemented by one or more classifiers, which may include NLP classifiers, machine learning classifiers, classifiers using statistical methods, and / or classifiers using other artificial intelligence methods. The determined classification can be any of the following labels (i.e., a list of cognitive distortions): The idea of ​​either having or not having Overgeneralization Psychological filtering Negating positive factors • Leap of conclusion Exaggeration or understatement • Emotional reasoning • Use statements that require or should Gratitude Trap · Attribution and self-blame Always right • The fallacy of change Fairness fallacy • Tags and error tags In some embodiments, the AI-powered psychotherapy system 100 uses a Transformer-based model and its variants in an automated cognitive distortion classifier 826. The input can be a patient message 818, such as “It’s all my fault,” and the output can be a cognitive distortion classification (i.e., a cognitive distortion classification label 828), such as “all-or-nothing thinking” and “attribution and self-blame.” The input might resemble the following: User ID, message ID, timestamp, patient's message, such as "It's all my fault". For example: user_ID: 45678023, message_ID: 00000040, 13:56 04 / 01 / 2021, "It's all my fault".

[0104] The output consists of user_ID, message_ID, cognitive distortion_ID (CD_ID), message timestamp, and a category label for cognitive distortion, such as "all-or-nothing thinking" and "egotization and self-blame." For example, user_ID: 45678023, message_ID: 00000040, CD_ID: 00000561, egotization and self-blame, 13:56 04 / 01 / 2021.

[0105] According to some embodiments, the AI-powered psychotherapy system 800 can respond to patients based on the above classification; and can use the psychotherapy generator 348 to generate personalized psychotherapy 330, in formats including but not limited to videos (e.g., 1- to 10-minute cartoons), images, emojis, text, GIFs, music, and other types. Cognitive distortion classification labels 828 (i.e., outputs) can be sent as inputs to the psychotherapy generator 348.

[0106] In some embodiments, to achieve functionality similar to the Automatic Cognitive Distortion Classifier 826, the following algorithms and models can be used: This includes: Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM); Convolutional Neural Network (CNN)-based models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN); Recurrent Neural Network (RNN)-based models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF); Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN); other specific or non-specific transfer learning methods and their variants; encoder / decoder-based models and their variants; and MAML-based models and their variants. • Other specific deep learning models and their variants, non-specific deep learning models and their variants, or combinations of multiple deep learning models and their variants can be used to achieve similar or equivalent purposes to the Automatic Cognitive Distortion Classifier 826. • Bayesian probabilistic models and their variants, logistic regression models and their variants, decision tree models and their variants, support vector machine models and their variants, random forest models and their variants, ensemble models and their variants (including but not limited to bagging models, boosting models, stacking models, XGBoost models, Bayesian optimal classifier models, Bayesian averaging models, and Bayesian weighted average models). • Other specific machine learning models and their variants, non-specific machine learning models and their variants, or combinations of multiple machine learning models and their variants can be used to achieve similar or equivalent purposes to the Automatic Cognitive Distortion Classifier 826. • Other specific statistical models and their variants, non-specific statistical models and their variants, or combinations of multiple statistical models and their variants can be used to achieve similar or equivalent purposes to the Automatic Cognitive Distortion Classifier 826. • Other specific artificial intelligence models and their variants, non-specific artificial intelligence models and their variants, or combinations of multiple artificial intelligence models and their variants can be used to achieve similar or equivalent purposes to the Automatic Cognitive Distortion Classifier 826. • Other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants, can be used to achieve models and variants that serve similar or equivalent purposes to the Automatic Cognitive Distortion Classifier 826. In some embodiments, the AI-powered psychotherapy system 100 can train model parameters from scratch on self-collected data and public datasets. To achieve similar or equivalent training of model parameters, methods include, but are not limited to, using pre-trained models of the aforementioned models, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods.

[0107] Features and functions: Psychotherapy selector and cognitive restructuring generator In some embodiments, after importing psychological and psychotherapy intentions 320, and / or named entities 318, and / or context 322, and / or a knowledge graph 344 of psychotherapy and mental disorders and psychological problems, and / or patient messages 302, an automatic cognitive distortion classifier 346 (sometimes referred to as an automatic cognitive distortion classifier 826) classifies cognitive distortions and outputs a cognitive distortion classification label 828, which is input to a psychotherapy selector 328. The selector 328 applies state-of-the-art deep learning NLP algorithms to select a psychotherapy 326 from a list 862 of 78 cognitive techniques, with or without NLP preprocessing method 400.

[0108] There are two types of cognitive skills: (i) static cognitive skills, which are interactive with the user without personalization, so all users experience the same interaction, such as writing a letter; and (ii) dynamic cognitive skills, which are personalized, so each user is assigned the same type of cognitive skill, but the experience for different users depends on previous conversations. For example, terminology definition (asking user A, “How do you define failure?”, asking user B, “How do you define independence?”). Static cognitive skills are selected by a psychotherapy selector 328 from a list 862 of 78 cognitive skills to obtain a selected psychotherapy 326 (sometimes referred to as a selected psychotherapy 864). Dynamic cognitive skills are first selected by the psychotherapy selector 328 to obtain a selected psychotherapy 326, and then the selected psychotherapy 326 is personalized using a cognitive reconstruction generator 348. The psychotherapy selector and cognitive reconstruction generator system 830 includes (i) the psychotherapy selector 328 and (ii) the cognitive reconstruction generator 348. According to some embodiments, the psychotherapy selector and cognitive reconstruction generator system 830 may be one or more NLP processors with machine learning, deep learning, or statistical algorithms.

[0109] In some embodiments, the AI ​​psychotherapist system 100 utilizes a cognitive reconstruction generator 348 (sometimes also referred to as a cognitive reconstruction generator 868) to synthesize the patient’s context and content 206, psychotherapy methods, treatment methods for specific mental disorders (such as major depressive disorder), and selected psychotherapy 864 to generate personalized, professional, and effective psychotherapy for cognitive reconstruction—personalized psychotherapy 330 (sometimes also referred to as personalized psychotherapy 832).

[0110] In some embodiments, a psychotherapy selector 328 (sometimes referred to as psychotherapy selector 860) can select from a list of cognitive skills 862 and output the selected psychotherapy 326 via a metric-based neural network NLP algorithm or a content-based recommendation engine, ranking and scoring the candidate responses and selecting the optimal candidate (typically the highest-scoring one). The inputs are user A's patient message, the list of cognitive skills, user A's feedback on each cognitive skill, and / or other users' feedback on each cognitive skill. Various implementations include collaborative filtering and its variants, as well as a hybrid of content-based filtering and collaborative filtering and its variants.

[0111] In some embodiments, the AI-powered psychotherapy system 100 can be used for these purposes using text representation, semantic analysis, and / or text mining.

[0112] According to certain embodiments, the AI ​​psychotherapist system 100 can generate psychotherapy (i.e., selected psychotherapy 864 and / or personalized psychotherapy 832) based on the aforementioned dialogue history, as executed in block 834, as follows: Example: The input to selector 328 is a dictionary data structure containing the user ID, message ID, CD_ID, cognitive distortion category label, message timestamp, and user feedback history (positive experiences) for various cognitive techniques. For example, User_ID: 45678023, Message_ID: 00000040, CD_ID: 00000561, attribution and self-blame, 13:56 04 / 04 / 01 / 2021, feedback history: [3.13.0.3: 50%, 3.13.0.21: 30%, 3.13.13.0.65: 85: 855: 855: 855: 855%. The output is user_ID, CD_ID, psychotherapy execution timestamp, selected cognitive restructuring method_ID (CT_ID), and user feedback. For example, user_ID: 45678023, CD_ID: 00000561, CT_ID: 3.13.048.

