An adaptive dignity interview interaction system and method

Abstract

The application provides an adaptive dignity interview interaction system and method, relates to the field of interview interaction, and comprises: an access and session layer used for interview interaction with a patient; an interview understanding and decision-making layer used for creating a dignity interview theme graph, acquiring an interaction result based on the dignity interview theme graph and the patient interview interaction, analyzing the coverage of interaction content, the substantial content of interaction and the emotional load of the patient during the interaction process, and controlling the interaction process; a content generation and security control layer used for generating a narrative unit through structured narration according to the interaction result, screening the narrative unit to generate a preliminary version of narrative content based on the emotional load of the patient, and performing security verification on the interaction result; and an auditing and output layer used for auditing the preliminary version of narrative content and outputting narrative content, wherein the narrative content at least comprises a narrative video and a narrative text. The scheme can solve the problems of theme coverage and inability to conduct an interview in combination with patient fatigue.

Description

Technical Field

[0001] This application relates to the field of interview interaction technology, and more specifically, to an adaptive dignity interview interaction system and method. Background Technology

[0002] Dignity therapy is an interview-based psychosocial intervention for patients with serious illnesses or terminal illnesses. A typical process involves trainees conducting a life review using semi-structured questions, recording experiences, values, and information the patient wishes to be remembered or passed on to family and friends. The interview recordings are then transcribed and edited into a legacy document, which is then reviewed, revised, and confirmed by the patient. Existing research also indicates that online methods can reduce time and geographical barriers, but technological availability, patient fatigue, and professional review remain critical conditions.

[0003] Clinical manual workflows heavily rely on the interviewer's questioning skills, transcription and editing experience, and risk identification abilities. The physical strength, attention span, speech rate, and emotions of palliative care patients can fluctuate during interviews; a fixed question sequence or continuous open-ended follow-up questions may lead to repetitive questioning, ineffective interaction, or interruptions. Furthermore, the transmitted content contains personal experiences and end-of-life wishes; adding new figures, events, causes, or evaluations not expressed by the patient to the generated text increases the review burden and distorts the content.

[0004] Existing large language model interview systems are already capable of generating semi-structured outlines, conducting voice interaction, generating follow-up questions in real time based on responses, and recording emotional information. For example, published patent CN118886503A discloses a semi-structured interview system based on a fine-tuned large language model, which includes outline generation, dynamic follow-up questioning, and multimodal information recording.

[0005] Existing digital psychological interview solutions can also analyze psychological states based on voice, text, video, and scale results. For example, published patent CN118098587A discloses a solution for digital doctors to conduct multiple rounds of psychological interviews and output suicide risk analysis reports. These technologies demonstrate that "AI interviews, dynamic follow-up questions, emotion recognition, suicide-related analysis, and text report generation" already have a considerable amount of publicly available information.

[0006] However, the existing technology has the following problems: 1. General interview systems typically generate the next question freely based on the current semantics, lacking a question node state mechanism that simultaneously encodes "the topic must be covered" and "the patient has the right to no content, refuse to answer, or postpone".

[0007] 2. Existing systems struggle to establish individual baselines based on changes in patient speech pauses, response lengths, and non-response patterns during the session. Consequently, they cannot combine patient-reported fatigue with behavioral deviations for question complexity, follow-up question limits, and pause recovery control.

[0008] 3. Judging completeness solely based on the number of words or semantic relevance of the answer makes it difficult to distinguish between different states such as "not fully answered", "clearly no relevant content", "unwilling to answer", and "to be completed next time due to fatigue".

[0009] 4. General generative models tend to mix substantive content from patients with transitional and polished content generated by the model, lacking anchor points for substantive information sources, reverse extraction verification, and backend source tags.

[0010] 5. Automated interviews, risk alerts, medical staff review, patient confirmation, family notification, and final draft solidification are usually completed in isolation by different modules, lacking auditable task status and version chain. Summary of the Invention

[0011] Embodiments of this application provide an adaptive dignity interview interaction system and method to address the problems existing in the prior art.

[0012] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0013] According to a first aspect of the embodiments of this application, an adaptive dignity interview interaction system is provided, comprising: The access and session layer is used for interviewing and interacting with patients. The interview understanding and decision-making layer is used to create a dignity interview theme map and obtain interaction results based on the dignity interview theme map and patient interviews. At the same time, it analyzes the coverage of interaction content, the substantive content of interaction and patient emotional load during the interaction process and controls the interaction process. The content generation and security control layer is used to generate narrative units based on the interaction results, filter the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and perform security verification on the interaction results. The review and output layer is used to review the initial version of the narrative content and output the narrative content, which includes at least: narrative video and narrative text.

[0014] In some embodiments of this application, based on the foregoing scheme, the interview understanding and decision-making layer includes: Create a unit to create a dignity interview topic graph based on the set of topic nodes and the optional jump relationships between nodes; wherein each topic node includes at least a node identifier, topic category, question template, set of required information slots, set of optional information slots, entry conditions, taboo marker, maximum number of follow-up questions, current status and source record; Each information slot can be set to any of the following states: not asked, asked and awaiting response, obtained substantive content, clearly no content, refused to answer, fatigued and postponed, forbidden to skip, transferred to human, terminated; The processing unit is used to perform voice / text processing on the interaction results of patient interviews; The first analysis unit is used to analyze the topic coverage and substantive content of the interaction results based on the status of the information slots. The second analysis unit is used to analyze the patient's fatigue during the interview and calculate the interaction load value; The control unit is used to control the interview interaction process based on topic coverage, substantive content, and interaction load.

[0015] In some embodiments of this application, based on the foregoing scheme, the formula for calculating topic coverage is as follows:

[0016] in, For topic coverage, For the first The first topic Individual information slot status, For indicator functions, To cover weights, This is a set of terminating states, which include: obtaining substantive content, clearly no content, refusing to answer, taboo skipping, transferring to human agent, and termination. The formula for calculating the content quality is as follows:

[0017] in, For the sake of substantive content, Weighting based on substantive content. For indicator functions, For the first The first topic Individual information slot status, To obtain the state of the substantive content.

