Psychological crisis intervention auxiliary method, device and equipment and storage medium

Abstract

The invention provides a psychological crisis intervention auxiliary method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence and psychological health education crossing. The method comprises the following steps: acquiring psychological interaction information of a target; calling a large language model subjected to fine adjustment of the target psychological corpus, performing semantic analysis on the psychological interaction information, identifying an emotion type and extracting a psychological crisis keyword; according to the number of the psychological crisis keywords, calculating a psychological crisis risk value of the current target; executing an intervention strategy on the current target according to the psychological crisis risk value of the current target; and collecting subsequent interaction information of the target within a preset time after the execution of the intervention strategy, substituting the subsequent interaction information into an intervention effect formula to calculate an intervention effect improvement rate, and if the intervention effect improvement rate is less than or equal to 0, adjusting the intervention strategy. According to the method, real-time semantic analysis, quantitative risk assessment and dynamic intervention strategy generation of the target psychological state are achieved through the large language model, and the efficiency and accuracy problems of a traditional intervention mode are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and mental health education, in particular to a psychological crisis intervention auxiliary method, device, equipment and storage medium. BACKGROUND

[0002] The current psychological crisis intervention work for college students faces many challenges: on the one hand, traditional manual intervention relies on psychological counselors to actively discover and students to actively seek help, which has problems of crisis recognition lag and limited coverage; on the other hand, the existing mental health assessment mostly uses scale research, which has the disadvantages of poor timeliness and single quantitative dimension. Large language model (LLM) provides new possibilities for early identification and dynamic intervention of psychological crisis with its powerful natural language processing and semantic understanding ability, but there is a lack of special system for college students in the prior art that integrates quantitative assessment logic and intelligent intervention process, which is difficult to accurately match the individualized and timely needs of college students' psychological crisis intervention. SUMMARY

[0003] The present application provides a psychological crisis intervention auxiliary method, device, equipment and storage medium, which realizes real-time semantic analysis, quantitative risk assessment and dynamic intervention strategy generation of college students' psychological state through a large language model, solving the efficiency and accuracy problems of traditional intervention methods.

[0004] In a first aspect, the present application provides a psychological crisis intervention auxiliary method, comprising: obtaining psychological interaction information of a target, if the psychological interaction information is voice information, converting it into text through voice recognition technology; calling a large language model fine-tuned by target psychological corpus, performing semantic analysis on the psychological interaction information, identifying emotion types and extracting psychological crisis keywords; calculating a current target psychological crisis risk value according to the number of psychological crisis keywords; executing an intervention strategy for the current target according to the current target psychological crisis risk value; collecting subsequent interaction information of the target within a preset time after the intervention strategy is executed, calculating an intervention effect improvement rate by substituting the intervention effect formula, and adjusting the intervention strategy if the intervention effect improvement rate is less than or equal to 0.

[0005] In a possible design, the target psychological corpus includes historical counseling records of a psychological counseling center, target group psychological theme forum texts, and psychological course learning discussion data; the large language model is adapted to target psychological expression habits and common psychological problem scenarios through supervised fine-tuning; the output result of the semantic analysis includes a list of emotion types with intensity scores and a set of psychological crisis keywords.

[0006] In one possible design, the psychological crisis risk value of the current target is calculated according to the number of psychological crisis keywords in the interactive text by the following formula: , wherein R is the psychological crisis risk value, and the value range is [0, 10], the greater the value, the higher the risk, is the number of psychological crisis keywords in the interactive text, is the total number of sentences in the interactive text, is the intensity score of the first i negative emotion, which is output by a large language model based on semantic sentiment analysis, and the value range is [0, 5], n is the number of recognized negative emotions, T is the number of times that the proportion of negative emotions in the recent continuous interaction exceeds the threshold value, α , β , gamma is a weight coefficient, and satisfies α + β + gamma =1.

[0007] In one possible design, an intervention strategy is performed on the current target according to the psychological crisis risk value of the current target, including: When R<3, the intervention strategy is to generate a light-weight soothing phrase and push a mental health popular science article; When 3≤R<7, the intervention strategy is to generate a deep soothing dialogue guide and send a concern reminder to the attention person associated with the target; When R≥7, the intervention strategy is to generate an emergency intervention phrase and trigger the emergency intervention process of the mental health on-duty personnel.

[0008] In one possible design, the intervention effect formula is: , wherein is the intervention effect improvement rate, E >0 indicates that the intervention is effective, E the greater the value, the better the effect, is the number of positive emotion sentences before intervention, is the number of positive emotion sentences after intervention.

[0009] In one possible design, the preset time is 1 week, and when the intervention effect improvement rate is less than or equal to 0, the intervention strategy is adjusted by the following method: re-performing semantic analysis on subsequent interactive information, and adjusting the weight coefficient α , β , gammare-calculate the psychological crisis risk value, and adjust the intervention strategy according to the re-calculated psychological crisis risk value.

[0010] In a possible design, when the intervention effect improvement rate is greater than 0, the current intervention strategy is maintained or adjusted to a light tracking strategy; the light tracking strategy includes regularly collecting short psychological state feedback texts of the target and performing semantic analysis to calculate a psychological crisis risk value.

[0011] In a second aspect, the present application provides a psychological crisis intervention auxiliary device, the device comprising: a data acquisition unit configured to acquire psychological interaction information of a target, and convert the psychological interaction information into text through voice recognition technology if the psychological interaction information is voice information; a keyword extraction unit configured to call a large language model fine-tuned by a target psychological corpus, perform semantic analysis on the psychological interaction information, identify an emotion type, and extract psychological crisis keywords; a risk value calculation unit configured to calculate a psychological crisis risk value of a current target according to a number of the psychological crisis keywords; an intervention strategy execution unit configured to execute an intervention strategy on the current target according to the psychological crisis risk value of the current target; an intervention strategy adjustment unit configured to collect subsequent interaction information of the target within a preset time after the intervention strategy is executed, calculate an intervention effect improvement rate by substituting the subsequent interaction information into an intervention effect formula, and adjust the intervention strategy if the intervention effect improvement rate is less than or equal to 0.