[0113] The input to generator 348 is user_ID, message_ID, CD_ID, cognitive distortion classification label, message timestamp, selector output, and short-term conversation history. For example, user_ID: 45678023, message_ID: 00000040, CD_ID: 00000561, CT_ID: 3.13.048, short-term conversation history: list of message_IDs [000037, 000038, 000039]. In some embodiments, after performing personalized therapy 832 or selected therapy 864, the patient's feedback 836 is solicited in a binary manner, positive and negative, by sending the option of "feeling good" 866 or "not so good" 838 to the user interface 102. Therefore, the average positive feedback score for the selected psychotherapy is returned to the user feedback dictionary. The 78 cognitive techniques 862 used by static and dynamic cognitive techniques are discussed below. Each cognitive technique forms the basis of a personalized cognitive technique by defining the main structure of the dialogue or method in the user interaction. If selector 328 selects a static cognitive technique from list 862, that static cognitive technique is directly output to the user without needing to be personalized for the performance. If selector 328 selects a dynamic cognitive technique from list 862, generator 348 takes the selected cognitive technique as input and refers to the main idea or structure of the cognitive technique to generate a personalized cognitive technique, which is then output to the user. Sentences in parentheses are either examples of specific cognitive techniques, indicated by "for example," or descriptions of the corresponding cognitive techniques. The list 862 of 78 cognitive techniques is summarized below: • Examine changes / degrees in behavior (e.g., help users track how their behavior changes and to what extent over time, place, conditions, and people). • Weigh the evidence supporting / opposing an idea (e.g., a user enters the message "I will be rejected," and is asked to write evidence supporting this idea: ["Someone doesn't like me"], and evidence opposing this idea: ["I am a good person" / "People go to parties to make new friends."], and is asked to calculate the percentage of support and the percentage of opposition, and then write a new conclusion: [There is not enough evidence to prove that I will definitely be rejected, so it's okay.]) • Imagine the patient is their own lawyer defending them (e.g., tell the user, "Imagine you hiring yourself as your own lawyer to defend yourself and write the explanation that is most compelling to you, even if you don't believe it.") • Examine internal contradictions (e.g., check if users believe, "I want everyone on Facebook to follow me, even if they don't know me."). • Reductio ad absurdum (e.g., detecting whether a user believes, "I'm divorced, therefore I'm not worthy of being loved.") Continuous media technology • More adaptive assumptions about costs and benefits (e.g., a user enters the message "Everything I want to do requires my mom's approval." Ask the user to write down the costs and benefits of doing so, cost: ["I lose the opportunity to try things my mom doesn't like, but I'm interested in."], benefit: ["I won't be called in for a talk for breaking the rules"], ask the user to calculate cost [%] versus benefit: [%]) • Differentiate between progress and perfection (guide users to explore and compare the benefits of progress and improvement, rather than striving for perfection.) • Borrow perspectives from others (e.g., ask users, "Don't keep getting stuck in your own way of thinking. Think about someone you admire and imagine how they would react to this?") • Terminology definition (For example, the system might ask the user to define a term by sending the message "How do you define failure?"). • Check the testability of the idea (check if the user's idea is an objective observation of the real world, and whether it can be verified or refuted by others?) • Examine the logic of the thinking (check if cognitive distortions are marked as "leap conclusions") • Examine user information limitations (using statistical modeling to check for bias in the evidence users provide to support their viewpoints). • Vertical progression (e.g., asking the user, "Based on your description of..., what will happen next?") • Double standards (e.g., asking users, "Do you think everyone in the world does XXX because of XXX?") • Challenge recursive self-criticism and self-doubt (e.g., detect if a user is trapped in a cycle such as "I am a failure, therefore I lost my job. I lost my job, therefore I am a failure."). • Differentiate between behavior and individual (e.g., detect whether a user believes, "I failed my final exam, so I'm a bad student.") • Make the challenge concrete [e.g., ask users to translate abstract or unobservable conclusions (e.g., worthless) into concrete expectations (e.g., I would prefer my boss to like my report)]. Self-reward mechanism • Scenario comparison (e.g., asking users to compare their "divorced" situation with their "orphan" situation). Write a letter to someone • Allocate responsibility using a pie chart (users are required to allocate other responsible parties and their percentage of responsibility in a given pie chart). • Examine mitigating factors and reattribute the cause (e.g., analyze whether the user uses medication, alcohol, etc.). • Externalize each other's thoughts through role-playing (the user is asked to play the role of "agreeing," and the system plays the role of "disagreeing." For example, the system says: ["You're divorced."], and the user replies: ["Maybe my next boyfriend will be a prince, like Meghan Markle."]) • Hypothesis thinking (guiding users to role-play as those who do not believe their current negative thoughts). • Challenge absolutist thinking (for example, ask users, "If you think you'll fail this exam, have you failed every exam you've ever taken before?") • Set a baseline for comparison to avoid extreme comparisons (when users compare themselves to the best, ask them: "How do you see yourself when compared to the worst? And how do you see yourself compared to the average person in the group?") • Positive restructuring (finding positive factors in negative situations) • De-catastrophize (e.g., ask the user, "What if {X} wasn't scary?") • Examine “frightening scenarios” (e.g., ask the user: “If {X} happens, what is the worst-case scenario? How would you handle it?”). • Predict future reactions (ask users, "Will {X} still be important to you or others after a certain month / year / quarter?") • Examine past predictions, failure to learn from false predictions, and self-fulfilling prophecies (ask the user: "Can you recall any negative predictions in the past that did not come true?"). • Evaluation criteria for patients (e.g., asking users: "What's on your daily to-do list?" "How much can you accomplish?" "Are your goals unrealistic?" "Have you considered the learning curve?") • Examine the patient's value system (ask the user, "What is your value system?" and guide the user to complete the value pyramid by inputting values ​​at each level.) • Examine societal standards (ask users: "Are you trying too hard to conform to societal standards? For example, men should be rich, and women should have families") • Question idealized assessments of others (e.g., assign users the task of asking whether someone they admire has ever made a mistake). Promote adaptability and flexibility (e.g., send users short videos explaining the benefits of greater flexibility). • Problem and solution review (e.g., ask the user: "What problems have you encountered recently? What solutions did you use?") • Emphasize curiosity, challenge, and growth, rather than the pursuit of perfection (for example, send users a short video explaining the relationship between curiosity, challenge, and success, and how to cultivate curiosity and a spirit of daring to take on challenges). • Imagination and Emotion [Play calming music and ask the user: "Now close your eyes, evoke your negative feelings (such as chest pain), and imagine a visual image related to your negative feelings, then complete this sentence—this picture makes me uncomfortable because it reminds me of…."] • Response Notifications (Send users response notifications with calming music, funny animal videos, or positive quotes.) • Minimize the frightening image (guide users to imagine the person or situation they are afraid of as someone much weaker than the patient). • Desensitize images (e.g., ask users to interact with people or situations they fear for varying durations). • Fostering self-statement (instructing the user: "Now imagine you are just a little child, and describe the kind of warm words you wish you had said to yourself.") • "Rights List" (requires users to list their rights, such as "the right to cry" or "the right to self-compassion") • Identify the problem (e.g., ask the user, "What problem needs to be solved? What is the problem?"). • Embrace questions and uncertainty (e.g., ask users to "recall any previous lottery winning experiences" and send them a short video about uncertainty). • Review the goals and generate alternative goals (ask the user: "What are your goals in the current situation? Set small goals and achieve them step by step, rather than striving for perfection"). • Steps to prevent procrastination (Ask the user to respond with: "Set your goal," "Can you break it down into smaller steps?", "What are the costs and benefits of the first step to achieve the goal and alternatives?", "Choose a time, place, and duration to perform this activity," and "How did you do?" Follow up with the user again at the selected time.) • Self-correction (e.g., sending users a short video about how to learn from mistakes and stop self-criticism) • Establish treatment contracts (e.g., requiring users to set specific times each day to communicate with the system and sending timely mobile notifications to remind them). • Bibliotherapy (assigning books to users, including titles and online / free links) • Explain how thoughts generate emotions (e.g., send users a short video about the connection between thoughts and emotions). • Distinguish between ideas and facts (e.g., ask users to gather evidence to support their ideas). • Delay making decisions (e.g., ask users to set a reasonable timeframe during which they will not make a decision, such as taking a good night's sleep before considering it). • Predict problems (e.g., ask users to "list all the problems you think will happen and your solutions.") • Assess confidence in the accuracy of the idea and the intensity of the emotion (e.g., “I feel shy [%] because I think no one will talk to me [%].”) • Examine the quality of the evidence (e.g., check whether the cognitive distortion label includes "affective reasoning") • Alternative explanations (e.g., suggesting to the user: "Perhaps others aren't talking to you because they don't know how to approach you. Or, perhaps there are more people who want to talk to me"). • Negative questions (requiring users to "list all cases where the current question is not a question") • Conduct experiments (encourage users to test or challenge their ideas; for example, user: "I will be rejected.", system: "Try to approach others at the party.") • Test predictions (requires users to "create a specific list of predictions for the next three days," and then continuously track the results by sending reminders and notifications.) • Examine past worries (for example, ask users to "recall and write down some things you used to worry about but no longer care about, list as many as possible," and after the user has entered their thoughts, ask them "Why are these things no longer important to you now?"). • Examine future distractions (e.g., ask users, "What unrelated events might occur in the next day, week, month, or even year that would make you less concerned about the current situation?"). • Differentiate between possibility and probability (e.g., ask users, "If you continue to be anxious, it could trigger a heart attack, but what is the probability?") • Calculate continuous probabilities (requires the user to list several events, assign a separate probability to each event, and then multiply them together). • Correct overgeneralization (e.g., ask the user, "Just because it happened once, does it mean it will inevitably happen again?") • Challenge the need for certainty (e.g., sending users a short video explaining that the world is full of uncertainty, and the benefits that uncertainty brings). • Promote acceptance (send users a short video about learning to accept). • Using the "positive and negative comparison method" to handle difficult ideas • Re-examine the initial negative thoughts and emotions, confidence in the accuracy of the thoughts, and the intensity of the emotions (redo this after the patient undergoes actual testing or other treatment). • Identify the contents of the patient's "rulebook"—statements such as "should," "must," and "if-then," which form the basis of distorted automatic thinking (once these word entities are detected, store the relevant messages in a database and analyze them). • Write self-directed guidance statements; create "coping cards" • Explain pattern processing (send short videos explaining how abnormal patterns form and how they systematically influence how people pay attention to and react to events.) • Strategies for determining avoidance / compensation patterns • Activate early memories to determine the source of the pattern (e.g., ask the user, "Who made you think in this abnormal way?" "Do you think their teachings are effective? Or are they bad role models?") • Source of the challenge mode through role-playing • Imagine reconstructing and rewriting your life script (for example, asking users to re-examine their negative life script and rewrite it in their journal to produce positive results). According to some embodiments, the AI-powered psychotherapy system 100 can use a family of Transformer-based algorithm models and their variants to generate personalized psychotherapy 832 for different emotional disorders. The process takes as input emotional expression, psychological and psychotherapy intentions 320, context and content 206, cognitive distortion classification, mental disorder type, psychotherapy, and treatment methods for one or more specific mental disorders (such as major depressive disorder). It can select psychotherapy and / or generate personalized psychotherapy 832 as output to provide support to patients in distress, such as the following: User ID: 98803311, Message ID: 00003023, 03:09 02 / 26 / 2021, Selected Psychotherapy 864: Differentiating Between Individual and Behavior, Personalized Psychotherapy 832: "I can sense that you're struggling with swimming, and because all the other students are learning so quickly, you feel like you're stupid. Anna, learning slower than others just means your learning methods are different, it doesn't mean you're a stupid student. You have unique learning methods that no one else has!" In some embodiments, in order for the task-oriented dialogue systems 204a to 204h to achieve similar functionality to the cognitive reconstruction generator 348, the following algorithms can be used: • Transformer-based models and their variants (including but not limited to Transformer, GPT, BERT, XLM); Convolutional Neural Network (CNN) models and their variants (including but not limited to TextCNN, DCNN, VDCNN, GCNN, TCN); Recurrent Neural Network (RNN) models and their variants (including but not limited to RNN, LSTM, GRU, BiLSTM, RCNN, RNN / LSTM Seq2Seq, Tree-based LSTM, Stack LSTM, RNN-based VAE, LSTM-CNN-based GAN, DMN, BiLSTM-CRF); Attention-based models and their variants (including but not limited to AT Seq 2Seq, ATAE-ST, ABCNN); Contextual NLG algorithms and their variants; Generative Adversarial Network (GAN) models and their variants (including but not limited to Fictitious GAN, Capsule...). GAN, treeGAN), Plan-based NLG algorithms and their variants, Class-based LM algorithms and their variants, Phrase-based algorithms and their variants, Corpus-based generation methods and their variants, Structural NLG algorithms and their variants, Controlled text generation algorithms and their variants, Transfer learning methods and their variants, Laser Tagger models and their variants, XLNet models and their variants, Pegasus series models and their variants, ULMFiT series models and their variants, Megatron series models and their variants, Variational Autoencoder (VAE) series models and their variants, Encoder / Decoder series models and their variants, MAML series models and their variants • Task-oriented dialogue systems can utilize specific or non-specific deep learning models and their variants, specific or non-specific machine learning models and their variants, specific or non-specific statistical models and their variants, other specific artificial intelligence methods and their variants, or any combination of multiple algorithms and their variants, to achieve models and variants that serve similar or equivalent purposes to those used in psychotherapy. • Through task-oriented dialogue systems, other specific meta-learning and few-shot learning models and their variants, non-specific meta-learning and few-shot learning models and their variants, or combinations of multiple meta-learning and few-shot learning models and their variants, can be used to achieve models and variants that serve similar or equivalent purposes to those used in psychotherapy. In some embodiments, the system and method can train model parameters from scratch on self-collected data and public datasets. To achieve a similar or equivalent purpose in training model parameters, methods include, but are not limited to, using pre-trained models of the models described above, fine-tuning, meta-training, few-shot training, or other specific and unspecified training methods.

[0114] Examples of cognitive skills This section describes several cognitive techniques from List 862 based on some embodiments of an AI psychotherapist 800. The AI ​​psychotherapist 800 will select the optimal cognitive technique based on the patient's condition (e.g., type and severity of mental disorder, type of cognitive distortion, treatment recommendations, and user feedback).

[0115] Features and functions: Self-reward mechanism techniques In some embodiments, the system may require the patient (user) to set a reward (e.g., eating ice cream or playing video games for 30 minutes) to reward themselves after achieving a specific goal or experiencing an unpleasant event. The system may require the user to type the reward in a dialog user interface and choose between "goal to be achieved" or "negative emotional event." If the user selects "negative emotional event," the self-reward mechanism will execute immediately, reminding the user to claim the reward by sending the message "Now it's time for ice cream!" If the user selects "goal to be achieved," the AI ​​psychotherapy system 800 may: 1) Wait for the user to manually click "Done" and send the message "Don't forget your ice cream"; or 2) Send notification 854 to remind the user that the goal has been achieved and to claim the reward after waiting for a certain period of time.

[0116] Features and Functions: Techniques for Writing a Letter to Someone In some embodiments, the AI-powered psychotherapy system 100 may select this cognitive technique when it detects that a user frequently mentions or has conflicts with someone. The system will suggest that the user write a letter within the system, suggesting something like, "I've noticed you're struggling with XXX. If they were willing to listen to your thoughts, what would you like to say?" The system may guide the user to a page containing a placeholder for writing the letter, which will be private. A language model will analyze the letter and track changes in the user's mood or warning signs.

[0117] Characteristics and functions: Psychological education techniques In some embodiments, the system educates patients on rational thinking through animated short films in multiple languages. For example, “When things happen, sometimes we beat ourselves up with negative, habitual thoughts. Pay attention to using precise language to defuse the negative emotions in these thoughts. When you think, ‘I feel so bad today, I’m a failure,’ you are actually generalizing a short-lived emotion into a long-term definition of your self-identity. Think about it! When everyone is feeling bad, do you still conclude that they are all failures? This is not to deny your emotions, but to pay attention to the words you choose to infer your feelings. This will completely change how you feel about things.”

[0118] Characteristics and functions: The techniques by which patients, as defense lawyers, defend themselves. In certain embodiments, this cognitive technique can be selected, particularly when the system detects that the user is experiencing guilt, self-blame, frustration, resentment, self-doubt, anger, and related negative thoughts. The system will suggest that the user write a letter within the system, such as, "If you were hired as a lawyer to defend a client who has experienced something similar to what you have gone through, how would you defend your client?" The system will then guide the user to a page containing a placeholder for writing their defense. All information will be kept confidential.

[0119] Features and Functions: Continuous Medium Techniques According to some embodiments, see Figure 11A and 11B Especially when the system detects that a user is anxious or worried about a variety of things, from small daily occurrences to major life events (which may or may not happen consecutively), continuity cognitive techniques can be selected. This can guide the user to self-correct negative thoughts. For example, in some embodiments, when a user learns that a family member who loves beef has been diagnosed with stomach cancer, the system will report that the user is constantly worried that "eating beef will cause cancer." In some embodiments, the AI ​​psychotherapy system 800 may suggest that the user brainstorm a behavior that carries a 100% risk, such as a "100% risk dietary habit." The system may ask the user to input three dietary habits that the user believes are 100% risky and cause cancer. The user can input, for example, "eating toxic food at every meal," "never eating vegetables," and "eating raw food every day." Then, the system will suggest that the user brainstorm a 0% risk behavior, such as a "0% risk dietary habit." The system will ask the user to input three dietary habits that the user believes have no possibility of causing cancer. For example, the user can input, "eating fresh vegetables at every meal," "never eating strange things," and "never drinking alcohol." Next, the system will suggest that the user identify a 50% risk habit that increases the risk of cancer, such as a "50% risk eating habit." The system will ask the user to input three eating habits that they believe have a 50% chance of causing cancer. Users could input, for example, "My fork and knife are sometimes clean, sometimes not," "I drink a bottle of vodka every month," and "I eat very salty food at every meal." Finally, the user will be asked to rank their anxiety-inducing events along a 0%-50%-100% risk habit axis. In this example, the system will ask the user to rethink how to place "eating beef" on the 0%-50%-100% cancer-risk eating habit axis.

[0120] Features and Functions: Losses and Gains Based on More Adaptive Assumptions According to some embodiments, when the AI-powered psychotherapy system 800 detects that a user has certain persistent irrational assumptions, it can select more adaptive assumption cognition skills. For example, a user might input: "My mother said that if I marry my boyfriend, I won't get any of his property." The AI-powered psychotherapy system 800 might detect an irrational assumption such as "Everything my mother told me will absolutely happen." According to some embodiments, the AI-powered psychotherapy system 100 can generate a more adaptive assumption, such as, "Some of the things my mother told me will happen, and some won't." This assumption will be sent to the user, and then the losses and gains of this new assumption will be discussed. For example, the AI-powered psychotherapy system 100 could send: "Let's think about this a little differently. If what your mother told you happened partly, and partly didn't, what would you think? Can you write down some of the losses caused by this assumption?" The user will expand their thinking, considering the losses of this more adaptive assumption, such as, "I have to try it myself, otherwise I'll never know what will happen." The benefits would be, "I will have many, even unlimited, opportunities to be exposed to new things," "I can do things my mother doesn't allow me to do," and "I no longer have to act according to her wishes every time." According to some embodiments, the AI-powered psychotherapy system 100 may ask the user to rate the losses and gains, with a maximum score of 10. For example, the user might rate the loss of trying anything as 5 points and the gain of experiencing something their mother forbids as 9 points. In some embodiments, if the gain of this more adaptive hypothesis outweighs the loss, the AI-powered psychotherapy system 100 might suggest that the user adopt this more adaptive hypothesis—"What my mother told me will happen partly and partly not." It will send the user a message: "Now that you see the gain outweighs the loss, let's try this more adaptive hypothesis [for 3 days]? It will help you if you remind yourself of this hypothesis every time you think [everything my mother told me will definitely happen]. Let me know how you feel if needed!" Otherwise, in cases where the loss outweighs the gain, the AI-powered psychotherapy system 800 might continuously generate different hypotheses and ask the user to interact until the gain outweighs the loss.

[0121] Features and Functions: Task-Oriented Dialogue System In some embodiments, natural language processing (NLP) methods are combined with artificial intelligence algorithms and statistical tools, such as knowledge graphs (KG).

[0122] In some embodiments, the task-oriented dialogue system can be two-layered. At the NLP level, NLP and machine learning models 116 are components of the example task-oriented dialogue system. The task-oriented dialogue system can be used to continue and guide the dialogue, and generate and perform personalized psychotherapy 330 (e.g., asking the user questions and generating questions and answers) using the NLP and machine learning models 116 to process, analyze, and understand the user's answers. At the evidence-based psychotherapy level, some embodiments interact with the task-oriented dialogue system. These two levels may run in parallel; that is, they can spontaneously collaborate with each other.