[0018] In some embodiments of this application, based on the foregoing scheme, the specific process by which the second analysis unit analyzes the patient's fatigue during the interview and calculates the interaction load value includes: At the start of the interview, the patient's self-reported fatigue score was obtained on a scale of 0-10, and the fatigue status was asked again when the preset time interval was reached. In addition to timed triggering, fatigue inquiry can be triggered in advance when the patient actively clicks fatigue / pause, does not respond for the first time, pauses significantly increase, response length continues to decrease, or other behavioral abnormalities occur; After obtaining several rounds of valid responses, establish individual patient baselines, including baseline pause duration, baseline response length, and optional baseline speech rate; Calculation time Interactive load value :

[0019] in, Represents the weight vector Transpose of; For a moment The interactive load feature vector; For the first The weights corresponding to each interactive load feature For a moment The One interactive load characteristic value; Interactive load levels are established based on interactive load values, including: normal, load increase, high load, and manual takeover.

[0020] In some embodiments of this application, based on the foregoing scheme, the process by which the control unit controls the interview interaction process according to topic coverage, substantive content, and interaction load includes: If the patient clearly has no content or refuses to answer, set the status of the relevant information slot to "clearly has no content" or "refuses to answer" and stop asking follow-up questions. If the interaction load level is high, pause or postpone the query and do not continue to replenish the information slot status; If the interaction load level is increased, the current number of follow-up questions will be set to the minimum, and simplified questions will only be generated for the highest priority missing slots. If the interaction load level is normal and the topic coverage has not reached the termination condition, and the current number of follow-up questions is less than the maximum number of follow-up questions, then follow-up questions will be asked for the missing information slots. If no substantial information is obtained after reaching the limit of follow-up questions, the node is marked as partially completed or the information slot status is set to fatigue delay, and infinite follow-up questions are not allowed.

[0021] In some embodiments of this application, based on the foregoing scheme, the content generation and security control layer includes: Transformation units are used to convert interaction results into narrative units that constitute a story, are relatively complete in themselves, and play a clear role in the overall narrative structure; The pre-generated filtering unit is used to break down the specific characters, places, events and intentions in the narrative unit into fact units. The reliability of the evidence before generation is determined based on the consistency of the transcription candidates and the reliability of the original audio time positioning. The status of the fact unit is determined by combining the interaction load value to determine whether it is released, pending confirmation or delayed confirmation. The generation unit is used to generate an initial version of the narrative content based on narrative units composed of released or confirmed factual units; The verification unit is used to verify the consistency of the source of the initial narrative content. The safety control unit is used to filter the interaction results based on dangerous words, and to control the interaction process and issue warnings based on the filtering results.

[0022] In some embodiments of this application, based on the foregoing scheme, the process by which the generation unit generates an initial narrative content based on narrative units composed of released or confirmed factual units includes: First, the narrative units composed of the released or confirmed factual units are sorted according to the organization method selected by the patient. Then, the substantive content channel and the transitional text channel are called to generate the initial narrative content. Each generated sentence segment of the initial narrative content saves the source type and source unit list. The substantive content channel only allows the use of narrative units with source_turn_id and original audio / text range to generate content expressing characters, experiences, relationships, feelings, evaluations, and intentions; The transition text channel only generates paragraph connections, pronoun references, word order adjustments, and language polishing; it does not add new information such as people, time, place, events, causes and effects, evaluations, or intentions.

[0023] In some embodiments of this application, based on the foregoing scheme, the process by which the verification unit verifies the consistency of the source of the initial narrative content includes: After the initial narrative content is formed, the source narrative units used for generation and their source interaction fragments are located based on the source unit list recorded for each generated sentence segment, and a source narrative graph is constructed. The draft semantic graph formed by reverse extraction of the initial narrative content is compared with the source narrative graph. Segments that lack source support are marked as unsupported and prevented from entering the final narrative content. After a patient or healthcare worker corrects a fact unit, based on the dependency relationship between the fact unit and sentence segments or video clips, only the affected segments are set to a state pending re-verification.

[0024] In some embodiments of this application, based on the foregoing scheme, the process by which the security control unit filters the interaction results according to dangerous words, and controls the interaction process and issues warnings based on the filtering results includes: The interaction results are filtered based on dangerous words, which include, but are not limited to: words related to general negative emotions, words related to passive death wishes, words related to active self-harm, words related to suicide thoughts, and words related to depression. After filtering out the interaction results that belong to dangerous words, pause the interaction process and determine whether the emergency conditions are met based on the verification fields. If the emergency conditions are met, immediately stop the query and push an emergency event warning. The verification fields include at least: recent and current related thoughts, methods, intentions, specific plans, availability of relevant means, past or recent preparation / implementation behaviors, whether currently alone, whether help is needed, whether consent to medical contact, and authorization of a designated contact person.

[0025] According to a second aspect of the embodiments of this application, an adaptive dignity interview interaction method is provided, applied to the system as described in the first aspect, comprising: Create a dignity interview theme map, and obtain interaction results based on the dignity interview theme map and patient interviews. At the same time, analyze the coverage of interaction content, the substantive content of interaction and patient emotional load during the interaction process, and control the interaction process. The system generates narrative units based on the interaction results, filters the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and performs security verification on the interaction results. The initial version of the narrative content is reviewed, and the resulting narrative content is output, which includes at least: narrative video and narrative text.

[0026] The technical solution of this application has the following beneficial effects: 1. By separating coverage from substantive content and using multiple status information slots, we avoid misidentifying "clearly no content, refused to answer, or forbidden to skip" as incomplete, thereby reducing repeated follow-up questions and status ambiguity.

[0027] 2. By using both timed and event-triggered mechanisms, along with the patient's own baseline deviation and the upper limit of follow-up questions, the complexity of questions and the duration of the conversation can be adjusted according to the current interaction capabilities, and precise recovery after interruption is supported.

[0028] 3. By generating content and transitional text through a dual-channel approach and comparing source semantic graphs, the probability of introducing unrelated information such as people, events, relationships, evaluations, or intentions into the model is reduced.