[0012] In a third aspect, an embodiment of the present application provides an electronic device, comprising at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the psychological crisis intervention auxiliary method as described in the first aspect and various possible designs of the first aspect.

[0013] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium storing computer execution instructions, when a processor executes the computer execution instructions, the psychological crisis intervention auxiliary method as described in the first aspect and various possible designs of the first aspect is implemented.

[0014] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, when the computer program is executed by a processor, the psychological crisis intervention auxiliary method as described in the first aspect and various possible designs of the first aspect is implemented.

[0015] The psychological crisis intervention auxiliary method, device, equipment and storage medium provided by the application have at least the following beneficial effects: 1) Traditional psychological crisis intervention relies on manual discovery or regular scale research, and has the defect of not timely crisis identification. The application can quickly complete semantic analysis and risk quantification calculation by acquiring target psychological interaction information in real time through multi-channel information collection and combining a large language model fine-tuned by target psychological corpus. The whole process from information collection to risk determination does not require manual intervention, can capture crisis signals at the first time when the target expresses psychological distress, and effectively solves the problem of lagging behind in traditional intervention identification.

[0016] 2) Existing intervention schemes rely on experience judgment and lack objective quantitative standards, which can easily lead to mismatch between intervention strategies and actual needs. The application provides scientific basis for intervention decision-making by constructing double quantification formulas: on the one hand, the risk value formula converts multi-dimensional indicators such as the number of keywords, the intensity of negative emotions and the number of times of negative emotion persistence into calculable values, divides the risk level into low, medium and high, and ensures that the intervention strategy is accurately corresponding to the risk level; on the other hand, the intervention effect formula quantitatively evaluates the intervention effect by comparing the number of positive emotional statements before and after intervention, avoiding subjective judgment errors. The double quantification logic makes the intervention from experience-driven to data-driven, significantly improving the accuracy and scientificity of intervention decision-making.

[0017] 3) Traditional intervention often lacks follow-up after one-time intervention, making it difficult to cope with the dynamic changes of the target's psychological state. The application continuously collects the target's subsequent interaction information through the effect tracking module within a preset time after intervention, and evaluates the intervention effect based on the intervention effect improvement rate E If E >0, maintain or adjust the light tracking strategy; if E ≤0, immediately re-analyze the semantics, adjust the weight coefficient or replace the intervention tactics, and execute the intervention process again. In this way, the changes in the target's psychological state can be responded to in real time, ensuring that the intervention strategy always adapts to the target's needs and ensuring the continuity and effectiveness of the intervention effect. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the application.

[0019] Figure 1 A structural diagram of a psychological crisis intervention auxiliary system provided for an embodiment of the application; Figure 2 A flowchart of a psychological crisis intervention auxiliary method provided for an embodiment of the application; Figure 3 Another flowchart of a psychological crisis intervention auxiliary method provided for an embodiment of the application; Figure 4 A structural diagram of a psychological crisis intervention auxiliary device provided by an embodiment of the present application is shown.

[0020] The specific embodiments of the present application have been shown in the above-described drawings, and will be described in more detail hereinafter. These drawings and the written description are not intended to restrict the scope of the present application by any means, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0021] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The same or similar components are denoted by the same reference numerals throughout the drawings and the following description, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0022] In the technical solution of the present application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data comply with relevant laws and regulations and do not violate public order and good customs.

[0023] It should be noted that in the embodiments of the present application, some existing industry solutions, components, models, etc. may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0024] The technical solution of the present application and how the technical solution of the present application solves the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0025] An embodiment of the present application provides a psychological crisis intervention auxiliary method, as shown in the figure. Figure 1 As shown in the figure, a psychological crisis intervention auxiliary system is provided by an embodiment of the present application, and the psychological crisis intervention auxiliary system is used to implement the psychological crisis intervention auxiliary method provided by the embodiment. The psychological crisis intervention auxiliary system includes an interaction module 110, a semantic analysis module 120, a risk quantification evaluation module 130, an intervention strategy production module 140 and an effect tracking module 150 connected in sequence, wherein the risk quantification evaluation module 130 is configured with a quantification formula calculation unit 131 and a weight coefficient configuration unit 132.

[0026] The interaction module 110 supports multi-channel information collection, including text messages of the campus psychological service platform, dialogue access of instant messaging tools, voice interaction of intelligent psychological terminals, etc. When receiving voice information, a speech recognition (ASR) model based on deep learning is used to convert the voice into text, ensuring that the subsequent large language model can process the unified text format. Taking college students as the target of psychological crisis intervention assistance as an example, the instant messaging tool can be a chat window built in the campus APP, and the intelligent psychological terminal can be a psychological self-service device.

[0027] The semantic analysis module 120 is configured to call a large language model fine-tuned on the target psychological corpus, perform semantic analysis on the psychological interaction information, identify emotion types and extract psychological crisis keywords. An exemplary fine-tuning process of a large language model: select historical counseling records of a university psychological counseling center, college student psychological forum texts, psychological course learning discussion data, etc. to construct a college student psychological corpus, and use supervised fine-tuning (SFT) to adapt the large language model to the college student psychological expression habits and common psychological problem scenarios, such as “pre-examination anxiety” and “dormitory interpersonal relationship conflicts”. After fine-tuning, the model can perform semantic analysis on the input interaction text and output an emotion type list and a psychological crisis keyword set. The emotion type list includes an intensity score for each emotion S i .