[0123] In some embodiments, the AI-based psychotherapy system 100 employs a combination of various AI-related algorithms, such as machine learning and deep learning algorithms, as well as statistical computing methods and tools, focusing on English and Chinese languages ​​and other Chinese dialects (including but not limited to Simplified Chinese, Traditional Chinese, Cantonese, Shanghainese, Teochew and other Chinese dialects).

[0124] The AI-powered psychotherapy system 100 can process text messages and perform psychotherapy when patients are facing emotional difficulties. The system provides timely responses and support. In some embodiments, it also provides additional voice interaction to communicate with patients in English, Chinese, and other Chinese dialects.

[0125] In some embodiments, the AI-powered psychotherapy system 100 can provide professional, appropriate, practical, customized, systematic, continuous, rapid, and low-cost psychotherapy responses to guide patients. In some embodiments, the AI-powered psychotherapy system 800 interacts with patients in various situations during their emotional disturbances, providing relief techniques and emergency psychotherapy counseling. Patients can use customized treatments and psychotherapy anytime, anywhere to alleviate pain and stress, reduce the likelihood of suicide, and shorten the treatment period.

[0126] Some implementations may utilize knowledge graphs of various mental disorders (such as depressive disorders and anxiety disorders) as a supplement to (i) help natural language processing (NLP) algorithms better and more accurately understand each patient's disease characteristics, treatment methods, treatment cycles, suicide risk levels, etc.; and (ii) provide customized psychotherapy, emotional support, cognitive guidance and / or behavior modification programs for each patient based on their different mental state, communication style, social and cultural environment.

[0127] Figure 2 A block diagram of example system 204a-h of a task-oriented dialogue system using open-domain models and / or closed-domain models according to some embodiments is shown. Figure 2 This describes the input flow for open-domain and / or closed-domain models.

[0128] Various embodiments of the AI ​​psychotherapist system 100 utilize a closed-domain retrieval model 204a, a closed-domain retrieval-generative hybrid model 204b, a closed-domain generative model 204c, an open-domain retrieval model 204d, an open-domain retrieval-generative hybrid model 204e, an open-domain generative model 204f, a closed-domain rule-based model 204g, or an open-domain rule-based model 204h. In some embodiments, each of models 204a-h may be based on artificial intelligence technologies, deep learning and machine learning algorithms, statistical methods, and other related technologies and tools, combined with cognitive behavioral therapy (CBT), dialectical behavioral therapy (DBT), interpersonal relationship therapy (IPT), and other related treatments and therapies.

[0129] Figure 2 An example of an AI-powered psychotherapy system 200 (which may be implemented using system 100) is shown, according to some embodiments, which uses natural language processing (NLP) through a task-oriented dialogue system 204a-h. This example demonstrates an end-to-end NLP process implementation. In some embodiments, the method includes retrieving the patient message 202 as an input query from a user interface after the patient inputs a patient message 202 and sends it to the task-oriented dialogue system 204a-h. Deep learning or machine learning algorithms in the task-oriented dialogue system 204a-h analyze the patient message, historical dialogue (i.e., context and content 206), knowledge and information 208 related to psychotherapy and mental disorders and psychological problems, treatment history, etc., to generate personalized psychotherapy 330 and / or perform selected psychotherapy 864.

[0130] In the closed-domain retrieval-generative hybrid model 204b of the AI ​​psychotherapy system 100, the retrieval-generative hybrid model improves the diversity and accuracy of the AI ​​psychotherapy system 100's responses. This embodiment of a task-oriented dialogue system is based partly on retrieval-based deep learning or machine learning algorithms and partly on generative deep learning or machine learning algorithms. The retrieval part provides an answer framework containing key psychological suggestions, while the generative part provides a more comprehensive response to the user by generating personalized answers that are easier to understand and more engaging. For example, the retrieval part might return a response like "Open your chest and take a deep breath," while the generative part might translate it into a more human-like response, such as "To relieve your anxiety, take a deep breath with me and feel your chest expand." In the closed-domain generative model 204c, the generative model adds human-like interaction to the system. In the open-domain retrieval model AI psychotherapy system 100, the AI ​​psychotherapy system 200 uses the open-domain retrieval model 204d, increasing the freedom and flexibility of patient input, as well as the stability of the returned messages. In the open-domain retrieval-based generative hybrid AI psychotherapy system 100, the AI ​​psychotherapy system 200 uses an open-domain retrieval-based generative hybrid model 204e to improve the quality of psychotherapist interventions and the freedom of patient input. In the open-domain generative AI psychotherapy system 100, the AI ​​psychotherapy system 200 uses an open-domain generative model 204f to make the patient feel as if a real human psychotherapist is talking to them. In the closed-domain rule-based model 204g, the simplest dialogues are performed through limited input understanding and generation. In the open-domain rule-based model 204h, simple dialogues cover a variety of topics and domains.

[0131] In some embodiments, the AI-powered psychotherapy system 200 and method focus on a closed-domain retrieval model 204a. In some embodiments, the AI-powered psychotherapy system 200 and method use a closed-domain retrieval generative hybrid model 204b. In some embodiments, the AI-powered psychotherapy system 200 uses a closed-domain generative model 204c. In some embodiments, the AI-powered psychotherapy system 200 uses an open-domain retrieval model 204d. In some embodiments, the AI-powered psychotherapy system 200 uses an open-domain retrieval generative hybrid model 204e. In some embodiments, the AI-powered psychotherapy system 200 uses an open-domain generative model 204f. In some embodiments, the AI-powered psychotherapy system 200 uses a closed-domain rule-based model 204g. In some embodiments, the AI-powered psychotherapy system 200 uses an open-domain rule-based model 204h.

[0132] In various embodiments, the artificial intelligence psychotherapy system 200 may be based on artificial intelligence technology, deep learning and machine learning algorithms, statistical methods and other related technologies and tools, combined with evidence-based psychotherapy and other related treatments and therapies.

[0133] According to some embodiments, various natural language understanding and generation application programming interfaces (APIs) can be task-oriented dialogue systems. The objectives of such dialogue systems are (1) to collect specific information about the user through dialogue; (2) to implement cognitive behavioral therapy and dialectical behavioral therapy by guiding the user to perform selected cognitive skills (tasks); and (3) to improve the user's mood and encourage the user to revisit the therapeutic dialogue system (the application). According to some embodiments, this is not a non-task-oriented dialogue system (for casual / entertaining dialogue with the user without a specific purpose), but rather a system with specific task orientation.

[0134] In some embodiments, after a patient inputs a patient message 126, the AI-powered psychotherapy system 100 retrieves the patient message 126 from the user interface 102 and sends it to the server 104. The server 104 then sends it as an input query to a natural language processing (NLP) and machine learning model 116. In this task-oriented dialogue system, deep learning and / or machine learning algorithms consider the patient's message, historical dialogue (i.e., contextual text and content 206), psychotherapy-related knowledge and information 208, and treatment history to generate a personalized psychotherapy plan.

[0135] Features and Functions: Closed-Domain Retrieval-Based Task-Oriented Dialogue System According to some embodiments, the AI-powered psychotherapy system 100 uses a closed-domain retrieval model 204a. It can be applied using natural language processing interfaces (APIs) of the closed-domain retrieval model, combined with deep learning and / or machine learning and / or statistical modeling. This provides a high-level closed-domain retrieval digital psychotherapist for concept understanding.

[0136] Features and Functions: Closed-Domain Retrieval Generative Task-Oriented Dialogue System In some embodiments, the AI ​​psychotherapy system 100 uses a retrieval-based closed-domain generative hybrid model 204b to improve the diversity and accuracy of the AI ​​therapist's responses. This can be applied in the following ways: (1) using closed-domain retrieval-based generative hybrid model natural language processing interfaces (APIs), or combining deep learning, machine learning, and statistical modeling; or (2) using model / processing interfaces (APIs) divided into two parts, one using closed-domain retrieval-based deep learning, machine learning, and statistical modeling algorithms, and the other using closed-domain generative deep learning, machine learning, and statistical modeling algorithms, both applied within a closed domain.

[0137] Features and Functions: A Task-Oriented Dialogue System Based on Closed-Domain Generation According to some embodiments, the AI-powered psychotherapy system 100 uses a closed-domain-based generation model 204c to enhance its human-like interaction. This can be achieved by applying a closed-domain-based generation natural language API, combined with deep learning and / or machine learning and / or statistical models (e.g., GPT and its variants) within the closed domain.

[0138] Features and Functions: A Task-Oriented Dialogue System Based on Closed-Domain Rules In some embodiments, the AI-powered psychotherapy system 100 uses a closed-domain rule-based model 204g to conduct dialogues with the user. The system may apply a (static) closed-domain rule-based natural language API combined with deep learning, machine learning, and / or statistical models (e.g., GPT and its variants) within the closed domain.

[0139] Features and Functions: A Task-Oriented Dialogue System Based on Open Domain Retrieval In some embodiments, the AI-based psychotherapy system 100 uses an open-domain retrieval-based model 204d to increase the freedom and flexibility of patient input and the stability of psychotherapy 128. It can apply an open-domain retrieval-based natural language API, combined with deep learning / machine learning / statistical models in the open domain (e.g., GPT and its variants).

[0140] Features and Functions: A task-oriented dialogue system based on open-domain retrieval generation In some embodiments, the AI-powered psychotherapy system 100 uses a hybrid model 204e based on open-domain retrieval generation to improve the quality of psychotherapist interventions and the freedom of patient input. It can apply a retrieval-based hybrid natural language API, combined with deep learning and / or machine learning and / or statistical models in the open domain.

[0141] Features and Functions: A task-oriented dialogue system based on open-domain generation In some embodiments, the AI-powered psychotherapy system 100 additionally uses an open-domain generative model 204f to make the patient feel as if they are communicating with a real human psychotherapist. It can apply open-domain generative natural language APIs, as well as deep learning and / or machine learning and / or statistical models in the open domain.

[0142] Features and Functions: A Task-Oriented Dialogue System Based on Open Domain Rules In some embodiments, the AI-powered psychotherapy system 100 uses an open-domain rule-based model 204h to perform dialogues with users in various domains. It can apply an open-domain rule-based natural language API combined with open-domain deep learning and / or machine learning and / or statistical models (e.g., GPT and its variants).

[0143] NLP module for psychotherapy: Features and Functions: NLP Deep Learning Model System Figure 1 Detailed information on the architecture and processes of the NLP and machine learning model 116 in some embodiments is also shown. In some embodiments, upon receiving an NLP request 134, the NLP and machine learning model 116 interprets and generates a psychotherapy response, and utilizes a knowledge graph 316 (sometimes also referred to as a knowledge graph 420 for psychotherapy, psychology, mental disorders, and psychological problems) involving psychotherapy, psychology, mental disorders, and psychological problems, as well as treatment, patient behavior, and other relevant information to support the patient (e.g., if they are experiencing a period of emotional distress) and perform psychotherapy based on the patient's individual condition. In some embodiments, the NLP and machine learning model 116 uses deep learning techniques to process word representations, such as word embeddings, semantic analysis, text mining, and text generation.

[0144] In some embodiments, the artificial intelligence psychotherapy system 100 can generate intelligent psychotherapy.

[0145] In some embodiments, the response of the AI-powered psychotherapy system 100 varies from patient to patient. In some embodiments, the cognitive reconstruction generator 348 and / or the psychotherapy selector 328 can learn and analyze historical context and dialogue content (i.e., context 322) and their associated metrics to create customized responses for patients.

[0146] Features and Functions: Patient Messaging System In some embodiments, such as Figures 3A-3BAs shown, upon receiving patient message 302 (sometimes referred to as patient message 126) from user interface 102, two parallel processes simultaneously analyze the query message: (a) named entity recognizer 314 and / or sentiment classifier 822, and (b) psychological and psychotherapy intent classifier, NLP intent recognizer, topic classifier and / or sentiment analysis model 312.

[0147] Features and Functions: NLP Preprocessing Methods In some embodiments, different parts of the AI ​​psychotherapy system 100 (e.g., a psychotherapy intention classifier and / or an NLP intention recognizer and / or a topic classifier and / or an emotion analysis model 312, a cognitive reconstruction generator 348, and a psychotherapy selector 328) analyze the context 322 and patient messages 302 during the NLP preprocessing method 400, one example of which is shown below. Figure 8A and 8B This is to ensure that psychotherapy 130 is meaningful in terms of context 322, patient messages 302, and the handling of medical and psychotherapeutic matters.

[0148] Figure 4 A flowchart of an example method 400 for psychoanalyzing patient messages using an NLP preprocessing approach, according to some embodiments, is shown. Method 400 is sometimes referred to as NLP preprocessing method 400. According to some embodiments of NLP preprocessing method 400, patient messages 402 are first analyzed by spell checking 404. Patient messages 402 may be segmented into sentences 406, further segmented into words 408, and their parts of speech 410 tagged. Then, patient messages 402 may undergo lemmatization processing 412 and be sent to an emotion and psychological named entity recognizer 414 (sometimes referred to as emotion and psychological named entity recognizer 314). Next, patient messages 402 are sent to a method 414 for looking up terms, concepts, and / or synonyms related to psychology, mental disorders, psychotherapy, and psychological problems. In the process of determining 414, knowledge graphs 420 of psychology, psychotherapy, and various mental disorders and psychological problems (e.g., depression and anxiety), suicide domain libraries 422, psychotherapy domain libraries 424, rehabilitation domain libraries 426, and other relevant domain libraries 428 may be used. Based on the determination results of 414, NLP preprocessing method 400 can identify named entities 418 that are related to psychotherapy, psychological problems and / or mental disorders.

[0149] like Figure 4 As shown, in some embodiments, the NLP preprocessing method 400 is provided as the first part of the overall NLP processor 112. After receiving the patient message 402 from the user interface 102, the following modules perform their functions as follows.