[0029] 4. By using sentence / segment source tags, audio time anchors, and non-overridable version chains, medical staff can quickly verify the source of generated content and track modifications.

[0030] 5. Through conditional branching automatic verification, task time limits, and emergency bypass, a workflow that is executable and auditable is formed from the identification, review, and contact of security clues to a closed loop.

[0031] 6. Before the initial version of the narrative content is generated, fact units can be screened based on the consistency of the transcription candidates, the reliability of the original audio time positioning, and the interaction load, which can reduce the number of questionable facts entering the initial version of the content; after the initial version of the content is generated, source consistency verification can be performed to identify content without a source added during the generation stage; based on the dependency relationship between fact units and sentences / segments / video clips, the scope of re-verification after fact correction can be narrowed.

[0032] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0033] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 A block diagram of an adaptive dignity interview interaction system according to one embodiment of this application is shown. Detailed Implementation

[0034] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0035] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0037] It should be noted that "multiple" in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0038] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such uses of these terms can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described.

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0040] The following detailed description of some embodiments of this application will be provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0041] See Figure 1 The diagram shows a block diagram of an adaptive dignity interview interaction system according to one embodiment of this application.

[0042] like Figure 1 As shown, the system includes: The access and session layer is used for interviewing and interacting with patients. The interview understanding and decision-making layer is used to create a dignity interview theme map and obtain interaction results based on the dignity interview theme map and patient interviews. At the same time, it analyzes the coverage of interaction content, the substantive content of interaction and patient emotional load during the interaction process and controls the interaction process. The content generation and security control layer is used to generate narrative units based on the interaction results, filter the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and perform security verification on the interaction results. The review and output layer is used to review the initial version of the narrative content and output the narrative content, which includes at least: narrative video and narrative text.

[0043] like Figure 1As shown in this embodiment, the access and session layer includes a patient terminal, which can be a mobile phone or a chatbot. The patient terminal can support voice interaction with the patient, upload information data such as photos, and support human-computer interaction operations.

[0044] During interactions with patients via patient terminals, session and authorization management mechanisms are implemented to protect patient privacy.

[0045] like Figure 1 As shown in this embodiment, the system has a data storage support layer, which is connected to the access and session layer, the interview understanding and decision-making layer, the content generation and security control layer, and the review and output layer, respectively, to store the data generated by each layer in its work.

[0046] The data storage support layer consists of a relational database and an object repository. The relational database is mainly used to store data such as node status, version, tasks, and logs. The object repository is mainly used to store data such as audio, photos, documents, and videos.

[0047] In practical applications, the data storage support layer can be deployed in a hospital's private cloud, dedicated cloud, or cloud environment that meets the requirements for medical data governance.

[0048] In some feasible embodiments, based on the foregoing scheme, the interview understanding and decision-making layer includes: Create a unit to create a dignity interview topic graph based on the set of topic nodes and the optional jump relationships between nodes; wherein each topic node includes at least a node identifier, topic category, question template, set of required information slots, set of optional information slots, entry conditions, taboo marker, maximum number of follow-up questions, current status and source record; Each information slot can be set to any of the following states: not asked, asked and awaiting response, obtained substantive content, clearly no content, refused to answer, fatigued and postponed, forbidden to skip, transferred to human, terminated; The processing unit is used to perform voice / text processing on the interaction results of patient interviews; The first analysis unit is used to analyze the topic coverage and substantive content of the interaction results based on the status of the information slots. The second analysis unit is used to analyze the patient's fatigue during the interview and calculate the interaction load value; The control unit is used to control the interview interaction process based on topic coverage, substantive content, and interaction load.

[0049] In some feasible embodiments, based on the aforementioned scheme, the specific process of creating a dignity interview theme map by the creation unit includes: Establish a dignity interview theme graph G=(V,E). Where V= Let E be the set of topic nodes, and E represent the optional navigation relationships between nodes. Each node It should include at least node identifier, topic category, question template, set of required information slots, set of optional information slots, entry conditions, taboo marker, maximum number of follow-up questions, current status, and source record.

[0050] Each message slot takes one of the following states: U (not asked), A (asked and awaiting response), F (substantial content received), N (no content specified), R (refusal to answer), D (fatigue and delay), B (taboo skip), H (transfer to human operator), or X (terminated). F, N, R, B, H, and X are terminating states; D is a paused state in the current session but can be restored to A in a subsequent session.

[0051] In some feasible embodiments, based on the aforementioned scheme, the specific process by which the first analysis unit analyzes the topic coverage and substantive content of the interaction results according to the information slot status includes: Obtain the status of the information slots and calculate the topic coverage and substantive content. The formula for calculating topic coverage is as follows:

[0052] in, For topic coverage, For the first The first topic Individual information slot status, For indicator functions, To cover weights, This is a set of terminating states, which include: obtaining substantive content, clearly no content, refusing to answer, taboo skipping, transferring to human agent, and termination. The formula for calculating the content quality is as follows:

[0053] in, For the sake of substantive content, Weighting based on substantive content. For indicator functions, For the first The first topic Individual information slot status, To obtain the state of the substantive content. When hour, ,otherwise .

[0054] It should be noted that, in this embodiment, the topic coverage... Used to determine whether all required slots have been properly queried and processed; Substantial content level This is used to select whether to ask follow-up questions regarding missing substantive information. When the patient explicitly states "no," "can't think of," or "unwilling to answer," the corresponding slot is counted for coverage, but not for substantive content.

[0055] In some feasible embodiments, based on the foregoing scheme, the specific process by which the second analysis unit analyzes the patient's fatigue during the interview and calculates the interaction load value includes: At the start of the interview, the patient's self-reported fatigue score was obtained on a scale of 0-10, and the fatigue status was asked again when the preset time interval was reached. In addition to timed triggering, fatigue inquiry can be triggered in advance when the patient actively clicks fatigue / pause, does not respond for the first time, pauses significantly increase, response length continues to decrease, or other behavioral abnormalities occur; After obtaining several rounds of valid responses, establish individual patient baselines, including baseline pause duration, baseline response length, and optional baseline speech rate; Calculation time Interactive load value :

[0056] in, Represents the weight vector Transpose of; For a moment The interactive load feature vector; For the first The weights corresponding to each interactive load feature For a moment The One interactive load characteristic value; Interactive load levels are established based on interactive load values, including: normal, load increase, high load, and manual takeover.