[0028] The risk quantification evaluation module 130 is configured to calculate the current target psychological crisis risk value according to the number of psychological crisis keywords. The quantification formula calculation unit 131 is configured with a quantification formula, and the weight coefficient configuration unit 132 is configured to adjust the weight coefficient in the quantification formula.

[0029] The pre-strategy production module 140 is configured to execute an intervention strategy on the current target according to the current target psychological crisis risk value. For example, when a college student is taken as the target of intervention assistance, personalized intervention content can be generated according to the risk level and the current semantic analysis result. For example, when “pre-examination anxiety” related semantics are identified, soothing language is generated: “It is normal to have moderate anxiety before an exam. Try to break down your review tasks into small goals and reward yourself with a small prize for each completed task.” When high-risk keywords (such as “suicide”) are identified, emergency intervention language is generated: “I understand your feelings. The teachers in the school psychological center are available 24 / 7. You can call this number: XXX, or go directly to the XX building psychological center to find a teacher.” The effect tracking module 150 is configured to collect subsequent interaction information of the target within a preset time after the intervention strategy is executed, and calculate the intervention effect improvement rate by substituting the intervention effect formula. If the intervention effect improvement rate is less than or equal to 0, the intervention strategy is adjusted.

[0030] Exemplarily, the subsequent interactive text of the collection target assumes the number of positive emotion sentences before intervention N p,pre = 5, the number of positive emotion sentences after intervention N p,post = 8, the intervention effect improvement rate is 60%, which shows that the intervention of the alumni maintains The psychological crisis intervention auxiliary method realized by the above system realizes the real-time, accurate and dynamic intervention auxiliary of the psychological crisis of college students through the combination of the semantic analysis ability of the large language model and the quantitative evaluation formula, which not only plays the advantage of the large language model in understanding natural language, but also provides objective basis for intervention decision through quantitative logic, effectively improving the efficiency and scientificity of the psychological crisis intervention of college students.

[0031] As Figure 2 shown, the following takes college students as the target of psychological crisis intervention auxiliary, and the psychological crisis intervention auxiliary method includes the following steps S10 to S50, wherein the steps S10 to S50 can be implemented by the interactive module 110, the semantic analysis module 120, the risk quantitative evaluation module 130, the intervention strategy production module 140 and the effect tracking module 150 connected in sequence as shown in the signal connection. Figure 1

[0032] It should be noted that the target to which the method disclosed in the present application can be applied includes but is not limited to college students, and it can also be applied to other groups of targets, such as primary and secondary school students or staff, which all belong to the target category to which the present application can be applied. Here, only examples are given, and they do not constitute a limitation on the present application.

[0033] S10: Obtain psychological interaction information of the target. If the psychological interaction information is voice information, it is converted into text through voice recognition technology.

[0034] Exemplarily, taking college students as an example. First, information collection entrances are built relying on psychological service related carriers, which include three types of channels. The first type is an online platform channel, which covers the text message board of the psychological service exclusive platform and the dialogue window of the instant messaging tool. The second type is an offline intelligent terminal channel, which is deployed in schools, communities, enterprises and other scenes. The intelligent psychological self-service equipment is equipped with a voice interaction module, which supports the target to directly express the psychological state through voice. The third type is a related system docking channel, which connects the data interface of the psychological health management subsystem of the organization (school in this embodiment) where the target is located, and synchronizes the psychological state records submitted by the target.

[0035] ​Based on the above channels, active submission and / or periodic guidance are used to trigger collection. On the one hand, the target can initiate the submission of psychological interaction information at any time according to its own needs through any of the above channels, such as inputting "recent emotional depression" in the instant messaging window or speaking "I feel very stressed" through the intelligent terminal. On the other hand, lightweight collection guidance is pushed to the target according to the preset period, such as "whether you are willing to share your recent psychological state? It will help you to provide more suitable psychological support", which guides the target to participate in information interaction and ensures the continuity of information collection.

[0036] After receiving the information submitted by the target, the information type is first determined by the format identification module to distinguish between text information and voice information. For text information, it is directly stored in the text interaction database and marked with collection time, collection channel and target identification; wherein the target is represented by a unique code after anonymization processing; for voice information, it is stored in the voice interaction database, and a corresponding voice file number is generated, the target identification and the collected information are associated, and data traceability basis is provided for the subsequent conversion link.

[0037] The voice file to be converted is extracted from the voice interaction database, and preprocessing is performed to improve the recognition accuracy. The preprocessing includes three core steps: first, noise filtering, using an adaptive noise cancellation algorithm to remove environmental noise in the voice; second, voice segmentation, using endpoint detection technology to identify the starting point and ending point of the voice signal, and eliminating the silent audio segments at the beginning and end; third, format unification, converting voice files of different sampling rates and different encoding formats into 16 kHz sampling rate, 16-bit single-channel WAV format. An automatic speech recognition (ASR) model based on deep learning is called to convert the semantics of the preprocessed voice file. The ASR model uses a hybrid architecture of convolutional neural network (CNN), recurrent neural network (RNN) and connectionist temporal classification (CTC). The CNN layer is used to extract local features of the voice, the RNN layer uses a bidirectional long short-term memory network BiLSTM to capture the time sequence dependence of the voice, and the CTC layer is used to solve the alignment problem between voice and text, and to improve the recognition accuracy of continuous voice. The model input is the preprocessed voice feature vector, and the output is the corresponding text content; if there are ambiguous semantics in the recognition process, such as dialect words and unclear pronunciation sentences, the system will generate a "suspected text candidate set", such as "I recently sleep badly" and "I recently water badly" as candidates, and mark the ambiguity degree for further verification by the subsequent semantic analysis module.