[0150] Example of Sentence Segmentation System 406: In some embodiments, the NLP preprocessing method 400 is configured to perform sentence splitting 406, as shown in the following example: "Yesterday my mind just wouldn't stop. I tried to make it stop but it was always so excited that I lost sleep all night." "Yesterday my mind just wouldn't stop. I tried to make it stop. But it was always so excited. That caused me to lose sleep all night." Example of Word Segmentation System 408: In some embodiments, the NLP preprocessing method 400 may be configured to perform word splitting 408, as shown in the following example: "My appetite is very poor" "My" "appetite" "very" "poor" Example of Part-of-Speech (POS) Tagging System 410: In some embodiments, the NLP preprocessing method 400 may be configured to perform part-of-speech tagging 410 to tag each word. As shown in the following example: "I" "like" "my" "job" "I" – Noun "like" – Verb "my" – Adjective "job" – Noun.

[0151] Example of Lemmatization System 412: In some embodiments, the NLP preprocessing method 400 may be configured to perform lemmatization 412. Different from stemming, lemmatization 412 relies more on part-of-speech tags 410, adjacent sentences, and the current context and content, as shown below: Reducing "Dad", "father", "daddy" to "father" As Figure 4 shown, some embodiments of the artificial intelligence psychotherapy system 100 use an emotion and psychological named entity recognizer 414 — a deep learning NLP algorithm of artificial intelligence, to classify the processed words into psychotherapy entities 418 or non-psychotherapy entities, as shown in the following example: "Self-esteem" — psychotherapy named entity, "ego" — psychotherapy named entity, "noodle soup" — not a psychotherapy named entity.

[0152] After marking the named entities 418 according to the emotion and psychological named entity recognizer 414, the artificial intelligence psychotherapy system 100 may find psychotherapy, mental disorder, psychology or related terms, concepts and / or synonyms in the multi-domain library 414 (for example, rehabilitation domain library 426, suicide domain library 428, psychotherapy domain library 424) to match the medical treatment or diagnosis related to the named entities 418. The artificial intelligence psychotherapy system 100 may send the named entities 418 to the cognitive distortion classifier 328 and the emotion and / or mood tracker and predictor 824, as shown in Figure 3 and Figure 8B shown.

[0153] Features and Functions: Voice Interaction According to the settings in the AI-powered psychotherapy system 100 described above, in some embodiments, voice or text feedback can be provided based on the patient's performance. In some embodiments, if the patient chooses to receive text, the AI-powered psychotherapy system 100 can provide psychotherapy in text format (i.e., AI-powered psychotherapy (text format) 338). If the patient chooses voice reception mode, the AI-powered psychotherapy system 100 will convert the text into voice and respond to the patient's psychotherapy needs via voice (i.e., AI-powered psychotherapy (voice format)), as shown in block diagrams 340 and 342. The AI-powered psychotherapy system 100 can send psychotherapy content in both voice and text output formats. In some embodiments, depending on the specific questions and answers, the best psychotherapy method, the patient's previous psychotherapy experience, etc., the psychotherapy content may include images, videos, music, and / or background notifications.

[0154] According to some embodiments, Figure 7 A schematic diagram of the speech synthesis workflow 700 is shown. Figure 7 This is an illustration of a text-to-speech conversion module for artificial intelligence psychotherapy, which, according to some embodiments, can be used by an artificial intelligence psychotherapy system 100. In some embodiments, once the patient-preferred "speech" format output 702 (sometimes referred to as patient-preferred "speech" format output 336) is determined, the psychotherapy can be sent to a text-to-speech converter 704 (sometimes referred to as a text-to-speech converter 340). The psychotherapy can then be sent from the text-to-speech converter 704 to a speech synthesis generator 706 (i.e., a speech clone) to generate artificial intelligence psychotherapy (speech format) 710 (sometimes referred to as artificial intelligence psychotherapy (speech format) 342).

[0155] In some embodiments, the AI-powered psychotherapy system 100 supports text-to-speech and speech-to-text voice interaction. Through speech recognition (or speech-to-text), the AI-powered psychotherapy system 100 can recognize user-uploaded audio 708 or Mandarin or other Chinese dialects (e.g., Shanghainese, Cantonese, Minnan, etc.) and convert it into text that can be edited, analyzed, and processed by the AI-powered psychotherapy system 100. Furthermore, in some embodiments, the AI-powered psychotherapy system 100 supports voice psychotherapy (i.e., AI-powered psychotherapy (voice format) 710). In some embodiments, AI-powered psychotherapy (voice format) 710 can be presented in various voice forms of Mandarin or other Chinese dialects through a text-to-speech function similar to a text-to-speech converter 704, to enhance the patient's experience with different voices, depending on the patient's individual preference 332 for the psychotherapy output format and target voice.

[0156] Features and Functions: Voice Interaction Triggered Prompts According to some embodiments, the artificial intelligence psychotherapy system 100 may support the following example trigger modes: • Example of a touch-triggered system: Press the bottom right button and the “microphone” button.

[0157] • Example of a voice-triggered system: The patient utters a specific phrase to activate the AI ​​psychotherapy system 100, which then begins processing the patient's voice, including but not limited to "Are you there?" • Example of an action-triggered system: The patient performs a specific action in front of the device 132, including but not limited to waving in front of the sensor (phone camera).

[0158] • Example of a self-triggered system: The events or settings scheduled by the patient will activate the AI ​​psychotherapy system 100, including but not limited to the patient scheduling a therapy session with the AI ​​psychotherapy system 100 every Saturday at 10:00 AM via AI psychotherapy (voice format) 710.

[0159] Features and Functions: Speech-to-Text and Text-to-Speech Interface System In some embodiments, the AI-powered psychotherapy system 100 may use the "pyaudio" library and Google's speech-to-text and text-to-speech APIs to collect speech input from a microphone on a patient's device and convey the speech of the AI-powered psychotherapy (speech format) 710 to the patient in a speech manner. The AI-powered psychotherapy system 100 may use a speech-to-text converter 310 and / or a text-to-speech converter 704 via a speech recognition-related software development kit (SDK). In some embodiments, to achieve functionality similar to the text-to-speech converter 704, the following methods and related series may be used: • Implement speech-to-text, speech recognition, text-to-speech, speech synthesis, TTS and / or other APIs or modifications with equivalent or similar functions (including but not limited to Huawei Cloud Voice Interaction Service, Google Voice API, IBM Watson API, Voice API, Speech-to-Text API, Siri API, Wit API, Speech2Topics API, ReadSpeacker API, Rev.AI API, iFlytek API, iFlytek speech-to-text and / or speech synthesis or other APIs, frameworks or similar software from iFlytek, Google Cloud TTS (cloud-based text-to-speech) and / or speech-to-text or other similar APIs, Microsoft Azure speech-to-text, Amazon speech-to-text, Amazon Polly, CMU Sphinx, Microsoft Bing speech recognition, Houndify API, Snowboy Hotword Detection, etc.). The use of this software can be based on different electronic operating systems and their auxiliary or modified operating systems, cloud platforms, cloud computing services, and cloud operating systems, including but not limited to iOS, Android, Windows, Symbian, MacOS, MacOS X, DOS, Unix, Linux, Chrome OS, Deepin Linux, Anchao OS, UOS, Vi m / Vi, Neokylin OS, Ubuntu, Ubuntu Kylin, WiOS, AliOS, Start OS, Blackberry OS, Raspberry Pi OS, Harmony OS, AWS, Euler OS, Tencent Cloud, JD Cloud, Alibaba Cloud, Huawei Cloud, Wangsu Cloud, ChinaCache, Dilian Cloud, Kingsoft Cloud, StallerCloud, MOS, Google Cloud, Microsoft Azure, Shijie Cloud, China Telecom Cloud, Amazon Web Services, QingCloud, etc.

[0160] • Implement specific or non-specific libraries and their modifications to collect voice input from a device microphone, including but not limited to PyAudio, python-sounddivce, speech recognition (PyPI), CFFI, etc., based on specific or non-specific programming languages, including but not limited to Python, Java, PHP, C++, C, Cython, CSS, HTML family (such as HTML5), JavaScript, R, Perl, Swift, Ruby, Go, Rust, C#, TCL, Scala, Node.js, Android, Lua, QML, AngularJA, Cordova, Ionic, RN, weex, Kotlin, Flutter, DCould, Delphi, Mono, Haskell, etc.; and / or • Specific or non-specific APIs, SDKs, interfaces, software, tools, and their modifications that can achieve similar or equivalent speech-to-text and text-to-speech conversion purposes under specific or non-specific programming languages, based on specific or non-specific operating systems or cloud services.

[0161] Features and Functions: Speech Synthesis Generator In some embodiments, the text-to-speech converter 704 may attempt to generate speech feedback based on the text and message, as well as the user-defined target speech preference 332. The customer can select different voices as the communication tool in the system, and the output speech can be implemented as follows: 1. When the user does not specify a preference or the user selects a voice from the built-in voices of the AI ​​psychotherapy system 100 according to the methods and tools described herein, the built-in voices of the AI ​​psychotherapy system 100 may be used.

[0162] 2. A specific voice specified by the user after performing speech synthesis (e.g., voice cloning) technology in the artificial intelligence psychotherapy system 100, which can be achieved through speech synthesis technology.

[0163] In some embodiments, users can upload voice recordings 708 or collect voice input from the device's microphone, which can serve as input to the system, i.e., the user-uploaded voice audio 708. Voice input can come from important figures in their lives, such as deceased family members or people living far away. The AI-powered psychotherapy system 100 supports users' urgent needs and uses the voices of people important to them to assist their therapeutic process.

[0164] In some embodiments, the AI ​​psychotherapist system 100 uses a speech synthesis generator 706 (e.g., a speech clone) to clone speech input for communication with clients and generate synthesized speech to converse with clients in AI psychotherapy (speech format) 710, which sounds similar to the uploaded speech. The synthesized speech may have similar speech (phrasing, details, imagery, syntax, tone, etc.) and voice (volume, pitch, rhythm, pauses, resonance, intonation, speed, timbre, etc.) characteristics.

[0165] This system and method achieve digital representation of speech in short audio clips (several seconds or minutes) through speech cloning. In some embodiments, a speech cloning API based on deep learning or machine learning can be used. To achieve functionality similar to a speech synthesis generator (e.g., speech cloning) 706, the following methods can be used: • In some embodiments, the AI ​​psychotherapy system 100 may use specific, non-specific, or combinations of models and algorithms based on machine learning, deep learning, statistics, artificial intelligence, quantitative analysis, numerical simulation, computer science, interfaces, or modifications and variations thereof, which have equivalent or similar functionality (including but not limited to WaveNet, MelNet, Deep Voice series, SV2TTS, synthesis networks, vocoder networks, speaker encoder networks, speaker encoders, speaker adapters, World vocoders, Griffin-Lim vocoders, encoder-decoders, real-time speech cloning, WaveRNN, Tacotron series, GE2E, synthesizers, converters, VoiceLoop, parallel WaveNet, concatenated TTS, parameterized TTS, SampleRNN, Char2Wav, sequence-to-sequence models, few-sample generation models, meta-learning generation models, speaker encoding, neural TTS, multi-speaker TTS, Mellotron, etc.).

[0166] • In some embodiments, speech-to-text, speech recognition, text-to-speech, speech synthesis, TTS, speech cloning, and other APIs or modifications with equivalent or similar functions (including but not limited to Huawei Cloud Voice Interaction Service, Google Voice API, IBM Watson API, Voice API, Speech-to-Text API, Siri API, Wit API, Speech2Topics API, ReadSpeaker API, Rev.AI API, iFlytek APIs, iFlytek speech-to-text / speech synthesis and other APIs, frameworks or similar software from iFlytek, Google Cloud TTS (cloud-based text-to-speech) / speech-to-text and other similar APIs, Microsoft Azure speech-to-text, Amazon speech-to-text, Amazon Polly, CMUSphinx, Microsoft Bing speech recognition, resemble.ai speech cloning and related APIs, Houndify API, Snowboy Hotword Detection, etc.) can be used, based on different electronic operating systems and their secondary or modified operating systems, cloud platforms, cloud computing services and cloud operating systems (including but not limited to iOS, Android, Windows, Symbian, MacOS, MacOSX, DOS, Unix, Linux, Chrome) based on different electronic operating systems and their secondary or modified operating systems, cloud platforms, cloud computing services and cloud operating systems (including but not limited to iOS, Android, Windows, Symbian, MacOS, MacOSX, DOS, Unix, Linux, Chrome) OS, Deepin Linux, Anchao OS, UOS, Vim / Vi, Neokylin OS, Ubuntu, Ubuntu Kylin, WiOS, Ali OS, Start OS, Blackberry OS, Raspberry Pi OS, Harmony OS, AWS, Euler OS, Tencent Cloud, JD Cloud, Alibaba Cloud, Huawei Cloud, Wangsu Cloud, ChinaCache, Dilian Cloud, Kingsoft Cloud, StallerCloud, MOS, Google Cloud, Microsoft Cloud, Shijie Cloud, China Telecom Cloud, Amazon Web Services, QingCloud, etc.

[0167] • In some embodiments, specific or non-specific libraries or toolkits and their modifications are used for data collection for speech synthesis or speech generation based on specific or non-specific programming languages ​​(including but not limited to Python, Java, PHP, C++, C, Cython, CSS, HTML family (e.g., HTML5), JavaScript, R, Perl, Swift, Ruby, Go, Rust, C#, TCL, Scala, Node.js, Android, Lua, QML, AngularJA, Cordova, Ionic, RN, weex, Kotlin, Flutter, DCould, Delphi, Mono, Haskell, etc.).

[0168] • In some embodiments, specific or non-specific APIs, SDKs, interfaces, software, tools and their modified versions may be used, all of which can achieve similar or equivalent purposes to the artificial intelligence psychotherapy system 100, and are all based on specific or non-specific programming languages, and are carried out under specific or non-specific operating systems or cloud services.

[0169] Front-end and full-stack: Features and Functions: User Interface System Features and functions: devices, operating systems like Figure 1 As shown, in some embodiments, the system includes various electronic devices, such as device 132 and platforms, including but not limited to websites, mobile operating systems (e.g., iOS, Android), smartwatches (e.g., iWatch), wearable devices, computers, laptops, and any other electronic terminals that the patient can access.

[0170] Features and Functions: Conversational UI / UX System In some embodiments, the example UI / UX is a communication interface between the digital psychotherapist and the patient, such as... Figure 9-1 As shown in Figure 1.