[0057] In some feasible embodiments, based on the foregoing scheme, the interactive load feature vector is represented as:

[0058] in, Normalized results of patients' self-reported fatigue scores in the range of 0 to 10; This is a normalized result of the duration of consecutive interviews; This indicates the degree to which the current pause duration deviates from the individual's baseline. This represents the degree to which the current response length has decreased relative to the individual's baseline. This is the normalized result of the number of consecutive no-response events; These are optional emotional load features.

[0059] Normalized to each interactive load characteristic value The weights of each feature satisfy ,and

[0060] therefore, The range of values ​​is .

[0061] It should be noted that, in this embodiment, the preset time interval can be 10-30 minutes, preferably 20 minutes. In addition to being triggered at preset time intervals, the system will inquire about fatigue status in advance when the patient actively clicks "fatigue / pause," shows no response for the first time, experiences a significant increase in pauses, a continuous decrease in response length, or other abnormal behaviors.

[0062] It should be noted that, in this embodiment, the interaction load levels are specifically divided as follows: Characteristic values ​​of each interactive load Normalized to Weights of each feature It is a non-negative number and satisfies Therefore, the interactive load value The range of values ​​is .

[0063] Based on interaction load value Establish interactive load levels: When 0 This is considered normal. when At that time, it was determined to be an increase in load; when At that time, it was determined to be a high load; when At that time, it was determined to be a case of manual intervention.

[0064] in, These thresholds can be set to 0.40, 0.70, and 0.85 respectively; the above thresholds can be adjusted according to the target audience, interview scenario, and historical interaction data.

[0065] If the patient actively requests to terminate the interview or switch to a human interview, or if the safety control unit determines that the emergency conditions are met, the above-mentioned interaction load value range will not apply, and the system will directly enter the human takeover state.

[0066] In some feasible embodiments, based on the foregoing scheme, the process by which the control unit controls the interview interaction process according to topic coverage, substantive content, and interaction load includes: If the patient clearly has no content or refuses to answer, set the status of the relevant information slot to "clearly has no content" or "refuses to answer" and stop asking follow-up questions. If the interaction load level is high, pause or postpone the query and do not continue to replenish the information slot status; If the interaction load level is increased, the current number of follow-up questions will be set to the minimum, and simplified questions will only be generated for the highest priority missing slots. If the interaction load level is normal and the topic coverage has not reached the termination condition, and the current number of follow-up questions is less than the maximum number of follow-up questions, then follow-up questions will be asked for the missing information slots. If no substantial information is obtained after reaching the limit of follow-up questions, the node is marked as partially completed or the information slot status is set to fatigue delay, and infinite follow-up questions are not allowed.

[0067] When the interaction load level is normal, continue the interaction according to the current interview strategy; when the interaction load level is high, reduce the number of follow-up questions and only generate simplified questions for high-priority missing information slots; when the interaction load level is high, pause or postpone the current interview; when entering manual takeover mode, interrupt the automatic interview and push the interview record, load characteristics and trigger reasons to medical staff for processing.

[0068] Among them, thematic coverage The termination conditions are as follows: For the first... A set of topic nodes, where all required information slots of that topic node are in a terminating state. When the topic node is determined to meet the termination condition, that is... Stop further inquiry into this topic node and proceed to the next available topic node according to the optional jump relationships in the topic graph. This includes the set of terminating states.

[0069] These correspond to obtaining substantive content, explicitly stating no content, refusing to answer, prohibiting skipping, transferring to human assistant, and termination, respectively.

[0070] If any required information slot is still in the state of not queried (U), queried and awaiting response (A), or fatigued and delayed (D), it is determined that the topic node has not yet met the coverage termination condition. Among them, fatigued and delayed (D) means that the current session is temporarily suspended, and the status of the information slot will be judged again after the subsequent session resumes.

[0071] In some feasible embodiments, based on the foregoing scheme, the content generation and security control layer includes: Transformation units are used to convert interaction results into narrative units that constitute a story, are relatively complete in themselves, and play a clear role in the overall narrative structure; The pre-generated filtering unit is used to break down the specific characters, places, events and intentions in the narrative unit into fact units. The reliability of the evidence before generation is determined based on the consistency of the transcription candidates and the reliability of the original audio time positioning. The status of the fact unit is determined by combining the interaction load value to determine whether it is released, pending confirmation or delayed confirmation. The generation unit is used to generate an initial version of the narrative content based on narrative units composed of released or confirmed factual units; The verification unit is used to verify the consistency of the source of the initial narrative content. The safety control unit is used to filter the interaction results based on dangerous words, and to control the interaction process and issue warnings based on the filtering results.

[0072] Understandably, in this embodiment, after the conversion unit forms a narrative unit, the pre-generation screening unit first calculates the pre-generation evidence reliability of the fact units in the narrative unit and arranges confirmation based on the patient interaction load value. The generation unit only uses the released or confirmed fact units to form the initial version of the narrative content. The verification unit then constructs a source narrative graph based on the source unit list saved in the initial version of the segment, and compares the initial version of the narrative content with the source narrative graph; segments that fail the source consistency verification are not included in the final text or video. When the patient or medical staff corrects a fact unit, only the output segment that depends on that fact is re-verified.