[0038] The text result output by the ASR model is double-checked. The first check is grammar check, which detects whether the text has obvious grammatical errors through a grammar rule library, such as incomplete sentences like “I have been very anxious recently and don't want to go out every day”. If there is, the model will be triggered to re-identify. The second check is semantic consistency check, which compares the length of the voice file with the number of characters in the converted text. If the ratio of the number of text characters to the length of the voice exceeds the preset threshold, such as less than 5 or more than 30 characters per 10 seconds of voice, it is determined that the recognition is abnormal, and the preprocessing and conversion steps are re-executed. After the check is passed, the converted text is stored in the text interaction database, and the original voice file number is associated. At the same time, the voice interaction database is deleted or transferred to the backup database, and the storage space is released.

[0039] The converted text and directly collected text information are subjected to format standardization operations, including removing special characters in the text, unifying punctuation usage specifications, and correcting errors, to ensure that the text format received by the subsequent semantic analysis module is uniform and the content is accurate.

[0040] Finally, the standardized text is associated with metadata such as target identification, collection time, collection channel, and information type to generate an interactive text data unit containing complete information dimensions. At the same time, an index system is established in the text interaction database, with target identification and collection time as core index fields, and collection channel and information type as auxiliary index fields, to facilitate the subsequent semantic analysis module to quickly call the target's historical interactive text and current interactive text, and support continuous analysis of risk assessment.

[0041] S20: Call the large language model fine-tuned by the target psychological corpus, perform semantic analysis on the psychological interaction information, identify the emotion type, and extract the psychological risk keywords.

[0042] In this embodiment, step S20 can be implemented through steps S201-S206 as follows.

[0043] S201: Construct a target psychological corpus.

[0044] Taking the college student group as the core target, three types of core corpus are collected to build a dedicated psychological corpus. The first type is the historical counseling records of the university psychological counseling center. The counseling records of the past 5-10 years containing descriptions of psychological distress of college students, responses of counselors and emotional labels are selected. After filtering out sensitive information involving personal privacy, desensitization processing is performed. The second type is the psychological forum text of college students. Posts and comments containing psychological state expressions are collected from campus BBS, college student psychological health community and other platforms. Advertisements and irrelevant chatting content are filtered out. The third type is the psychological course learning discussion data. Classroom discussion records and texts about self-psychological state analysis in homework after class are collected to ensure that the corpus is consistent with the daily psychological expression scene and high-frequency psychological problem types of college students.

[0045] S202: Corpus annotation and preprocessing.

[0046] Annotation and preprocessing operations are performed on the constructed target psychological corpus. In the annotation stage, manual annotation and machine-assisted verification are used. The annotators with basic knowledge of psychology annotate each corpus with emotion type, psychological crisis keyword and emotion intensity level. The emotion type includes but is not limited to anxiety, depression and anger. The psychological crisis keyword includes but is not limited to "suicide", "self-harm", "life has no meaning", "insomnia" and "irritability". The emotion intensity level corresponds to the S i value output by the subsequent semantic analysis, with a value range of [0, 5].

[0047] Machine-assisted verification cross-verified the annotation results by using a trained basic emotion analysis model to correct annotation errors, such as adjusting the mislabeled "depression" corpus to the correct category. The preprocessing stage includes text segmentation, stop word removal and text vectorization to provide standardized input for model fine-tuning. Text vectorization is used to convert text into a vector format that can be recognized by the model.

[0048] S203: Fine-tuning of large language model.

[0049] A basic large language model is selected as the pre-training model, and a supervised fine-tuning (SFT) method is used to adapt to the target psychological corpus. The basic large language model includes but is not limited to BERT and GPT series models. The fine-tuning process is divided into three stages: The first stage is initialization, the pre-trained model parameters are loaded to the training framework, part of the bottom layer parameters are frozen, the calculation amount is reduced and the model basic semantic understanding ability is reserved, only the top layer parameters related to psychological semantics are trained; the second stage is iterative training, the labeled target psychological corpus is used as training data, the cross entropy loss function is used as loss function, the model parameters are continuously adjusted through back propagation, so that the model learns the college students' psychological expression habits, such as identifying the semantics of college students' exclusive expressions such as "pressure of failing a course" and "GPA anxiety"; the third stage is model verification and optimization, 20% of the corpus is divided from the target psychological corpus as a test set, the test set is used to evaluate the performance of the model, if the emotion type recognition accuracy is less than 90% and the keyword extraction recall rate is less than 85%, the training rounds are increased or the corpus is supplemented to fine-tune again, until the model meets the performance requirements of accurately identifying the keywords of college students' psychological emotions and dangers, and finally generates an exclusive large language model fine-tuned by the target psychological corpus, which is deployed to the semantic analysis module for standby.

[0050] S204: Obtain psychological interaction text and perform preprocessing.

[0051] From the text interaction database generated in step S10, the psychological interaction text of college students is retrieved, including the original text and the text converted from voice, and the same operation as the model fine-tuning preprocessing is performed on the text to ensure that the input text format matches the model input requirements. If the text length exceeds the maximum input length limit of the model, text truncation or segmentation processing is used, such as splitting long text into "learning pressure description" and "interpersonal relationship description" fragments according to semantic logic to avoid information loss.

[0052] S205: Input the preprocessed psychological interaction text into the fine-tuned large language model to realize emotion type recognition and intensity score output.

[0053] The processed text is input into the fine-tuned large language model, the model analyzes the semantic features in the text, such as "fear of failing a course" and "still have no idea after midnight" corresponding to "anxiety" category emotion, "not interested in anything" and "don't want to talk" corresponding to "depression" category emotion, and outputs the target emotion type list; at the same time, based on the emotion intensity level annotated by the corpus, combined with the intensity of emotion expression in the text, only as an example, "occasional insomnia" corresponds to S i =2, "insomnia every day and unable to sleep" corresponds to S i =4, the intensity score corresponding to each emotion type S i is output, and the number of negative emotions n is counted, such as "anxiety" and "depression" are recognized at the same time, then n=2.