[0171] Features and Functions: Communication Interface System like Figure 1 As shown, in some embodiments, the front-end and full-stack architecture of the AI-powered psychotherapy system 100 can implement network sockets as interfaces to build and execute communication operations. Some embodiments may use the following tools, methods, frameworks, and / or modifications: • Specific or non-specific network socket-based technologies and their modifications and updates (including but not limited to Websockets, easysocket, quickerver, netty2, colouradoFTP, apache James, Java mail server, Java sockets, raining sockets, HP-socket, Cindy, brynet, teleport, Winsock, and workerman); • Based on specific or non-specific network sockets and their modified and updated versions based on different programming languages ​​(including but not limited to python, Java, PHP, C++, C, CSS, HTML series (such as HTML 5), JavaScript, R, Perl, Swift, Ruby, Go, Rust, C#, TCL, Scala, Node.js, Android, Lua, QML, AngularJA, Cordova, Ionic, RN, weex, Kotlin, Flutter, DCould, Delphi, Mono, Haskell, etc.); • Any specific socket-based modification and update version interface in other specific programming languages, unspecified socket-based modification interfaces and updates, and all means of achieving a communication purpose similar to or equivalent to the AI ​​psychotherapy system 100; • In any programming language, socket-based modified and updated versions of interfaces or combinations thereof, unspecified socket-based modified interfaces and updated versions or combinations thereof, and all means of achieving communication purposes similar to or equivalent to the AI ​​psychotherapy system 100. • The SendMessage() and PostMessage() functions; • UdpClient, TcpClient, Remote Direct Memory Access (RDMA), Jro, WinPcap OpenOnload, .NET Remoting, Windows Communication Foundation (WCF), and other specific and non-specific means that achieve the same or similar communication purposes as this system and solution through network socket interfaces; and / or • Interfaces, software, APIs, SDKs, and other tools and their modifications that can be used to achieve similar or equivalent network communication purposes in specific or non-specific programming languages.

[0172] Features and Functions: Server According to some embodiments, Figure 1An example of server 104 is also shown. The chat center is responsible for maintaining and managing registered patients and directing the AI-powered psychotherapy system 100 to deliver specific messages to specific patients. In some embodiments, the AI-powered psychotherapy system 100 provides different psychotherapy for different patients, rather than providing a uniform psychotherapy. Patients can communicate with the chat center through registered user channels, unregistered user channels, and broadcast channels. According to some embodiments, the registered channel may be open to registered patients, while the unregistered channel may be open to unregistered patients.

[0173] According to some embodiments, when a patient enters a text or voice message on user interface 102, the message is sent via a broadcast channel (a type of user messaging channel). The patient's message is broadcast to server 104 of the AI ​​psychotherapy system 100 using the "SendMessage()" function, the "Send()" function in a network socket, and other methods serving the same or similar purpose as the "SendMessage()" or "send()" functions. In some embodiments, the message contains pointers related to the patient, and according to some embodiments, the message content is sent via the patient channel in UTF-8 encoding. Psychotherapy responses generated by the AI ​​psychotherapy system 100 are returned to the patient's user interface 102 via the "recv()" and "PostMessage()" functions in a network socket, and other similar methods in network sockets.

[0174] According to some embodiments, for registered users, a client mapping can be used, where key-value pairs stored in the mapping include: a key, a pointer to a user instance, and a Boolean value indicating whether the user is in a live connection state. In some embodiments, the client type acts as an intermediary between the chat center and the network socket connection. Each user value includes a pointer to the chat center, a pointer to the network socket connection, and a buffered channel for messages with a "send" function. Messages can be written using a write method. In some embodiments, server 104 can register a server method as a hypermedia document transfer handler for registering users to the chat center and upgrading standard connections to network socket connections.

[0175] In some embodiments, the following tools, methods, frameworks, and / or variations thereof may be used: any specific or non-specific API and its variations and updated versions interfaces in a specific or non-specific programming language, and all means of achieving server communication purposes similar to or equivalent to the AI ​​psychotherapy system 100, including but not limited to REST, SOAP, Facebook Messenger, Slack Bot, Wit.AI, Dialogflow, Amazon Lex, Bronshop.AI, Botlibre, Notify.eu, BotDelive, Microsoft Bot Framework, Hangouts, Telegram Bot, ChatBot, CrispChat, NativeChat, GraphQL, Falcor, gRPC, JSON-Pure, oData, Postman; interfaces, software, APIs, SDKs, other tools, and variations thereof capable of achieving similar or equivalent server communication purposes in a specific or non-specific programming language; rear end: Features and Functions: Authorization System According to some embodiments, the AI-powered psychotherapy system 100 can be activated by connecting the user interface 102 to the server 104, verifying registered or unregistered patients and other authorization requests, such as membership authorization 106 (also known as user authentication). Each request may contain a server-based signature, such as a patient ID and / or token. If the patient is successfully authorized, a server code "200" is generated, indicating that the patient has successfully logged into the AI-powered psychotherapy system 100. If the signature and token verification fails security checks (including SSL / TLS communication with the server 104), a code "401" is returned to the user interface 102, indicating that the user is not authorized. Codes "401" and "200" are merely examples of signal codes, and other types of codes or communication signal formats with similar or identical purposes may also be used. For privacy verification, the AI-powered psychotherapy system 100 encourages patients to undergo real-time facial recognition verification each time they log in. These facial images and videos may also be stored in a data storage management system and applied to mental disorder detection, patient recovery speed monitoring, and treatment assessment by utilizing facial image detection and recognition technology for depression.

[0176] Features and Functions: Queue Generation System According to some embodiments, patient statement format checking 304 is used to identify whether the patient's message is in voice format 306 or text format 308. After converting the voice input to text format (if applicable), or if the patient directly inputs text, the AI ​​psychotherapy system 100 can output the text-formatted input to queue 110 via a patient message processing procedure, one example being... Figures 3A-3B As shown in the diagram. Each entry in queue 110 from queue generator 108 may contain a patient's signature, patient identifier (ID), NLP request, and message ID. When multiple patients send patient messages 126 simultaneously, mapping method 114 can use queue 110 to identify different patients. Queue 118, as update information and data for queue 110, consists of user ID and message ID. In some embodiments, NLP requests can be sent to the next step within their own thread to enable personalized dialogue.

[0177] Features and functions: Mapping system According to some embodiments, in mapping method 114, the AI-powered psychotherapy system 100 can store different patient messages 126 from different patients in different spaces of a data storage management system 122 based on the patient ID and / or token associated with the patient message 126. In some embodiments, the AI-powered psychotherapy system 100 can generate customized psychotherapy, treatment plans, schedules, solutions, guidance, etc., through an AI and statistical processor (i.e., psychotherapy 128). For example, suppose Amy and John both suffer from major depressive disorder and are currently experiencing an episode. Amy is struggling with the loss of her father, while John is struggling with unemployment. The AI-powered psychotherapy system 100 can add different patient IDs and tokens to Amy and John's messages and store their personal and health information respectively. Then, according to some embodiments, the AI-powered psychotherapy system 100 can send the patient message 126 to a natural language processing (NLP) processor 112, where the NLP processor 112 can provide customized psychotherapy, treatment plans, schedules, guidance, and solutions. Some implementations utilize a patient’s behavior or lifestyle to address the root cause of their major depressive disorder (e.g., emotion and / or mood trackers and predictors 824).

[0178] Features and Functions: Data Storage Management System Interface According to some embodiments, the AI-powered psychotherapy system 100 may use a data storage management interface 120, which may be an Open Database Connectivity (ODBC) interface, to access the data storage management system 122. The AI-powered psychotherapy system 100 may also use other technologies that can achieve the same or similar functionality as this system and method through interfaces including, but not limited to: Database Access Objects (DAO), Remote Database Objects (RDO), Object Linking and Embedded Databases (OLE DB), ActiveX Data Objects (ADO), ADO.NET, Java Database Connectivity (JDBC), PHP Data Objects (PDO), etc. Figure 1 In this context, queue 118 is an enhanced version of queue 110, which includes at least the user ID and message ID.

[0179] In some embodiments, the AI-powered psychotherapy system 100 can receive a queue of patient messages and psychotherapy responses generated by an NLP model. Both messages (patient messages and responses to patients) can be stored in a data storage management system 122 to establish a patient’s historical health record in order to track and predict the patient’s recovery process, medical condition, suicide risk, feedback, etc.

[0180] Features and Functions: Data storage management and customer relationship management (CRM) system Figure 5 This shows, according to some embodiments, by Figure 1 The example shown is of 500 patient data collected and / or used by an AI-powered psychotherapy system. Figure 5 Examples of data relationship designs for data storage management 122, according to some embodiments, are also shown to store patient-related in-hospital and out-of-hospital data.

[0181] like Figure 5 As shown, based on the common key ID—patient ID—hospital data 506 and external data 504 can be associated with a single patient record 502. By merging hospital data 506 and external data 504, an embodiment of the artificial intelligence psychotherapy system 100 can utilize these two sets of data to develop and generate specific psychotherapy for the patient.

[0182] According to some embodiments, Figure 6 This is a graphical example of a patient's treatment timeline. Figure 1 This example can be used in an AI-powered psychotherapy system. The example illustrates, according to some embodiments, the development of a general schedule 600 related to patient records 502 by merging in-hospital data 506 and out-of-hospital data 504. The general schedule 600 can utilize the methods of the AI-powered psychotherapy system 100 to provide patients with a more comprehensive psychological analysis.

[0183] In some embodiments, the data and information 124 of the artificial intelligence psychotherapy system 100 include, but are not limited to, each patient's historical and current health data and records. Two important components are shown below: • The patient's basic personal information, including but not limited to: name, age, occupation, health information (including physical and mental health), doctor's name, diagnosis record, hospital code, doctor's record, medications, laboratory test reports and results, records of the psychologist, and records of the nurse.

[0184] • Communication and health information based on some embodiments of the system, including but not limited to: patient ID, message ID and content, psychotherapy ID and content, mental and emotional type ID and content, solution ID and content, dialogue topic ID and content, severity of illness, timeline, treatment cycle, suicide risk, anxiety level, depression level, etc.

[0185] Features and Functions: Data Storage Management System Implementation Examples According to some embodiments, after the AI ​​psychotherapy system 100 receives patient messages from the queue generator 108 and the NLP and machine learning model 116 generates psychotherapy 130 (sometimes referred to as psychotherapy 210), these two messages (patient message 126 and psychotherapy 130) can be stored in the data storage management system 122 to establish the patient's historical healthcare record in order to track and predict their recovery process, healthcare status, suicide risk, feedback, etc.

[0186] In some embodiments, MySQL, based on the SQL language, is used to implement the data storage management system 122. In some embodiments, functions similar to the data storage management system 122 are implemented, including but not limited to storage, transmission, processing, modification, addition, deletion, and retrieval. When using the artificial intelligence psychotherapy system 100 to store data, the following tools, methods, frameworks, and variations thereof can be used to construct an equivalent or similar data storage management system 122 based on various data manipulation languages: Suitable database software includes Oracle RDMS, Microsoft SQL Server, Microsoft Access, PostgreSQL, Apache Derby, Berkeley DB, eXist, HSQL, Ingres, LevelDB-Google, mSQL, SQLite, XinDice, 4th Dimension, Adabas, askSam, Cache, c-tree, and also IBM DB1, IBM DB2, dBase, FileMaker, FoxBase, Google Fusion Tables, GuptaSQLBase, HyperFileSQL, IDMS, IMS, Informix, InterBase, MaxDB, MongoDB, Knack, Google Cloud BigTable, SAP HANA, TeamDesk, Amazon RDS, Microsoft Visual FoxPro, SyBase, MS SQL-Server, Paradox Borland, PrimeBase, RRDtoo, Tamino XML, Td beengine, Teradata, Visual dBase, MariaDB, Cassandra, CouchDB, Amazon DynamoDB, Neo4j, Firebirdsql, RavenDB, RethinkDB, Orient... DB, Arango DB, JAM Stack, Cockroach DB, Citus, Timescale DB, ToroDB, MySQLDocStore, TiDB, CouchBase, Elastic, ElasticSearch, HDFS, Redis, ClickHouse, ScyllaDB, HBase, Google Cloud Data Storage, Apache Spark, Cosmos DB, Percona Server for MongoDB, Solr, Lucene-based data storage, datalog, TigerGraph, RedisGraph, etc. Various data manipulation languages, including but not limited to the following languages ​​and their modified and updated versions: SQL-based languages ​​(e.g., pgsql, Psql, SQL PL, NZPL SQL, PSQL, SQL / PSM, PL / SQL, T-SQL, PL / pgSQL, SQL scripts, Gsql, Watcom-SQL), Cypher, XQuery, HQL, Gremlin, Spark, SPL, ABAP, SSP, NoSQL, MQL, dBase, etc. • Tools, methods, frameworks, and modifications for achieving similar and equivalent data-related functions, including but not limited to storing, transmitting, processing, modifying, adding, deleting, and retrieving data through the AI-powered psychotherapy system 100 in specific or non-specific data manipulation languages. Features and functions: Psychotherapy response system Features and Functions: Response Provision System In some embodiments, the artificial intelligence psychotherapy system 100 provides psychotherapy, and the psychotherapy 130 generated by the NLP and machine learning model 116 is sent to the server 104.

[0187] Features and functions: Psychotherapy system In some embodiments, the AI ​​psychotherapy system 100 sends personalized psychotherapy 832 (e.g., “Are you still troubled by insomnia yesterday?”) as a specific psychotherapy 128 of the AI ​​psychotherapy system 100, which is sent to the user interface 102 as “intelligent psychotherapy” 128. The patient can then view the “intelligent psychotherapy” 128 and its related instructions.