[0073] It should be noted that in this embodiment, after the interview phase ends, the system converts the patient's answers into structured narrative units. Each narrative unit includes at least: unit_id, patient_id, session_id, node_id, source_turn_id, audio_start, audio_end, transcript_start, transcript_end, person entity, time entity, location entity, relationship, event, evaluation / meaning, intention expression, source confidence, patient confirmation status, and medical staff review status. Fact unit records are created under each narrative unit, storing at least fact_id, source_turn_id, audio_start, audio_end, transcription candidates and their scores, the character range corresponding to the transcription candidates, fact type, information slot, pre-generation evidence reliability, importance, estimated confirmation cost, and confirmation status. After generating the initial version of the content, the corresponding sentence segment identifiers and source narrative graph node or edge identifiers are backfilled; after the video is generated, video segment identifiers are backfilled. The confirmation status includes at least automatic release, pending patient confirmation, fatigue delay, refusal to confirm, medical staff confirmation, and confirmation failure; refusal to confirm does not automatically indicate that the fact is false, and specific experiences that have not been confirmed will not be included in the initial version of the content as confirmed facts.

[0074] In addition, this embodiment also anchors the source of each narrative unit. For example, "wishes" and "messages" may overlap in natural language. The system does not force mutually exclusive classifications, but instead models them uniformly as "intentional expression units," which can be attached with one or more auxiliary tags such as blessings, suggestions, entrustments, thanks, apologies, arrangements, or wishes to be remembered. Tags are only used for organization and retrieval and do not determine the validity of the content.

[0075] In some feasible embodiments, the specific process of pre-generating the filtering unit for filtering includes: For a fact unit u intended to be used to generate the initial version of the content, calculate A(u) and T(u). A(u) represents the degree of consistency of the corresponding person's name, place name, event, or expression of intent among multiple speech-to-text candidates; T(u) represents the location reliability of the character range of the fact unit relative to the time range of the original audio. Both are normalized to 0 to 1, and the pre-generation evidence reliability E(u) is the smaller of the two values. E(u) is only used to identify the input facts that need to be verified, without using the initial version of the content or the source narrative graph.

[0076]

[0077] Assign importance H(u) based on fact type, whether it contains specific names or locations, and whether it involves patient intent; assign confirmation cost C(u) based on expected short question length and answer complexity, where C(u) is a positive number. Let L(t) be the current interaction load value and B0 be the preset maximum confirmation load per round, thus obtaining the confirmation budget B(t) for this round. For fact units with E(u) below the threshold, calculate the confirmation priority Q(u), and prioritize confirming those with higher Q(u) if the sum of candidate confirmation costs does not exceed B(t). Fact units with E(u) reaching the threshold or obtaining confirmation from patients or medical staff are marked as approved; the remaining fact units are not used to generate the initial version of the content.

[0078]

[0079] When a patient explicitly refuses to answer, no confirmation question is generated, and the refusal status is retained. When the interaction load is too high or the current confirmation budget is insufficient, questionable fact units are marked as fatigue-delayed, and the confirmation point is saved. When the load allows, at most one short question involving only one fact is issued per round. After the patient or medical staff completes the confirmation based on the original audio, the confirmation subject, time, and audio range on which it is based are recorded. The generated unit only reads the fact units that have been approved; fact units that are pending confirmation, fatigue-delayed, refused confirmation, or failed confirmation are not included as confirmed facts in the initial narrative content.

[0080] In some feasible embodiments, based on the foregoing scheme, the process by which the generation unit generates an initial version of the narrative content according to the narrative units composed of released or confirmed factual units includes: First, the narrative units composed of the released or confirmed factual units are sorted according to the organization method selected by the patient. Then, the substantive content channel and the transitional text channel are called to generate the initial narrative content. Each generated sentence segment of the initial narrative content saves the source type and source unit list. The substantive content channel only allows the use of narrative units with source_turn_id and original audio / text range to generate content expressing characters, experiences, relationships, feelings, evaluations, and intentions; The transition text channel only generates paragraph connections, pronoun references, word order adjustments, and language polishing; it does not add new information such as people, time, place, events, causes and effects, evaluations, or intentions.

[0081] In some feasible embodiments, based on the aforementioned scheme, the process by which the verification unit verifies the consistency of the source of the initial narrative content includes: After the initial narrative content is formed, the source narrative units used for generation and their source interaction fragments are located based on the source unit list recorded for each generated sentence segment, and a source narrative graph is constructed. The draft semantic graph formed by reverse extraction of the initial narrative content is compared with the source narrative graph. Segments lacking source support are marked as unsupported and prevented from entering the final narrative content. After the patient or medical staff corrects the fact unit, based on the dependency relationship between the fact unit and the sentence / segment and video clip, only the affected clip is set to a state of pending re-verification.

[0082] It should be noted that, in this embodiment, the generation process of the source narrative graph specifically includes: For each generated sentence segment in the initial narrative content, the source type and source unit list saved in the generated sentence segment are read. When the source type is a substantial content channel source, the corresponding narrative unit is obtained according to the source unit identifier in the source unit list. The corresponding source interaction segment is located according to the source_turn_id carried by each narrative unit and the original audio / text range. Using the same entity category, event category, relationship category and standardization rules as the draft semantic graph, entity extraction, event extraction, relationship extraction and reference resolution are performed on the source interaction segment. The extracted entities and events are constructed as nodes, and the semantic relationships between entities, between entities and events, and between events are constructed as edges. The source unit identifier, source_turn_id, original audio / text range and source type are written into the source attribute of the corresponding node or edge. Entity nodes pointing to the same object and event nodes representing the same experience are merged, and all source attributes of each node or edge before merging are retained. The nodes and edges corresponding to each source unit are merged to form a source narrative graph. When the source type is a transitional text channel source, nodes or edges are not created based on the transitional text to support substantive information such as people, time, place, events, causes and effects, evaluations, or intentions.

[0083] For example, a source narrative graph is represented as ,in, It is a collection of entity nodes and event nodes. Entity nodes include at least people, time, location and object nodes, and event nodes include at least experience, behavior and intention event nodes. It is a set of semantic relationship edges, including at least the relationships between people, the relationships between participants in an event, the relationships between the time of occurrence, the relationships between the location of occurrence, the relationships between feelings, the relationships between evaluations, the relationships between intentions, and explicit causal relationships. This is a set of source attributes used to record the source unit identifier, source round identifier, audio start and end time or text start and end position, and source type for each node or edge. Only entities, events, or relationships that can be located to the source interaction segment are written into the source narrative graph.