[0054] S206: Extract psychological danger keywords The model extracts the psychological crisis keywords from the input text by learning the annotated crisis keyword features in the target psychological corpus, such as the semantic direction of the words and the context of the appearance. During the extraction process: on the one hand, the model directly extracts the explicit keywords appearing in the text by matching the preset crisis keyword library learned during the fine-tuning process of the model, such as "suicide" and "self-harm"; on the other hand, the model identifies implicit keywords by semantic association, such as in the text "I feel that life has no hope, it is better to end everything", although "suicide" does not appear directly, the model extracts "end everything" as an implicit crisis keyword through semantic understanding, ensuring that no crisis keyword is missed. After the extraction is completed, the "emotion-keyword" analysis result table of the target psychological interactive text is generated, which contains the emotion type, S i

[0055] S30: According to the number of psychological crisis keywords, calculate the psychological crisis risk value of the current target.

[0056] In this embodiment, the psychological crisis risk value of the college student is calculated by a quantitative formula, which is: , In the formula, R is the psychological crisis risk value, the value range is [0, 10], the larger the value, the higher the risk, R is the number of psychological crisis keywords in the interactive text, is the total number of sentences in the interactive text, is the intensity score of the th negative emotion, which is output by the large language model based on semantic sentiment analysis, and the value range is [0, 5], i is the number of negative emotions identified, n is the number of times that the proportion of negative emotions in the recent continuous interaction exceeds the threshold value, T , α , β , gamma are weight coefficients, which satisfy α + β + gamma =1.

[0057] Exemplarily, taking a college student's interaction as an example, the interactive text contains 20 sentences, =20, among which "insomnia" and "irritability" are identified as two psychological crisis keywords, =2, two negative emotions "anxiety" (S1=4) and "depression" (S2=3) are identified, n=2), and the proportion of negative emotions in the last three continuous interactions exceeds 40%, T=3. Set the weight coefficients α=0.4, β=0.3, and γ=0.3, then according to the above quantitative formula, the psychological crisis risk value R ​= 2.0 < 3, trigger light-weight appeasement strategy S40: execute intervention strategy on the current target according to the current target's psychological crisis risk value.

[0058] In this embodiment, college students are taken as an example. According to the psychological crisis risk value R The following intervention strategies are executed: If 0 < R < 2.0, trigger light-weight appeasement strategy R < 3: generate light-weight appeasement rhetoric and push psychological health popular science articles. An exemplary light-weight appeasement rhetoric is "Have you been feeling tired recently? Why not try deep breathing relaxation for 10 minutes every day?"

[0059] If 3 < R < 7, trigger medium-weight intervention strategy R <7: generate deep appeasement dialogue guidance and send attention reminders to counselors. An exemplary deep appeasement dialogue guidance is "Would you like to talk to me more about what's making you feel sad?"

[0060] If R > 7, trigger emergency intervention strategy R > 7: immediately generate emergency intervention rhetoric and trigger the emergency intervention process of the psychological center on duty. An exemplary emergency intervention rhetoric is "I understand how you feel right now. The teachers at the school psychological center are on duty 24 hours a day. Here is the contact number: XXX."

[0061] In some embodiments, step S40 is implemented through steps S401-S403 as follows.

[0062] S401: set risk value grading standards.

[0063] Based on the psychological crisis risk value R calculated in S30, which ranges from 0 to 10, three risk grading standards are set, including low-risk, medium-risk, and high-risk intervals, as follows: Low-risk interval: 0 < R < 3, corresponding to a target psychological state of mild emotional distress without obvious crisis tendency, requiring light-weight intervention; Medium-risk interval: 3 < R < 7, corresponding to a target psychological state of moderate emotional problems with potential crisis risk, requiring medium-weight intervention; High-risk interval: R > 7, corresponding to a target psychological state of severe emotional crisis, possibly leading to extreme behaviors such as self-harm and suicide, requiring emergency intervention.

[0064] S402: establish intervention strategy matching rule library.

[0065] For different risk intervals, a multi-dimensional strategy matching rule library is constructed, including intervention subjects, intervention content, intervention timeliness, and intervention channels. The rule library associates risk values with intervention measures through formalized logic, as follows.

[0066] Definition of rule base core parameters: I : intervention strategy type; I 1 for light intervention, I 2 for moderate intervention, I 3 for emergency intervention; A : intervention execution subject, A 1 for system automatic execution, A 2 for counselor and system collaborative execution, A 3 for psychological center on-duty personnel, counselor and system collaborative execution; C : intervention content complexity; C 1 for basic pacification and popularization, C 2 for deep guidance and attention reminder, C 3 for emergency rhetoric and intervention process; t : intervention response timeliness; t 1 for response within 12 hours, t 2 for response within 4 hours, t 3 for response within 15 minutes; Ch : intervention execution channel; Ch 1 for online text / push, Ch 2 for online dialogue and offline follow-up, Ch 3 for multi-channel synchronization, multi-channel including online pop-up window, telephone notification and offline intervention.

[0067] Based on the above core parameters, the matching formula of intervention strategy is as follows: If 0≤ R <3, then I = I 1, A = A 1, C = C 1, t = t 1, Ch = Ch 1; If 3≤ R <7, then I = I 2, A = A 2, C = C 2, t = t 2, Ch = Ch 2; If R ≥7, thenI = I 3、 A = A 3、 C = C 3、 t = t 3、 Ch = Ch 3.

[0068] S402: Execute the risk interval intervention strategy.

[0069] In the low-risk interval, execute the light intervention I 1, including intervention content generation and push.

[0070] In this embodiment, the light soothing rhetoric is generated by the following method: calling a large language model fine-tuned on the target psychological corpus, combining the emotional type identified by S20, generating personalized soothing rhetoric, and the rhetoric generation satisfies the formula: S comf = S base + S emo， wherein, S comf is the final soothing rhetoric, S base is the basic soothing template, S emo is the emotional adaptation supplement sentence.