[0188] Scope of application According to some embodiments, the artificial intelligence psychotherapy system 100 can handle general psychological problems, severe psychological problems, neurotic psychological problems, and mental disorders. The system and methods encompass examples of psychological problems and mental disorders, their classifications, types, categories, names, and International Classification of Diseases (ICD-10-CM) and classification systems. "General psychological problems" include, but are not limited to, psychological problems related to work, study, social life, personal life, and / or interpersonal relationships arising from real-world stimuli at various stages of development. The patient's emotional reactions are short-lived, under control, do not cause serious impairment to social functioning, and have not generalized. These problems include, but are not limited to, common life problems such as parent-child relationships, love, marriage and family, career choices and employment, career confusion, interpersonal relationships, and personal growth. "Severe psychological problems" include, but are not limited to, mental health conditions resulting from relatively strong real-world stimuli. The patient initially experiences strong emotional reactions that are prolonged and whose content is sufficiently broad to generalize into an adverse mental health condition. Examples include the breakdown of significant relationships (such as romantic relationships ending, divorce, bereavement), academic failure (failing important exams, unemployment), moral and legal issues (domestic violence), and other problems that lead to prolonged distress and anxiety. "Neurotic psychological problems" include an individual's excessive worry about having a serious illness. Although there is no actual medical diagnosis, this debilitating condition may be due to inaccurate perceptions of physical or mental health. General characteristics of "mental disorders" include abnormal thoughts, perceptions, moods, behaviors, and relationships with others, and may include depression, bipolar disorder, schizophrenia and other psychoses, dementia, and developmental disorders including autism.

[0189] Characteristics and Functions: Psychotherapy Methods According to some implementation schemes, an example method for performing psychotherapy using an artificial intelligence psychotherapy system 100 is provided. The method includes the following steps: obtaining patient messages from a user interface; identifying emotional and psychotherapy status based on user input using (i) multiple machine learning classifiers, (ii) natural language processing (NLP) techniques, and (iii) a knowledge graph related to psychotherapy, mental disorders, and psychological problems; determining treatment options for the emotional and psychotherapy status using one or more machine learning classifiers based on user input. These treatment options include user interactions and a schedule of interactions; displaying the treatment options and tracking the patient's user interactions and schedule.

[0190] In some implementations, user input can be in text or voice format. The method also includes converting any voice input into text input and using the text input to identify emotions, psychotherapy status, and / or therapy options.

[0191] In some implementations, the method also includes outputting treatment options in a voice format based on the patient's voice output preferences.

[0192] In some implementations, an emotion collector 816, an emotion tracker and predictor 824, an emotion and psychological named entity recognizer 820, and / or an emotion classifier 822 may be used.

[0193] In some implementations, the emotion collector (sometimes referred to as the emotion collector) associated with the aforementioned user interface / user experience, including the emojis shown in the image above, is not a classifier, but a tool used to collect user input data.

[0194] In some implementations, classification and natural language understanding models are applied to document-level or paragraph-level text. In some implementations, the aforementioned user interface / user experience uses cognitive techniques to handle diary and / or defense attorney tasks. In some implementations, the aforementioned automatic cognitive distortion classifier is also included.

[0195] As described in this article, mental disorder is a term that encompasses all mental illnesses, such as anxiety disorders and depressive disorders, and requires a physician, mental health professional, or questionnaire to determine whether a user has a mental disorder. Emotions may include feelings, which include sensations / perceptions / thoughts, and moods (emotional states and psychological states). In some implementations, the categorization of psychotherapeutic terms (e.g., named entity recognition) helps identify emotions or conditions and can be used to determine the severity of specific symptoms to track patient recovery progress. For example, for major depressive disorder, a patient may experience weight loss or gain after developing the depressive disorder; therefore, a named entity recognizer would look for terms such as “diet / appetite / weight” to see any related symptoms and changes in those symptoms.

[0196] In some implementations, natural language processing (NLP) technology categorizes terms in user input to determine their relevance to psychotherapy, emotions, mental disorders, and psychological problems.

[0197] In some implementations, the method also includes using a bioinformatics analyzer to predict future moods and further determining treatment options based on those future moods.

[0198] In some implementations, firstly, a first or more classifiers are used to label cognitive distortions (e.g., using 14 labels). Training data may include sentences with cognitive distortions. Next, a second or more classifiers can be used for further processing (e.g., a classifier with 78 categories). In some implementations, the input to the classifiers may include user feedback and / or labels from the classifier. In some implementations, user feedback sets a subjective preference for a particular cognitive skill among the 78 labels, while the classification results from the first or more classifiers are objective. In some implementations, the first or more classifiers include a classifier for classification, while the second or more classifiers include a classifier designed to recommend the best skills for personal use based on personal preferences; therefore, its function is essentially recommendation.

[0199] In some embodiments, the training data consists of balanced sentences with cognitive distortions, such as "Because I'm divorced, I'm a complete failure; label: affective reasoning" (this label was determined by multiple psychotherapists). There are 8000 sentences per label, for a total of 14 labels, modeled using BERT, GTP, LSTM, and bi-LSTM, with BERT selected as the best model based on performance. The system uses supervised learning, but semi-supervised / unsupervised / reinforcement learning can achieve similar results. In some embodiments, the models BERT, GTP, LSTM, and bi-LSTM are used for feature extraction. Some embodiments choose BERT as the final feature extractor. Some embodiments use deep learning, thus eliminating the need for manual feature extraction; features are automatically extracted by the neural network. Some embodiments use cross-validation for validation, for example, dividing a dataset containing 10,000 samples with standard answer labels into 8,000 samples for training and 2,000 samples for validation.

[0200] In some embodiments, the method further includes using a list of cognitive distortions to identify cognitive distortions, thereby identifying psychotherapeutic conditions.

[0201] In some embodiments, the method further includes tracking user interactions, including: receiving confirmation from the patient that a treatment option is invalid, obtaining second user input from the patient, and repeatedly identifying emotional and psychotherapy status, treatment options, and displaying new treatment options in response to the second user input.

[0202] In some embodiments, displaying treatment options includes using an open-domain generative model to communicate with the patient.

[0203] In some embodiments, the tracker and predictor 824 are used for classification (e.g., natural language understanding models and classifiers) and prediction (e.g., time series machine learning models) related to the UI / UX sentiment tracker. In some embodiments, the sentiment recognizer is based on a named entity recognition model.

[0204] In some embodiments, classification and natural language understanding models are applied to article-level or paragraph-level text. In some embodiments, the UI / UX described above uses cognitive techniques from diaries and / or defense lawyers. In some implementations, the aforementioned automatic cognitive distortion classifier is included.

[0205] Some embodiments select a treatment for user A based on feedback from other users (e.g., 70% of other users like the treatment). In some embodiments, after treatment is completed, the system asks user A to provide feedback on the treatment with "like" or "dislike" as options. In this way, the system continuously collects user A's feedback on all the different treatments recommended to him / her. The more a user likes a particular treatment, the more likely the system is to recommend subsequent treatments. In some embodiments, a random coefficient, such as 10%, is used to randomly recommend treatments that user A has never used before.

[0206] In some embodiments, the treatment has a timeline and a user interface. An example user interface may include examples of all treatments and / or cognitive techniques. In some embodiments, the emotion tracker's user interface includes a timeline in the form of a line graph.

[0207] In some embodiments, the system tracks an action if the user is cooperating with a particular therapy, such as self-reward (an example of a cognitive correction approach). For example, if a user sets up 30 minutes of swimming, the system will check in with the user after 30 minutes. Some embodiments solicit user opinions and / or user feedback, and / or collect such feedback. Some embodiments analyze the severity of the user's condition and / or use voice cloning. Voice output cloning includes converting text to audio as a different form of communication, applying specific voices to a particular user to express or encourage their emotions. According to some embodiments, in addition to voice, a machine learning classifier can suggest the severity of the user's condition. Some embodiments use voices unrelated to the therapy and mental disorder analysis, and offer multiple output formats selectable by the user.

[0208] In some embodiments, there are cognitive distortion labels, such as statements that "must" or "should" be done. A "should / must statement list" is activated when a machine learning model assigns this label to a specific input sentence by user A. The system can save sentences input by a specific user A in this list. This "should / must statement list" is not for identifying types of cognitive distortion, but rather for collecting all sentences input by user A that are labeled with "must" or "should". In some embodiments, the "should / must statement list" is used to determine the content of a patient's "principle manual"—"should," "must," "if-then" statements—which are markers of distorted habitual thinking. In some embodiments, upon detecting the aforementioned entities, the system stores the relevant information in a database for analysis.

[0209] Another example of psychotherapy Based on some embodiments, this document describes another example method for performing psychotherapy using an AI psychotherapy system 100. The method includes obtaining patient messages from a patient via a user interface. The method also includes determining one or more classifiers based on the patient messages and other patient information. This other patient information includes in-hospital and out-of-hospital data. The method further includes selecting a psychotherapy for the patient based on a list of candidate psychotherapy treatments using one or more classifiers. The method also includes providing the psychotherapy to the patient in one of text or voice formats, based on a determination partially based on patient preferences.

[0210] In some embodiments, the selection of psychotherapy also includes analyzing the context of psychotherapy to ensure that the context is logically, medically, and psychotherapeutoriously meaningful.

[0211] In some embodiments, obtaining patient information includes using biometric verification to verify the patient's identity.

[0212] In some embodiments, biometric verification includes facial recognition.

[0213] In some embodiments, selected psychotherapy includes a combination of rule-based closed-domain generation models and open-domain generation models.

[0214] In some embodiments, the system includes a task-oriented dialogue system. The task-oriented dialogue system gathers specific information about the patient by driving dialogue with the patient, performs cognitive behavioral therapy and dialectical behavioral therapy by guiding the user to execute selected cognitive correction techniques, and encourages the patient to revisit the AI-based psychotherapy system.

[0215] In some embodiments, selecting psychotherapy includes collecting heart rate data from the patient and using a time-series-based model with deep neural networks to analyze the heart rate data to monitor treatment progress and symptom development.

[0216] In some embodiments, selecting psychotherapy includes: collecting medical records from the patient; and using OCR text recognition technology to read non-electronic information from the medical records.

[0217] On the other hand, without any human intervention, this system and method provide a way to intelligently control psychotherapy using artificial intelligence, based on evidence-based psychotherapy and other related treatments and therapies. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient messages from the patient through a user interface; identifying emotions in user input as affective terms based on the patient messages using one or more named entity recognizers; and / or identifying terms in user input as terms related to psychotherapy, and / or mental disorders, and / or psychological problems, and / or psychology, based on the patient messages using one or more named entity recognizers.

[0218] In some embodiments, the method further includes using a knowledge graph related to psychology, and / or psychotherapy, and / or mental disorders, and / or psychological problems, and / or emotions. Both emotion recognition and terminology classification require knowledge graphs (e.g., those referenced above). Figure 3A The knowledge graph described is 316). Named entity recognizers may only be used to recognize terms in sentences, and the domain in which terms need to be recognized depends on the knowledge graph. A knowledge graph provides a network graph with many nodes (e.g., a tree structure in a data structure) that (1) uses concepts / terms as nodes and (2) describes the relationships between different concepts.

[0219] In some embodiments, user input is in text or voice format, and any voice input is converted to text input and the text input is used to identify emotions and terms in the user input.

[0220] On the other hand, without any human intervention, this system and method provide a way to intelligently control psychotherapy using artificial intelligence, based on evidence-based psychotherapy and other related treatments and therapies. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient information from the patient through a user interface; and, based on the patient information, identifying and classifying user input into one or more terms related to psychotherapy and / or emotions and / or mental disorders and / or psychological problems and / or psychology using one or more NLP classifiers.

[0221] In some embodiments, the method further includes using a knowledge graph related to psychology, and / or psychotherapy, and / or mental disorders, and / or psychological problems, and / or emotions.

[0222] In some embodiments, the user input is in text or voice format, and any voice input is converted to text and the text is used to identify and classify terms in the user input.

[0223] On the other hand, without any human intervention, this system and method provide a way to intelligently control psychotherapy using artificial intelligence, based on evidence-based psychotherapy and other related treatments and therapies. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient messages from the patient through a user interface; and, based on the patient messages, identifying intent 320 in the user input using one or more NLP intent recognizers and / or one or more topic classifiers and / or one or more sentiment analysis models and / or one or more intent classifiers.

[0224] In some embodiments, the method further includes using knowledge graphs related to psychology, psychotherapy, mental disorders, psychological problems and / or emotions.

[0225] In some embodiments, the user input may be in text or voice format, and the method further includes: converting any voice input into text input and using the text input to recognize intent 320.

[0226] On the other hand, an intelligently controlled psychotherapy method is provided, which utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other related treatments and therapies without any human intervention. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient messages and timestamps from the patient via a user interface; or obtaining one or more emojis and timestamps from the patient via a user interface; and predicting future psychological states and treatment plans using one or more machine learning prediction models based on the timestamps and patient messages or emojis.

[0227] In some embodiments, user input may be in text or voice format, and the method further includes converting any voice input into text input and using the text input to predict future mental conditions and treatment plans.

[0228] In some embodiments, the method further includes predicting future psychological conditions and treatment plans, including: acquiring facial images and videos of the patient; and using psychological condition recognition technology based on facial images and videos.

[0229] In some embodiments, the method further includes predicting future psychological conditions and treatment plans, including: collecting the patient's heart rate data; and analyzing the heart rate data using a time series model.

[0230] In some embodiments, the method further includes predicting future psychological conditions and treatment plans, including: collecting the patient's electrocardiogram (ECG) data; and analyzing the ECG data using a machine learning pattern recognition model.

[0231] In some embodiments, the method further includes predicting future mental states and treatment plans, including: responding to received timestamps, new patient messages, emojis, facial images, facial videos, heart rate data, electrocardiograms, and / or medical records; repeatedly predicting future mental states and treatment plans; and storing new predictions of future mental states and treatment plans.

[0232] On the other hand, a method for intelligently controlled psychotherapy is provided, which utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other relevant treatments and therapies without any human intervention. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient messages from the patient through a user interface; and identifying cognitive distortions based on the patient messages using one or more natural language processing (NLP) classifiers.

[0233] In some embodiments, the method further includes using a list of cognitive distortions to identify cognitive distortions. In some embodiments, psychotherapy employs cognitive techniques. Some embodiments use a variety of cognitive techniques (e.g., 78 cognitive techniques), which may correspond to an equal number of psychotherapies (e.g., 78 psychotherapies). Feedback from psychotherapy can be used to identify cognitive distortions.

[0234] In some embodiments, the user input may be in text or voice format, and the method further includes: converting any voice input into text input and using the text input to identify cognitive distortions.