[0084] In some feasible embodiments, based on the aforementioned scheme, the process by which the security control unit filters the interaction results according to dangerous words, and controls the interaction process and issues warnings based on the filtering results includes: The interaction results are filtered based on dangerous words, which include, but are not limited to: words related to general negative emotions, words related to passive death wishes, words related to active self-harm, words related to suicide thoughts, and words related to depression. After filtering out the interaction results that belong to dangerous words, pause the interaction process and determine whether the emergency conditions are met based on the verification fields. If the emergency conditions are met, immediately stop the query and push an emergency event warning. The verification fields include at least: recent and current related thoughts, methods, intentions, specific plans, availability of relevant means, past or recent preparation / implementation behaviors, whether currently alone, whether help is needed, whether consent to medical contact, and authorization of a designated contact person.

[0085] It should be noted that, in this embodiment, the specific process for determining whether the emergency condition is met includes: After filtering out the interaction results that belong to dangerous words, pause the current interview and obtain the verification fields in sequence, including whether there are self-harm or suicide-related thoughts at present or recently, whether there is an intention to carry them out, whether a specific method or plan has been formed, whether the relevant means are available, and whether there has been any preparation or implementation behavior recently.

[0086] The determination of whether the emergency conditions are met based on the verification fields includes: (1) The patient currently has self-harm or suicidal thoughts and clearly intends to carry them out, and there are specific methods, plans or means available to carry them out; (2) The patient has recently engaged in self-harm or suicidal behavior, or has exhibited behaviors that prepare for, attempt, or rehearse such behavior; (3) The patient indicates that he or she is about to commit self-harm or suicide, or is currently in the process of doing so.

[0087] If any of the above conditions are met, an emergency condition is determined, the automatic interview is immediately interrupted, and the risk verification results and corresponding source interaction content are pushed to medical staff for emergency warning and manual takeover; if the above emergency conditions are not met, a corresponding risk warning is generated based on whether the patient is currently alone, whether they need help, whether they agree to medical staff contact, and the authorization status of the designated contact person, and the case is transferred to medical staff for further review.

[0088] It should be noted that in this embodiment, the generated initial narrative content includes the original audio and written text. The original audio and written text, as the initial version V0, cannot be overwritten by subsequent versions. The original audio and written text V0 are sequentially processed through the review and output layers to form the AI ​​initial draft V1, the medical staff review draft V2, the patient confirmation draft V3, and the final legacy document V4. The data storage support layer stores the subject, time, differences, and review results of each addition, deletion, and modification operation. Medical staff can click on the substantive information segments to listen back to the corresponding audio, view the AI ​​transition content, delete sensitive content, or revert to regeneration.

[0089] Patients can confirm, delete, supplement, or modify the draft reviewed by healthcare professionals. The modified draft becomes Patient Confirmation Draft V3; content requiring healthcare professional review can be returned to V2. Patient confirmation is a preferred and necessary step in creating the final legacy document V4. Whether proxy confirmation is allowed when the patient cannot confirm should be configured separately according to institutional policy and specific research / service plans.

[0090] When the patient explicitly chooses to generate a video, the system retrieves the final inheritance document V4 (confirmed by the patient), the patient-selected static photo, and a generic AI-synthesized narration to generate video V5. The system does not clone the patient's voice; the patient can choose to generate only text. Before outputting video V5, the system re-checks the text version number and the patient's authorization status to ensure the video uses only confirmed content. The final video is delivered to the patient and their designated family member. During video V5 generation, the system saves the identifier and confirmed version of the factual unit upon which each narration segment and corresponding video clip depends; only segments that have passed pre-generation screening and post-generation source consistency checks are used. If a relevant factual unit is later corrected or fails to be confirmed, the segments and video clips that depend on that fact are marked as invalid. The system only regenerates the affected clips and releases a new version after successful re-verification.

[0091] Based on the same inventive concept, this application also provides an adaptive dignity interview interaction method, applied to the system described in any of the above embodiments, the method comprising: Step S100: Create a dignity interview theme map, and obtain the interaction results by interacting with the patient interview based on the dignity interview theme map. At the same time, analyze the coverage of the interaction content, the substantive content of the interaction, and the patient's emotional load during the interaction process, and control the interaction process. Step S200: Generate narrative units based on the interaction results, filter the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and verify the security of the interaction results. Step S300: Review the initial version of the narrative content and output the narrative content, which includes at least: narrative video and narrative text.

[0092] Below is a specific example.

[0093] In this example, a home-based mobile app was used to conduct dignity interviews.

[0094] The specific process includes: Step A1: The patient logs in using the mobile app, reads the options for data processing, AI participation, medical staff review, family contact, and video generation authorization, and makes a selection for each.

[0095] Step A2: Import patient basic information, disease / nursing information, previous assessments, historical interviews, and pre-set contraindications topics into the system, and create a session_id.

[0096] Step A3: The patient completes a 0-10 point fatigue self-assessment and selects to begin the dignity interview. The system outputs a voice question from the first non-contraindicated and unterminated topic node.

[0097] Step A4: After the patient responds with their voice, the system saves the original audio and timestamp and generates a transcription; the parser fills the information slots and calculates the topic coverage and substantive content.

[0098] Step A5: The system establishes a baseline for individual pause and response length, and asks about fatigue again every 20 minutes or when an event is triggered.

[0099] Step A6: Under normal circumstances, each topic can be followed up a maximum of 3 times; when the load increases, the questions are simplified and followed up a maximum of 1 time; when the load is high, the questioning is paused and saved.

[0100] Step A7: After all core themes enter the termination state such as F, N, R, B, H or X, the system generates structured narrative units and initial narrative content.

[0101] Step A8: Medical staff review the source and content, and the final draft is generated after the patient confirms the revisions; when the patient selects, a static photo and a general narration video are generated.

[0102] In another implementation, patients complete the aforementioned process using a chatbot equipped with a microphone, speaker, and display screen, instead of a mobile phone. The chatbot acts only as an input / output terminal; topic status, interaction load, generation verification, and version control are performed on the server side. When a patient experiences high load or a security incident, the chatbot displays a neutral message such as "Your request has been paused and we have contacted medical staff," and provides an entry point for human assistance.