[0071] The push method of the mental health popular science article is: matching popular science content based on emotional type, and the push priority is calculated by the formula: P push = omega 1· R + omega 2· N emo， wherein, P push is the push priority, taking value [1, 5], the higher the value, the higher the push priority, omega 1=0.3、 omega 2=0.7 are weight coefficients, N emo is the number of negative emotions identified by S20, such as identifying 1 kind of negative emotion, N emo =1; identifying 2, N emo =2. For example, R =2.5、 Nemo = 1, P push = 1.45, push light anxiety relief tips class articles; R = 2.0, N emo = 2, P push = 2.0, push multi-emotional distress self-regulation method class articles.

[0072] The system completes the script and article pushing through Ch 1 channel at t1, and records the pushing time T send , target click view time T view , generates a low-risk intervention execution record table, and stores it in the intervention archive library.

[0073] In the medium-risk range, moderate intervention is performed I 2.

[0074] When performing moderate intervention, generate a deep pacification dialogue guide in the following way: Combine the psychological crisis keywords extracted by S20 and the negative emotion intensity score S i , generate a guided dialogue that meets the formula: D guide = D open + D focus， Among them, D guide is the deep pacification dialogue guide content, D open is the open-ended guide opening, D focus is the focus keyword supplement guide.

[0075] Based on the risk value R Set the interaction frequency, the formula is: , Among them, F chat is the weekly dialogue interaction frequency, taking values [1, 3], is the floor function. For example, R = 3.5, interact once a week; R = 6.0, interact 3 times a week.

[0076] The content used to remind the counselor to pay attention is generated in the following way: The system automatically generates a target anonymous identification containing ID anon , risk value R , emotion type Type emo , keywords K cris Attention reminder, reminder content format:

Medium risk attention reminder

[0077] The above content is delivered through the Ch 2 channels of counselor work APP push and SMS notification, and the counselor confirms receipt, and the system records the confirmation time T confirm If it is not confirmed within 1 hour, a second SMS reminder will be automatically triggered.

[0078] After the counselor completes the follow-up, fill in the medium risk follow-up record table in the system, record the follow-up content Content follow , target state feedback Feedback state ; The system generates a moderate intervention execution report in combination with online dialogue interaction records, and updates the target intervention archives in real time.

[0079] In the high-risk range, execute emergency intervention I3.

[0080] The generation method of the emergency intervention dialogue is: based on the high-risk keywords and risk value R extracted by S20, generate an emergency intervention dialogue, the formula is: S urgent = S empathy + S resource + S action， Among them, S urgent is the emergency intervention dialogue, S empathy is the empathy sentence, S resource is the resource notification, S action is the action guide.

[0081] The emergency intervention process of the psychological center on duty personnel is as follows: The system automatically generates high-risk emergency intervention instructions, including target anonymous identification ID anon , real-time location (if the target authorizes APP positioning), risk value R , high-risk keywords K high , through the psychological center on-duty system pop-up window + on-duty personnel mobile phone hotline Ch 3 channels to deliver, to ensure response within t 3 hours.

[0082] Set the key nodes of the intervention process, and calculate the time limit for completing the nodes by formula: T node =15 / N node ; wherein, T node is the longest completion time limit of each node (unit: minutes), N node is the number of intervention process nodes, fixed at 3, including instruction reception confirmation, first contact with the target, and on-site / remote intervention. For example, N node =3, T node =5 minutes, that is, instruction reception confirmation needs to be completed within 5 minutes, the first contact with the target needs to be completed within 5 minutes, and on-site / remote intervention needs to be started within 5 minutes.

[0083] The system synchronously pushes the collaborative intervention notice to the target's department counselor and security department (if necessary), and clearly defines the responsibilities of each subject, among which the counselor is responsible for contacting the target's roommate / classmate to confirm safety, and the security department is responsible for assisting the on-site intervention route guidance; During the intervention process, all subjects share the intervention progress in real time through the system until the target state is stable, and finally generate a high-risk emergency intervention report and archive it to the psychological center special archive library.

[0084] S403: Preliminary verification of intervention strategy execution effect.

[0085] After the intervention strategy is executed, the real-time response rate of each risk interval intervention is calculated, and the formula is: , wherein, is the real-time response rate, is the actual intervention execution time, is the response time limit required for the risk interval, is the number of interventions that meet the response time limit, is the total number of interventions for the risk interval. It is required that Need to achieve: low risk ≥ 95%, medium risk ≥ 98%, high risk ≥ 100%, if not up to standard, trigger system alarm, investigate channel or process problems.

[0086] S50: Collect the subsequent interaction information of the target within the preset time after the intervention strategy is executed, substitute it into the intervention effect formula to calculate the intervention effect improvement rate, and if the intervention effect improvement rate is less than or equal to 0, adjust the intervention strategy.

[0087] In this embodiment, the intervention effect formula is: , In the formula, is the intervention effect improvement rate, E > 0 indicates that the intervention is effective, E the larger the effect is better, is the number of positive emotion sentences before intervention, is the number of positive emotion sentences after intervention.

[0088] In an exemplary embodiment, as shown in Figure 3 Another flowchart of a psychological crisis intervention auxiliary method provided by the embodiment of the application is provided, and in this embodiment, the psychological crisis intervention auxiliary method is based on the system as shown in Figure 1 The method is implemented through the following steps 1 to 5.