[0235] On the other hand, a method for intelligently controlled psychotherapy is provided, which utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other relevant treatments and therapies without any human intervention. This method can be executed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient information from the patient via a user interface; and determining personalized psychotherapy based on the patient information using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills (e.g., any one of the 78 cognitive skills in list 862).

[0236] In some embodiments, the method further includes using knowledge graphs in the fields of psychology, psychotherapy, mental disorders, psychological problems, and / or emotions.

[0237] In some embodiments, the user input may be in text or voice format, and the method further includes: converting any voice input into text input and using the text input to generate personalized psychotherapy.

[0238] In some embodiments, the method further includes determining personalized psychotherapy, including: using the patient's feedback history; using the feedback history of each psychotherapy session; and / or using a list of cognitive distortions.

[0239] In some embodiments, determining personalized psychotherapy includes: after receiving confirmation from the patient that the psychotherapy is ineffective: obtaining a second user input from the patient; repeating the determination of personalized psychotherapy; and providing a new psychotherapy for the second user input.

[0240] On the other hand, a method of intelligently controlled psychotherapy is provided, which utilizes artificial intelligence to perform psychotherapy based on evidence-based psychotherapy and other relevant treatments and therapies without any human intervention. This method can be performed using the AI ​​psychotherapy system 100 described above. The method includes: obtaining patient messages from the patient through a user interface; and providing personalized psychotherapy based on the patient messages using (i) one or more natural language processing (NLP) task-oriented dialogue systems and (ii) a list of cognitive skills.

[0241] In some embodiments, the method further includes using knowledge graphs relating to the following domains: psychology, psychotherapy, mental disorders, psychological problems, and / or the emotional domain.

[0242] In some embodiments, the user input may be in text or voice format, and the method further includes: converting any voice input into text input and using the text input to generate personalized psychotherapy.

[0243] In some embodiments, the method further includes performing psychotherapy in a speech format based on the patient's preference for speech output.

[0244] In some embodiments, providing personalized psychotherapy includes: after receiving confirmation from the patient that the psychotherapy is ineffective; obtaining a second user input from the patient; re-determining personalized psychotherapy; and providing new psychotherapy based on the second user input.

[0245] In some embodiments, providing psychotherapy includes engaging in dialogue with a patient using the following models: (i) a closed-domain rule-based model, (ii) a closed-domain retrieval-based model, (iii) a hybrid closed-domain retrieval-generated model, (iv) a closed-domain generation-based model, (v) an open-domain rule-based model, (vi) an open-domain retrieval-based model, (vii) a hybrid open-domain retrieval-generated model, or (viii) an open-domain generation-based model.

[0246] In some embodiments, the method further includes: collecting the patient's medical records; and using OCR text recognition technology to read non-electronic information from the medical records.

[0247] In some embodiments, the method further includes generating a preferred speech using speech synthesis technology based on the patient's speech preferences; and then using the generated preferred speech to converse with the patient.

[0248] Data information content in data storage management system The following are sample data information contents (including in-hospital data and out-of-hospital data records) in the data storage management system.

[0249] Hospital records: • Medical and treatment history (e.g., "major depressive disorder, with an episode 5 years ago") • Basic personal information, see Table 2.1, registration form (Leahy, Holland and McGinn, 2011) (e.g., family medical history, age, gender, etc.) • Nurse records, medication records, doctor's records, current and past diagnoses, radiographic images, family and related assessments • Laboratory test reports and results (e.g., blood and heart tests: coagulation test, disposable vacuum blood collection tube coagulation test, EDTA anticoagulation test, computerized multi-lead electrocardiogram, heart rate variability trend graph, infrared thermography, medical history and mental status examination, venous blood collection, complete blood count, biochemical tests, endocrine tests) • Self-assessment (e.g., multiple-choice symptom scales and assessments: Self-Rating Anxiety Scale, Self-Rating Depression Scale, Beck Depression Scale, Symptom Checklist-90-Revised (SCL-90-R), Yale-Brown Obsessive-Compulsive Disorder Scale, Pittsburgh Sleep Quality Index, Hypomania Checklist (HCL-32)). Hospital outside records: • Biometric information and records (e.g., voice messages, sound information, facial images and videos, DNA reports, heart rate), text conversation messages, emotion records [including Table 10.4 Patient Event-Emotion-Thought Records (Leahy, Holland, and McGinn, 2011)], regular self-assessments, family and related assessments, suicide (suicide attempt) records. • Behavioral records [e.g., exercise (type, frequency, intensity), meditation] See Table 2.8, Patient Weekly Activity Schedule (Leahy, Holland, and McGinn, 2011). Classification of mental disorders and international diagnostic codes. The classifications and international diagnostic codes of mental disorders to which this system and method are applicable and covered are shown below, with the International Classification of Diseases 10th Revision (ICD-10-CM) diagnostic codes and DSM-5 codes indicated.

[0250] • Anxiety and stress-related disorders: F93.0, F94.0, F40.228, F40.23x, F40.230, F40.231, F40.232, F40.233, F40.248, F40.298, F40.10, F41.0, F40.00, F41.1, F10.180, F10.280, F10.980, F15.180, F15.280, F15.980, F12.180, F12.280, F12.980, F16.180, F16.280, F16.980, F18.180, F18 .280、F18 .980、F11 .188、F11 .288、F11.988、F13 ​​.180、F13 .280、F13 .980、F15 .180、F15 .280、F15 .980、F14 .180、F14 .280、F14 .980, F19 .180, F19 .280, F19 .980, F06 .4, F41 .8, F41 .9 • Depressive Disorders: F34.8, F32.0, F32.1, F32.2, F32.3, F32.4, F32.5, F32.9, F33.0, F33.1, F33.2, F33.3, F33.4, F33.5, F33.9, F34.1, N94.3, F10.14, F10.24, F10.94, F16.14, F16.24, F16.94, F18.14, F18.24, F18.94, F11.14, F11.24, F11.94, F13.14, F13.24, F13.94, F15 .14, F15 .24, F15 .94, F14 .14, F14 .24, F14 .94, F19 .14, F19 .24, F19 .94, F06 .31, F06 .32, F06 .34, F32 .8, F32 .9 • Dissociative disorders: F44.81, F44.0, F48.1, F44.89, F44.9 • Eating and eating disorders: F98.3, F98.21, F50.8, F50.01, F50.02, F50.2, F50.8, F50.9 • Bipolar disorder and related disorders: F31.11, F31.12, F31.13, F31.2, F31.71, F31.72, F31.9, F31.31, F31.32, F31.4, F31.5, F31.75, F31.76, F31.9, F31.81, F34.0, F10.14, F10.24, F10.94, F16.14, F16.24, F16.94, F13.14, F13.24, F13.94, F15.14, F15.24, F15.94, F14.14, F14.24, F14 .94, F19 .14, F19 .24, F19 .94, F31.89, F06 .33, F06 .34 Obsessive-compulsive disorder and related disorders: F31.9, F42, F45.22, F63.3, L98.1; F15.188, F15.288, F15.988, F14.188, F14.288, F14.988, F19.188, F19.288, F19.988; F06.8, F42 • Personality Disorders: F60.0, F60.1, F21, F60.2, F60.3, F60.4, F60.81, F60.6, F60.7, F60.5, F07.0, F60.89, F60.9 • Schizophrenia spectrum and other psychotic disorders: F22, F23, F20.81, F20.9, F25.0, F25.1, F10.159, F10.259, F10.959, F12.159, F12.259, F12.959, F16.159; F16.259, F16.959, F18.159, F18.259, F18.959, F13.159, F13.259, F13.959, F15.159; F15.259, F15.959, F14.159, F14.259, F14.959, F19 .159, F19 .259, F28, F29 • Sexual behavior, gender identity disorder, sexual perversion, and other sex-related disorders: F64.2, F64.1, F64.8, F64.9, F65.3, F65.2, F65.81, F65.51, F65.52, F65.4; F65.0, F65.1, F65.89, F65.9 • Physical symptoms and related disorders: F45.1, F45.21, F44.4, F44.5, F44.6, F44.7, F54, F68.10, F45.8, F45.9 Substance-related and addiction disorders: F10.10, F10.20, F10.129, F10.229; F10.929, F10.239, F10.232, F10.99, F15.929, F15.93, F15.99, F12.10, F12.20; F12.129, F12.229, F12.929, F12.288, F12.99, F16.10, F16.20, F16.129, F16.229, F16.929, F16.983, F16.99, F18.10, F18.20, F18 .129、F18 .229、F18 .929、F18 .99; .99, F13.10, F13 .20, F13 .129, F13 .229, F13 .929, F13 .239, F13 .232, F13 .99; F15 .10, F15.20, F14 .10, F14 .20, F15 .129, 15 .229, F15 .929, F14 .129, F14 .229; F14 .929, F15.122, F15 .222, F15 .922, F14 .122, F14 .222; F14 .922, F15 .23, F14 .23; F15 .99, F14.99, Z72 .0, F17 .200, F17 .203, F17 .209, F19 .10, F19 .20, F19 .129, F19 .229, F19.929, F19 .239, F19 .99, F63 .0 • Post-traumatic stress disorder: Childhood reactive attachment disorder, F94.1, F94.2, F43.10, F43.0, F43.21, F43.22, F43.23, F43.24, F43.25, F43.20, F43.8, F43.9 • Disruptive impulse control and behavioral disorders: F91.3, F63.81, F91.1, F91.2, F91.9, F60.2, F63.1, F63.2, F91.8, F91.9 • Sleep-wake disorders: F51.01, F51.11, G47.419, G47.411, G47.429, G47.33, G47.31, G47.37, R06.3, G47.34, G47.35, G47.36, G47.21, G47.22, G47.23, G47.24, G47.26, G47.20; F51.3, F51.4, F51.5, G47.52, G25.81, F10.182, F10.282, F10.982, F12.188, F12.288, F12 .988、F13 ​​.182、F13 .282、F13 .982、F15 .182、F15.282、F15 .982、F14 .182、F14 .282、F14 .982、F19 .182、F19 .282、F19 .982, F11.182, F17.208, G47.09, G47.00, G47.19, G47.10, G47.8, G47.9 • Sexual dysfunction: F52.32, F52.21, F52.31, F52.22, F52.6, F52.0, F52.4, F10.181, F10.281, F10.981, F13.181, F13.281, F13.981, F15.181, F15.281, F15.981, F14.181, F14.281, F14.981, F19.181, F19.281, F19.981, F11.181, F11.281, F11.981 • Excretory disorders: F98.0, F98.1, N39.498, R15.9, R32 • Neurodevelopmental disorders: Intellectual disability (mental developmental disorder) F70, F71, F72, F73; Intellectual disability F70-F73, F88, F79, F80.2, F80.0, F80.81, F80.89, F80.9, F84.0, F90.2, F90.0, F90.1, F90.8, F90.9; F81.0, F81.81, F81.2, F82, F98.4, F95.2, F95.1, F95.0, F95.8, F95.9, F88 The codes for some general psychological problems, severe psychological problems, and neurotic psychological problems to which this system and method are applicable and covered are shown below. This system and method adopt the descriptions, interpretations, scope, and details of relevant problems and symptoms in the 10th edition of the Classification of Diseases (ICD-10-CM) and the Diagnostic and Statistical Manual of Mental Disorders (DSM-5).

[0251] Abuse and Neglect: T74 .12 XA, T74 .12 XA, T76.22XD, Z69.010, Z69.020, Z62.810, Z69.011, Z69.021, T74.02 XA, T74.02XD, T76.02XA, T76.02XD, Z69.010, Z69 .020, Z62 .810, Z69 .011, Z69 .021, T74 .32XA, T74 .32XD, T76 .32 XD, T76 .11 XA, T76 .11 XD, Z69 .11, Z91 .410, Z69 .12, T74 .21 XA, T74 .21XD, T76 .21 .01 XA, T74 .01XD, T76.01 XA, T76 .01XD, Z69 .11, Z91 .412, Z69 .12, T74 .31 .12, T74 .11XA, T74 .11 XD, T76 .11 XA, T76 .11 XD, T74 .21 XA, T74 .21 XD, T76 .21 .31 XD, Z69 .81, Z69 .82 • Family parenting related issues: Z62.820, Z62.891, Z62.29, Z62.898 • Other issues related to key support members: Z63.0, Z63.5, Z63.8, Z63.4 • Education and Career Issues: Z55.9, Z56.82, Z56.9 • Housing and economic related issues: Z59.0, Z59.1, Z59.2, Z59.3 • Economic issues: Z59.4, Z59.5, Z59.6, Z59.7, Z59.9, Z59.8 Other issues related to the social environment: Z60.0, Z60.2, Z60.3, Z60.4, Z60.5, Z60.9 • Issues related to interactions with crime or law: Z65.4, Z65.0, Z65.1, Z65.2, Z65.3 • Inquiries: Z70.9, Z71.9 • Issues related to sociopsychological, personal, and environmental conditions: Z65.8, Z64.0; Z64.1, Z64.4, Z65.4, Z65.5, Z71.81, Z65.9 Other details in personal history: Z91.49, Z91.5, Z91.82, Z91.89, Z72.9; Z72.811, Z72.810 • Issues related to access to medical and other healthcare services: Z75.3, Z75.4 • Failure to adhere to medical treatment: Z91.19, E66.9, Z76.5, Z91.83, R41.83 For ease of explanation, the foregoing description refers to specific embodiments. However, the exemplary discussion above is not exhaustive and is not intended to limit the scope of the claims to the precise forms disclosed. Many modifications and variations are possible in light of the foregoing guidance. These embodiments were chosen to better illustrate the basic principles of the claims and their practical application, thereby enabling others skilled in the art to fully utilize these embodiments and make various modifications according to the specific intended use.