[0103] Based on the above examples, the following is an example of fatigue adaptation and session recovery, including: At the 20-minute mark of the interview, the patient reported a fatigue score of 7, and the average length of their responses in the last three rounds had decreased compared to their baseline, while their pauses had increased. The system set the interaction load to high, stopped supplementing information slots, and asked, "Would you like to pause now and continue next time, or end today's interview?" The patient chose to pause. The system set the current node to D (fatigue delayed), and saved the missing slots, the number of follow-up questions, the last audio position, and the patient's fatigue score.

[0104] After the patient logs in the next day, the system first retrieves the fatigue value, reads the D status and prompts, "We paused here last time. Would you like to continue this topic today?" After the patient confirms, D is restored to A (inquired and awaiting response). The system only inquires about unprocessed slots and does not repeat completed content.

[0105] Based on the above example, the following is an example of automatic verification of security leads and review by medical staff, specifically including: The patient stated, "There's no point in living." This original sentence was marked as a candidate for passive death wish, and the general dignity interview was paused while automatic verification was initiated. The system sequentially asked whether the patient recently wished not to wake up again, whether they had thought about harming or ending their life, and whether these were current thoughts; only if the preceding positive result was confirmed did the system continue to ask about methods, intentions, plans, means, past behaviors, and current safety.

[0106] If the patient denies any active thoughts, plans, or behaviors, but can confirm current safety, the system generates a Level 1 emotional safety cue task and suggests a healthcare review within 12 hours. If the patient admits to recent active thoughts but denies current intentions, or provides vague key answers, the system generates a Level 2 task. If the patient confirms current thoughts and has a plan or available means, the system interrupts the remaining questions and immediately generates an emergency safety event.

[0107] Medical staff can review the trigger message, audio clips, scale information, and verification results in the background to confirm or adjust the task level. After confirming that family members need to be notified, the system reads the pre-authorization and allows medical staff to choose between a phone call or SMS. The system records the time of task receipt, initial contact, family notification, professional assessment, referral, and closure.

[0108] Based on the above examples, the following is an example of generating inheritance text with source constraints, specifically including: The system generates narrative units from patient responses: U1: "I worked in another city when I was young"; U2: "This experience taught me independence"; U3: "I hope my children won't give up easily when they encounter difficulties." Each unit saves the corresponding original audio time range.

[0109] The substantive content channel generates main sentence segments based on U1-U3; the transitional text channel can generate connecting phrases such as "looking back on those years," but must not add specific cities, colleague names, reasons for events, or family evaluations not expressed by the patient. If the initial draft contains "working in City A" but the source only shows "outside the city," the reverse extraction module identifies the newly added location entity "City A," marks it as unsupported, and triggers regeneration or medical review.

[0110] The substantive content channel generates main sentence segments based on U1-U3; the transitional text channel can generate connecting phrases such as "looking back on those years," but must not add specific cities, colleague names, reasons for events, or family evaluations not expressed by the patient. If the initial draft contains "working in City A" but the source only contains "outside the city," the reverse extraction module identifies the newly added location entity "City A," marks it as unsupported, and triggers regeneration or medical staff review. If, before generating the initial version of the content, the transcription candidates for the same original audio segment are "outside the city" and "City A," the pre-generation screening unit calculates the consistency of the transcription candidates and the reliability of the audio time location for the "work location" fact unit. If the reliability of the evidence before the generation of this fact unit is insufficient, if the patient's workload is normal, only one question is asked: "Did you just say 'outside the city' or 'City A'?"; if the patient's workload is high, confirmation is delayed or the original audio is listened to by medical staff. Before confirmation, the generation unit does not include "working in City A" or "working outside the city" as a confirmed fact in the initial version of the content. If the source fact that is approved after screening is 'outside the city', but the initial version of the generated content is written as 'working in City A', then in the generated S3, the draft semantic graph and the source narrative graph will be compared to find that the location lacks source support, and it will be marked as not supported and regenerated or submitted to medical staff for review.

[0111] Based on the above example, the following is an example of review and video generation, specifically including: In the initial draft V1, healthcare staff see patient-sourced sentences and AI transition sentence markers, and can click on the source sentences to listen to the audio again. Healthcare staff then remove sensitive paragraphs that might offend family or friends and modify the salutations, creating the healthcare staff review draft V2. The patient adds a thank-you note to the healthcare staff review draft V2, creating the patient confirmation draft V3. The system records the source type of each change. After the patient confirms the patient confirmation draft V3, the final legacy document V4 is created, and all backend tags are hidden in the patient document.

[0112] When a patient selects to generate a video, the system retrieves the final legacy document V4, the patient's selected still photo, and a generic composite narration. If the final legacy document V4 is subsequently modified by the patient, the original video is marked as an older version and must be regenerated based on the new version to avoid inconsistencies between the video and the final document.

[0113] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An adaptive dignity interview interaction system, characterized in that, include: The access and session layer is used for interviewing and interacting with patients. The interview understanding and decision-making layer is used to create a dignity interview theme map and obtain interaction results based on the dignity interview theme map and patient interviews. At the same time, it analyzes the coverage of interaction content, the substantive content of interaction and patient emotional load during the interaction process and controls the interaction process. The content generation and security control layer is used to generate narrative units based on the interaction results, filter the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and perform security verification on the interaction results. The review and output layer is used to review the initial version of the narrative content and output the narrative content, which includes at least: narrative video and narrative text.

2. The system according to claim 1, characterized in that, The interview understanding and decision-making level includes: Create a unit to create a dignity interview topic graph based on the set of topic nodes and the optional jump relationships between nodes; wherein each topic node includes at least a node identifier, topic category, question template, set of required information slots, set of optional information slots, entry conditions, taboo marker, maximum number of follow-up questions, current status and source record; Each information slot can be set to any of the following states: not asked, asked and awaiting response, obtained substantive content, clearly no content, refused to answer, fatigued and postponed, forbidden to skip, transferred to human, terminated; The processing unit is used to perform voice / text processing on the interaction results of patient interviews; The first analysis unit is used to analyze the topic coverage and substantive content of the interaction results based on the status of the information slots. The second analysis unit is used to analyze the patient's fatigue during the interview and calculate the interaction load value; The control unit is used to control the interview interaction process based on topic coverage, substantive content, and interaction load.