[0089] Step 1, information collection and conversion: the interaction module collects the psychological interaction information of college students, and if it is voice information, it is converted into text through voice recognition technology (such as an ASR model); Step 2, semantic analysis: the semantic analysis module calls a large language model fine-tuned by college student psychological corpus, performs semantic analysis on the interaction text, identifies emotion types (such as anxiety, depression, anger, etc.) and extracts psychological crisis keywords; Step 3, risk quantification calculation: the risk quantification evaluation module substitutes the risk value formula R to calculate the psychological crisis risk value of the current college student; Step 4, intervention strategy execution: the intervention strategy generation module executes the corresponding intervention strategy according to the risk value R: If R < 3: generate light comfort rhetoric and push psychological health popular science articles; If 3 ≤ R < 7: generate deep comfort dialogue guidance, and send attention reminders to counselors; If R ≥ 7: immediately generate emergency intervention rhetoric, and trigger the emergency intervention process of the on-duty personnel of the psychological center.

[0090] Step 5, effect tracking and strategy adjustment: the effect tracking module collects the subsequent interaction information of the college students within a preset time (1 week in this embodiment) after the intervention, evaluates the intervention effect through the intervention effect formula E; if E≤0 (intervention is invalid or deteriorates), the intervention strategy is adjusted and step 4 is repeated.

[0091] The embodiment of the application further provides a psychological crisis intervention auxiliary device, as shown in the accompanying drawings. Figure 4 The psychological crisis intervention auxiliary device comprises: A data acquisition unit 401 configured to acquire psychological interaction information of a target, and if the psychological interaction information is voice information, convert it into text through voice recognition technology; A keyword extraction unit 402 configured to call a large language model fine-tuned by a target psychological corpus, perform semantic analysis on the psychological interaction information, identify an emotional type and extract psychological crisis keywords; A risk value calculation unit 403 configured to calculate a psychological crisis risk value of the current target according to the number of psychological crisis keywords; An intervention strategy execution unit 404 configured to execute an intervention strategy on the current target according to the psychological crisis risk value of the current target; An intervention strategy adjustment unit 405 configured to collect subsequent interaction information of the target within a preset time after the intervention strategy is executed, calculate an intervention effect improvement rate by substituting the intervention effect formula, and adjust the intervention strategy if the intervention effect improvement rate is less than or equal to 0.

[0092] In some embodiments, the target psychological corpus comprises historical counseling records of a psychological counseling center, target group psychological theme forum texts, and psychological course learning discussion data; the large language model is adapted to target psychological expression habits and common psychological problem scenarios through supervised fine-tuning; and the output result of the semantic analysis comprises a list of emotional types with intensity scores and a set of psychological crisis keywords.

[0093] In some embodiments, the risk value calculation unit is further configured to calculate the psychological crisis risk value of the current target according to the number of psychological crisis keywords through the following formula: , In the formula, R is the psychological crisis risk value, the value range is [0, 10], the larger the value, the higher the risk, is the number of psychological crisis keywords in the interaction text, is the total number of sentences in the interaction text, is the intensity score of the first i negative emotion, which is output by the large language model based on semantic sentiment analysis, and the value range is [0, 5], n is the number of recognized negative emotions,T the number of times that the proportion of negative emotions in the recent continuous interactions exceeds a threshold value, α 、 β 、 gamma is a weight coefficient, satisfying α + β + gamma = 1.

[0094] In some embodiments, the intervention strategy execution unit is further configured to: when R < 3, the intervention strategy is to generate a light-weight soothing phrase and push a mental health popular science article; when 3≤R<7, the intervention strategy is to generate a deep soothing dialogue guide and send a concern reminder to the attention person associated with the target; when R≥7, the intervention strategy is to generate an emergency intervention phrase and trigger an emergency intervention process of the on-duty personnel in the mental center.

[0095] In some embodiments, the intervention effect formula is: , wherein, is an intervention effect improvement rate, E >0 indicates that the intervention is effective, E the larger the effect is better, is the number of positive emotional statements before intervention, is the number of positive emotional statements after intervention.

[0096] In some embodiments, the preset time is 1 week, and the intervention strategy adjustment unit is further configured to adjust the intervention strategy when the intervention effect improvement rate is less than or equal to 0 by: re-performing semantic analysis on subsequent interaction information and adjusting the weight coefficient α 、 β 、 gamma , recalculating the psychological crisis risk value, and adjusting the intervention strategy according to the recalculated psychological crisis risk value.

[0097] In some embodiments, the intervention strategy adjustment unit is further configured to maintain the current intervention strategy or adjust to a light-weight tracking strategy when the intervention effect improvement rate is greater than 0; the light-weight tracking strategy includes regularly collecting target short psychological state feedback texts and performing semantic analysis to calculate the psychological crisis risk value.

[0098] Embodiments of the present application provide an electronic device. The electronic device can include a processor, a memory, wherein the processor and the memory can communicate; for example, the processor and the memory communicate through a communication bus.

[0099] The processor executes computer-executed instructions stored in the memory, so that the processor executes the solutions in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0100] The communication bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries and read-only libraries). The memory can include random access memory (RAM), and can also include non-volatile memory.

[0101] The electronic device provided by the embodiments of the present application can be the terminal device of the above embodiments.

[0102] The embodiments of the present application further provide a computer readable storage medium, which stores computer instructions, and when the computer instructions run on a computer, the computer executes the technical solutions of the above-mentioned embodiment of the psychological crisis intervention auxiliary method.

[0103] The embodiments of the present application further provide a computer program product, which includes a computer program stored in a computer readable storage medium, and at least one processor can read the computer program from the computer readable storage medium, and when the at least one processor executes the computer program, the technical solutions of the above-mentioned embodiment of the psychological crisis intervention auxiliary method can be realized.

[0104] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other manners. For example, the embodiments of the device described above are merely schematic. For example, the division of the modules is merely logical function division. There can be another division manner for the actual implementation. For example, a plurality of modules or a component can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or modules, and can be electrical, mechanical or in other forms.