Claims

1. A non-transitory computer-readable storage medium storing one or more programs executable by a computer, the one or more said programs comprising an artificial intelligence (AI) psychotherapy system, said AI psychotherapy system including instructions for operation: From the patient, user input, including patient messages, is obtained through the user interface; Based on patient messages, emotions in user input are identified as sentiment terms using one or more natural language processing named entity recognizers. as well as Based on the patient information, the words in the user input are classified into at least one of the following categories using one or more named entity recognizers: psychotherapy, mental disorder, psychological problem, psychology; as well as Cognitive distortions are identified using an automatic cognitive distortion classifier based on the identified emotional terms and words. as well as Based on the identified cognitive distortions, personalized psychotherapy is determined using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills through a psychotherapy selector; Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided through a psychotherapy selector using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

2. The non-transitory computer-readable storage medium of claim 1, wherein the AI ​​psychotherapy system further includes instructions for the following operations: Use knowledge graphs that cover at least one of the following areas: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

3. The non-transitory computer-readable storage medium according to any one of claims 1-2, wherein the user input is in either text or voice format, and the AI ​​psychotherapy system further includes instructions for the following operations: Convert any voice input to text input, and use the text input to identify emotions and words in the user's input.

4. A non-transitory computer-readable storage medium storing one or more programs for computer execution, the one or more programs comprising an AI psychotherapy system, the AI ​​psychotherapy system including instructions for operating: obtaining user input, including patient messages, from a patient via a user interface; and Based on the patient's information, identify and classify the words in the user input as one or more of the following terms using one or more Natural Language Processing (NLP) classifiers: psychotherapy, emotion, mental disorder, psychological problem, and psychology. as well as Based on the identified words, cognitive distortions are identified using an automatic cognitive distortion classifier. as well as Based on the identified cognitive distortions, personalized psychotherapy is determined using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills through a psychotherapy selector; Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided through a psychotherapy selector using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

5. The non-transitory computer-readable storage medium of claim 4, wherein the AI ​​psychotherapy system further includes instructions for the following operations: Use knowledge graphs that cover at least one of the following areas: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

6. The non-transitory computer-readable storage medium according to any one of claims 4-5, wherein the user input is either text or voice format, and the AI ​​psychotherapy system further includes instructions for operation as follows: Convert any voice input to text input, and use the text input to identify and categorize words in the user input.

7. A non-transitory computer-readable storage medium storing one or more programs executable by a computer, the one or more programs including an AI psychotherapy system, the AI ​​psychotherapy system including instructions for operation: From the patient, user input, including patient messages, is obtained through the user interface; and Based on patient information, use at least one of the following to identify intent in user input: one or more NLP intent recognizers, one or more topic classifiers, one or more sentiment analysis models, and one or more intent classifiers; as well as Based on the identified intent, cognitive distortions are identified using an automatic cognitive distortion classifier; as well as Based on the identified cognitive distortions, personalized psychotherapy is determined using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills through a psychotherapy selector; Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided through a psychotherapy selector using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

8. The non-transitory computer-readable storage medium of claim 7, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Use knowledge graphs that cover at least one of the following areas: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

9. The non-transitory computer-readable storage medium according to any one of claims 7-8, wherein the user input is in either text or voice format, and the AI ​​psychotherapy system further includes instructions for operation: Convert any voice input to text input and recognize intent from the text input.

10. A non-transitory computer-readable storage medium storing one or more programs for computer execution, the one or more programs including an AI psychotherapy system, the AI ​​psychotherapy system including instructions for operating to: obtain user input, including patient messages, from a patient via a user interface; as well as Based on patient information, use one or more classifiers to identify cognitive distortions; as well as Based on the identified cognitive distortions, personalized psychotherapy is determined using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills through a psychotherapy selector; Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided through a psychotherapy selector using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

11. The non-transitory computer-readable storage medium of claim 10, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Use a list of cognitive distortions to identify cognitive distortions.

12. The non-transitory computer-readable storage medium according to any one of claims 10-11, wherein the user input is either text or voice, and the AI ​​psychotherapy system further includes instructions for operation: Convert any voice input to text input, and use the text input to identify cognitive distortions.

13. A non-transitory computer-readable storage medium storing one or more programs for computer execution, the one or more programs comprising an AI psychotherapy system, the AI ​​psychotherapy system including instructions for operating: obtaining user input, including patient messages, from a patient via a user interface; and Use an automatic cognitive distortion classifier to identify cognitive distortions based on patient messages; as well as Based on the identified cognitive distortions, personalized psychotherapy is determined using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills through a psychotherapy selector; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

14. The non-transitory computer-readable storage medium of claim 13, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Use knowledge graphs that cover at least one of the following areas: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

15. The non-transitory computer-readable storage medium according to any one of claims 13-14, wherein the user input is either text or voice, and the AI ​​psychotherapy system further includes instructions for operation: Convert any voice input to text input, and use the text input to determine personalized psychotherapy.

16. The non-transitory computer-readable storage medium according to any one of claims 13-14, wherein, The personalized psychotherapy is determined by using at least one of the following: the patient’s feedback history, the feedback history of each psychotherapy session, and a list of cognitive distortions.

17. The non-transitory computer-readable storage medium according to any one of claims 13-14, wherein determining the personalized psychotherapy comprises: Respond to confirmation from the patient that psychotherapy has been ineffective: Obtain a second user input from the patient; as well as The personalized psychotherapy was repeatedly selected, and a new psychotherapy was provided to the second user input.

18. A non-transitory computer-readable storage medium storing one or more programs for computer execution, the one or more programs comprising an AI psychotherapy system, the AI ​​psychotherapy system including instructions for operating: obtaining user input, including patient messages, from a patient via a user interface; and Use an automatic cognitive distortion classifier to identify cognitive distortions based on patient messages; as well as Based on the identified cognitive distortions, personalized psychotherapy is provided through a psychotherapy selector using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as The personalized psychotherapy is processed by a cognitive reconstruction generator to generate a personalized psychotherapy response.

19. The non-transitory computer-readable storage medium of claim 18, further comprising: Use knowledge graphs covering the following areas: psychology, and / or psychotherapy, and / or mental disorders, and / or psychological problems, and / or emotions.

20. The non-transitory computer-readable storage medium according to any one of claims 18-19, wherein the user input is in either text or voice format, and the AI ​​psychotherapy system further includes instructions for operation: Convert any voice input to text input, and use the text input to generate personalized psychotherapy.

21. The non-transitory computer-readable storage medium according to any one of claims 18-19, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Based on the determination of the patient's preferred speech output, psychotherapy is provided in the form of speech.

22. The non-transitory computer-readable storage medium according to any one of claims 18-19, wherein providing personalized psychotherapy further includes tracking user interactions, which includes: Respond to confirmation from the patient that psychotherapy has been ineffective: Obtain a second user input from the patient; as well as It repeatedly generates personalized psychotherapy and provides new psychotherapy based on the input of a second user.

23. The non-transitory computer-readable storage medium according to any one of claims 18-19, providing the psychotherapy through a model selected from the group, which consists of the following: (i) Closed-domain rule-based model, (ii) Closed-domain search-based model, (iii) Closed-domain search-based generative hybrid model, (iv) Closed-domain generative model, (v) Open-domain rule-based model, (vi) Open-domain search-based model, (vii) Open-domain search-based generative hybrid model, and (viii) Open-domain generative model.

24. The non-transitory computer-readable storage medium according to any one of claims 18-19, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Collect medical records from patients; and Using OCR text recognition technology to read non-electronic information on medical records.

25. The non-transitory computer-readable storage medium according to any one of claims 18-19, wherein the AI ​​psychotherapy system further comprises instructions for operating the following: Based on the determination of the patient's preferred speech, the preferred speech is generated using speech synthesis technology; and The patient is then spoken to using the generated preferred voice.

26. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. One or more natural language processing named entity recognizers are used to identify emotions in user input as emotion terms based on patient messages; as well as One or more named entity recognizers are used to classify words in user input into words of at least one of the following categories based on patient messages: psychotherapy, mental disorder, psychological problem, psychology; as well as An automatic cognitive distortion classifier is used to identify cognitive distortions based on the identified sentiment terms and words; as well as, A psychotherapy selector is used to determine personalized psychotherapy based on identified cognitive distortions using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills. Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

27. The AI ​​psychotherapy system of claim 26, wherein one or more natural language processing named entity recognizers, and one or more named entity recognizers are used to utilize a knowledge graph relating to at least one of the following domains: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

28. The AI ​​psychotherapy system according to any one of claims 26-27, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter used to convert any voice input into text input; as well as One or more natural language processing named entity recognizers are used to identify emotions in user input using the text input; as well as One or more named entity recognizers are used to recognize words in the user input using the text input.

29. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. as well as One or more NLP classifiers are used to identify and classify words in user input as one or more of at least one of the following, based on patient messages: psychotherapy, emotion, mental disorder, psychological problem, and psychology; as well as An automatic cognitive distortion classifier is used to identify cognitive distortions based on the identified words. as well as, A psychotherapy selector is used to determine personalized psychotherapy based on identified cognitive distortions using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills. Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

30. The AI ​​psychotherapy system of claim 29, wherein one or more NLP classifiers are used to utilize a knowledge graph relating to at least one of the following domains: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

31. The AI ​​psychotherapy system according to any one of claims 29-30, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter used to convert any voice input into text input; as well as, One or more NLP classifiers are used to identify and classify words in the user input using the text input.

32. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. as well as One or more NLP intent recognizers, one or more topic classifiers, one or more sentiment analysis models, and at least one of one or more intent classifiers are used to identify intent in user input based on patient messages; as well as An automatic cognitive distortion classifier is used to identify cognitive distortions based on the identified intent. as well as, A psychotherapy selector is used to determine personalized psychotherapy based on identified cognitive distortions using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills. Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

33. The AI ​​psychotherapy system of claim 32, wherein at least one of the one or more NLP intent recognizers, one or more topic classifiers, one or more sentiment analysis models, and one or more intent classifiers is used to utilize a knowledge graph relating to at least one of the following domains: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

34. The AI ​​psychotherapy system according to any one of claims 32-33, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter used to convert any voice input into text input; as well as One or more NLP intent recognizers, one or more topic classifiers, one or more sentiment analysis models, and at least one of one or more intent classifiers are used to recognize intent from text input.

35. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. as well as One or more classifiers are used to identify cognitive distortions based on patient messages; as well as A psychotherapy selector is used to determine personalized psychotherapy based on identified cognitive distortions using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills. Alternatively, based on the identified cognitive distortions, personalized psychotherapy can be provided using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

36. The AI ​​psychotherapy system of claim 35, wherein one or more classifiers are used to identify cognitive distortions using a list of cognitive distortions.

37. The AI ​​psychotherapy system according to any one of claims 35-36, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter used to convert any voice input into text input; as well as One or more classifiers are used to identify cognitive distortions using the text input.

38. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. as well as An automatic cognitive distortion classifier is used to identify cognitive distortions based on patient information; as well as A psychotherapy selector is used to determine personalized psychotherapy based on identified cognitive distortions using (i) one or more machine learning recommendation engines and (ii) a list of cognitive skills. as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

39. The AI ​​psychotherapy system of claim 38, wherein the automatic cognitive distortion classifier is used to utilize a knowledge graph relating to at least one of the following domains: psychology, psychotherapy, mental disorders, psychological problems, and emotions.

40. The AI ​​psychotherapy system according to any one of claims 38-39, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter used to convert any voice input into text input; as well as, The psychotherapy selector is used to determine personalized psychotherapy based on the text input.

41. The AI ​​psychotherapy system according to any one of claims 38-39, wherein, The psychotherapy selector is used to determine the personalized psychotherapy using at least one of the following: the patient's feedback history, the feedback history of each psychotherapy, and a list of cognitive distortions.

42. The AI ​​psychotherapy system according to any one of claims 38-39, wherein the psychotherapy selector is used for: Respond to confirmation from the patient that psychotherapy has been ineffective: Obtain a second user input from the patient; and The personalized psychotherapy was repeatedly selected, and a new psychotherapy was provided to the second user input.

43. An AI-powered psychotherapy system, the AI-powered psychotherapy system comprising: The user interface is used to obtain user input, including patient messages, from the patient. as well as An automatic cognitive distortion classifier is used to identify cognitive distortions based on patient information; as well as A psychotherapy selector for providing personalized psychotherapy based on identified cognitive distortions by using (i) one or more task-oriented dialogue systems and (ii) a list of cognitive skills; as well as A cognitive reconstruction generator is used to personalize the personalized psychotherapy to generate a personalized psychotherapy response.

44. The AI ​​psychotherapy system according to claim 43, wherein the automatic cognitive distortion classifier is used to utilize a knowledge graph relating to the following areas: psychology, and / or psychotherapy, and / or mental disorders, and / or psychological problems, and / or emotions.

45. The AI ​​psychotherapy system according to any one of claims 43-44, wherein the user input is either text format or voice format, and the AI ​​psychotherapy system further comprises: A speech-to-text converter for converting any voice input into text input; and The psychotherapy selector is used to generate personalized psychotherapy based on the text input.

46. ​​The AI ​​psychotherapy system according to any one of claims 43-44, wherein the AI ​​psychotherapy system further comprises: A text-to-speech converter is used to provide psychotherapy in speech form based on the patient's preferred speech output.

47. The AI ​​psychotherapy system according to any one of claims 43-44, wherein the psychotherapy selector is used for: Respond to confirmation from the patient that psychotherapy has been ineffective: Obtain a second user input from the patient; and It repeatedly generates personalized psychotherapy and provides new psychotherapy based on the input of a second user.

48. The AI ​​psychotherapy system according to any one of claims 43-44, wherein the psychotherapy selector is configured to provide the psychotherapy through a model selected from a group, and to converse with the patient, the group comprising: (i) Closed-domain rule-based model, (ii) Closed-domain search-based model, (iii) Closed-domain search-based generative hybrid model, (iv) Closed-domain generative model, (v) Open-domain rule-based model, (vi) Open-domain search-based model, (vii) Open-domain search-based generative hybrid model, and (viii) Open-domain generative model.

49. The AI ​​psychotherapy system according to any one of claims 43-44, wherein the AI ​​psychotherapy system further comprises a health record recognizer for: Collect medical records from patients; and Using OCR text recognition technology to read non-electronic information on medical records.

50. The AI ​​psychotherapy system according to any one of claims 43-44, wherein the AI ​​psychotherapy system further comprises: A text-to-speech converter is used to generate preferred speech based on the patient's preferred speech, using speech synthesis technology. as well as A text-to-speech converter for conversing with the patient using the generated preferred speech.

Citation Information

Patent Citations

  • Assisting psychological cure in automated chatting

    US20200082928A1

  • Artificial intelligence assisted service provisioning and modification for delivering message-based services

    US20200244605A1