3. The system according to claim 2, characterized in that, The formula for calculating topic coverage is as follows: in, For topic coverage, For the first The first topic Individual information slot status, For indicator functions, To cover weights, This is a set of terminating states, which include: obtaining substantive content, clearly no content, refusing to answer, taboo skipping, transferring to human agent, and termination. The formula for calculating the content quality is as follows: in, For the sake of substantive content, Weighting based on substantive content. For indicator functions, For the first The first topic Individual information slot status, To obtain the state of the substantive content.

4. The system according to claim 2, characterized in that, The specific process by which the second analysis unit analyzes the patient's fatigue during the interview and calculates the interaction load value includes: At the start of the interview, the patient's self-reported fatigue score was obtained on a scale of 0-10, and the fatigue status was asked again when the preset time interval was reached. In addition to timed triggering, fatigue inquiry can be triggered in advance when the patient actively clicks fatigue / pause, does not respond for the first time, has a significant increase in pauses, has a continuous decrease in the length of the answer, or other abnormal behaviors occur; After obtaining several rounds of valid responses, establish individual patient baselines, including baseline pause duration, baseline response length, and optional baseline speech rate; Calculation time Interactive load value : in, Represents the weight vector transpose; For a moment The interactive load feature vector; For the first The weights corresponding to each interactive load feature For a moment The One interactive load characteristic value; Interactive load levels are established based on interactive load values, including: normal, load increase, high load, and manual takeover.

5. The system according to claim 4, characterized in that, The process by which the control unit controls the interview interaction process based on topic coverage, substantive content, and interaction load includes: If the patient clearly has no content or refuses to answer, set the status of the relevant information slot to "clearly has no content" or "refuses to answer" and stop asking follow-up questions. If the interaction load level is high, pause or postpone the query and do not continue to supplement the information slot status; If the interaction load level is increased, the current number of follow-up questions will be set to the minimum, and simplified questions will only be generated for the highest priority missing slots. If the interaction load level is normal and the topic coverage has not reached the termination condition, and the current number of follow-up questions is less than the maximum number of follow-up questions, then follow-up questions will be asked for the missing information slots. If no substantial information is obtained after reaching the limit of follow-up questions, the node is marked as partially completed or the information slot status is set to fatigue delay, and infinite follow-up questions are not allowed.

6. The system according to claim 5, characterized in that, The content generation and security control layer includes: Transformation units are used to convert interaction results into narrative units that constitute a story, are relatively complete in themselves, and play a clear role in the overall narrative structure; The pre-generated filtering unit is used to break down the specific characters, places, events and intentions in the narrative unit into fact units. The reliability of the evidence before generation is determined based on the consistency of the transcription candidates and the reliability of the original audio time positioning. The status of the fact unit is determined by combining the interaction load value to determine whether it is released, pending confirmation or delayed confirmation. The generation unit is used to generate an initial version of the narrative content based on narrative units composed of released or confirmed factual units; The verification unit is used to verify the consistency of the source of the initial narrative content. The safety control unit is used to filter the interaction results based on dangerous words, and to control the interaction process and issue warnings based on the filtering results.

7. The system according to claim 6, characterized in that, The process by which the generation unit generates the initial narrative content based on narrative units composed of released or confirmed factual units includes: First, the narrative units composed of the released or confirmed factual units are sorted according to the organization method selected by the patient. Then, the substantive content channel and the transitional text channel are called to generate the initial narrative content. Each generated sentence segment of the initial narrative content saves the source type and source unit list. The substantive content channel only allows the use of narrative units with source_turn_id and original audio / text range to generate content expressing characters, experiences, relationships, feelings, evaluations, and intentions; The transition text channel only generates paragraph connections, pronoun references, word order adjustments, and language polishing; it does not add new information about people, times, places, events, causes and effects, evaluations, or intentions.

8. The system according to claim 7, characterized in that, The process by which the verification unit verifies the consistency of the source of the initial narrative content includes: After the initial narrative content is formed, the source narrative units used for generation and their source interaction fragments are located based on the source unit list recorded for each generated sentence segment, and a source narrative graph is constructed. The draft semantic graph formed by reverse extraction of the initial narrative content is compared with the source narrative graph. Segments lacking source support are marked as unsupported and prevented from entering the final narrative content. After the patient or medical staff corrects the fact unit, based on the dependency relationship between the fact unit and the sentence / segment and video clip, only the affected clip is set to a state of pending re-verification.

9. The system according to claim 6, characterized in that, The process by which the safety control unit filters the interaction results based on dangerous words, and controls the interaction process and issues warnings based on the filtering results includes: The interaction results are filtered based on dangerous words, which include, but are not limited to: words related to general negative emotions, words related to passive death wishes, words related to active self-harm, words related to suicide thoughts, and words related to depression. After filtering out the interaction results that belong to dangerous words, pause the interaction process and determine whether the emergency conditions are met based on the verification fields. If the emergency conditions are met, immediately stop the query and push an emergency event warning. The verification fields include at least: recent and current related thoughts, methods, intentions, specific plans, availability of relevant means, past or recent preparation / implementation behaviors, whether currently alone, whether help is needed, whether consent to medical contact, and authorization of a designated contact person.

10. An adaptive dignity interview interaction method, applied to the system as described in any one of claims 1-9, characterized in that, include: Create a dignity interview theme map, and obtain interaction results based on the dignity interview theme map and patient interviews. At the same time, analyze the coverage of interaction content, the substantive content of interaction and patient emotional load during the interaction process, and control the interaction process. The system generates narrative units based on the interaction results, filters the narrative units based on the patient's emotional load to generate an initial version of the narrative content, and performs security verification on the interaction results. The initial version of the narrative content is reviewed, and the resulting narrative content is output, which includes at least: narrative video and narrative text.

Images