[0105] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical units, i.e., can be located in one place, or can be distributed to a plurality of network units. Some or all of the modules can be selected according to actual needs to implement the embodiments of the present application.

[0106] In addition, each function module in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present separately, or two or more modules can be integrated in one unit. The unit formed by the above modules can be realized in the form of hardware, or in the form of hardware plus software function unit.

[0107] The integrated module realized in the form of software function module can be stored in a computer readable storage medium. The software function module stored in the storage medium includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute part of the steps of the method of each embodiment of the present application.

[0108] It should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor. The steps of the method disclosed in the present application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0109] The memory can include a high-speed RAM memory, and can also include a non-volatile storage NVM, such as at least one disk memory, and can also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0110] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0111] The storage medium can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0112] An exemplary storage medium is coupled to the processor so that the processor can read information from, and write information to, the storage medium. Of course, the storage medium can be a part of the processor. The processor and the storage medium can be located in an application specific integrated circuits (ASIC). Of course, the processor and the storage medium can exist as discrete components in an electronic control unit or host device.

[0113] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program, when executed, executes steps including the above-mentioned method embodiments; and the foregoing storage medium includes ROM, RAM, magnetic disk or optical disk and various storage media that can store program codes.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A mental crisis intervention assistance method characterized by, The method comprises: acquiring psychological interaction information of a target, and converting the psychological interaction information into text through voice recognition technology if the psychological interaction information is voice information; calling a large language model fine-tuned by a target psychological corpus to perform semantic analysis on the psychological interaction information, identify an emotion type, and extract psychological crisis keywords; calculating a psychological crisis risk value of the target according to the number of the psychological crisis keywords; executing an intervention strategy on the target according to the psychological crisis risk value of the target; collecting subsequent interaction information of the target within a preset time after execution of the intervention strategy, and calculating an intervention effect improvement rate by substituting the subsequent interaction information into an intervention effect formula, and adjusting the intervention strategy if the intervention effect improvement rate is less than or equal to 0.

2. The psychological crisis intervention assistance method according to claim 1, characterized by, The target psychological corpus comprises historical counseling records of a psychological counseling center, target group psychological theme forum texts, and psychological course learning discussion data; the large language model is adapted to target psychological expression habits and common psychological problem scenarios through supervised fine-tuning; and the output result of the semantic analysis comprises an emotion type list with intensity scores and a psychological crisis keyword set.

3. The psychological crisis intervention assistance method according to claim 1, characterized by, The psychological crisis risk value of the target is calculated according to the number of the psychological crisis keywords through the following formula: In the formula, R is a psychological crisis risk value, the value range is [0, 10], the larger the value, the higher the risk, The number of keywords of the interactive text centering on psychological crisis, The total number of sentences of the interactive text, The intensity score of the first i negative emotion, output by the large language model based on semantic sentiment analysis, the value range is [0, 5], n The number of identified negative emotions, T The number of times that the proportion of negative emotions in the recent continuous interaction exceeds the threshold, α , β , The intervention strategy on the target according to the psychological crisis risk value of the target comprises: The weight coefficient satisfies α + β + when R < 3, the intervention strategy is to generate a lightweight pacification phrase and push a psychological health popular science article; =1.

4. The psychological crisis intervention assistance method according to claim 3, characterized by, when 3 ≤ R < 7, the intervention strategy is to generate a deep pacification dialogue guide and send a concern reminder to a person concerned with the target; when R ≥ 7, the intervention strategy is to generate an emergency intervention phrase and trigger an emergency intervention process of a psychological center on duty. The intervention effect formula is: The preset time is one week, and the intervention strategy is adjusted in the following manner when the intervention effect improvement rate is less than or equal to 0:

5. The psychological crisis intervention assistance method according to claim 1, characterized by, When the intervention effect improvement rate is greater than 0, the current intervention strategy is maintained or adjusted to a lightweight tracking strategy; the lightweight tracking strategy comprises regularly collecting a target short psychological state feedback text and performing semantic analysis to calculate a psychological crisis risk value. In the formula, is the intervention effect improvement rate, E > 0 indicates that the intervention is effective, E The larger the effect is better, is the number of positive emotional statements before intervention, is the number of positive emotional statements after intervention. 6.The psychological crisis intervention assistance method according to claim 1, wherein, The device comprises: re-performing semantic analysis on the subsequent interaction information and adjusting the weight coefficient α , β , a data acquisition unit configured to acquire psychological interaction information of a target, and convert the psychological interaction information into text through voice recognition technology if the psychological interaction information is voice information; , recalculating the psychological crisis risk value, and adjusting the intervention strategy according to the recalculated psychological crisis risk value.

7. The psychological crisis intervention assistance method according to claim 1, characterized by, a keyword extraction unit configured to call a large language model fine-tuned by a target psychological corpus to perform semantic analysis on the psychological interaction information, identify an emotion type, and extract psychological crisis keywords; 8. A mental crisis intervention assistance device characterized by comprising: a risk value calculation unit configured to calculate a psychological crisis risk value of the target according to the number of the psychological crisis keywords; an intervention strategy execution unit configured to execute an intervention strategy on the target according to the psychological crisis risk value of the target; an intervention strategy adjustment unit configured to collect subsequent interaction information of the target within a preset time after execution of the intervention strategy, substitute the subsequent interaction information into an intervention effect formula to calculate an intervention effect improvement rate, and adjust the intervention strategy if the intervention effect improvement rate is less than or equal to 0. comprise: a processor, and a memory connected to the processor in communication; the memory stores computer execution instructions; 9. An electronic device, comprising: ​ ​ ​ The processor executes computer-executed instructions stored in the memory to implement the psychological crisis intervention assistance method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executed instructions, and the computer-executed instructions are executed by the processor to implement the psychological crisis intervention assistance method according to any one of claims 1-7.

Images