AI-based Emotional Dependence Risk Detection Method and Device Based on Multi-Dimensional Comprehensive Scoring

Abstract

This application relates to a method and apparatus for detecting AI emotional dependence risk based on multi-dimensional comprehensive scoring. The method includes: acquiring four types of data sequences during the user's interaction with the AI ​​system: interaction intensity, emotional substitution, dependence during vulnerable moments, and withdrawal reactions; determining, based on each sequence, a score for the acceleration of interaction intensity changes, a score for the substitution relationship between AI and real social interaction, a score for seeking help during emotional lows, and a score for the intensity of withdrawal reactions; weighting and synthesizing the four-dimensional scores to generate a Dependence Risk Index (DRI); and outputting an alarm message when the DRI meets preset alarm conditions. This method can quantitatively assess the depth and risk of a user's emotional dependence on the AI ​​system, effectively distinguishing between healthy companionship and pathological dependence.

Description

Technical Field

[0001] This application relates to the fields of affective computing and digital mental health technology, and in particular to an AI-based method and device for detecting the risk of affective dependence based on a multi-dimensional comprehensive score. Background Technology

[0002] With the rapid development of artificial intelligence technology, AI companion products with emotional interaction capabilities, such as Character.ai and Replika, have been widely used. These AI systems offer users an unprecedented companionship experience by providing always-on, always gentle, and highly personalized emotional responses. However, their structured interaction mechanisms, including unconditional availability, precise access to user history information, and predictable emotional feedback, also bring the potential risk of users developing pathological emotional dependence. Between 2024 and 2026, there have been multiple reports globally of serious emotional dependence caused by AI companion products, leading to negative consequences. This makes the effective monitoring and assessment of AI-induced emotional dependence an urgent issue to be addressed.

[0003] Currently, monitoring methods for user interactions with AI primarily focus on screen time statistics or subjective psychological scale assessments. Screen time statistics only reflect the duration of interaction and cannot measure the depth of emotional investment during the interaction, thus failing to distinguish between healthy daily companionship and pathological, deep dependence. For example, an elderly person chatting with AI for an hour daily may be experiencing healthy emotional support, while a teenager chatting with AI for an hour daily could be a warning sign of replacing real social interaction. On the other hand, questionnaire-based psychological scale dependence assessment methods are subjective and lagging, unable to achieve real-time, continuous, and objective monitoring. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and device for detecting AI emotional dependence risk based on a multi-dimensional comprehensive score, which can quantitatively assess the depth and risk level of users' emotional dependence on AI systems and distinguish between healthy companionship and pathological dependence, in order to address the above-mentioned technical problems.

[0005] An AI-based method for detecting the risk of emotional dependence based on multi-dimensional comprehensive scoring, comprising: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal reaction data sequences during the period when the AI ​​system is unavailable; Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on this emotional substitution data sequence, an emotional substitution dimension score was determined to represent the degree of correlation between AI interaction and real social activities. Based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, which represents the degree of favoritism towards the AI ​​system in emotionally low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The dependence risk index (DRI) is generated by weighting and combining the scores of the interaction intensity dimension, the emotional substitution dimension, the vulnerability moment dependence dimension, and the withdrawal reaction dimension. When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

[0006] In one embodiment, based on the interaction intensity data sequence, determining an interaction intensity dimension score characterizing the acceleration of interaction intensity change includes: Interaction frequency, interaction duration, and emotional depth parameters are extracted from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which a user's emotional state deviates from their personal baseline. The second derivatives of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter are calculated within the first preset time period to quantify the acceleration trend of each parameter. The interaction intensity dimension score is generated by combining the second derivative values ​​of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter.

[0007] In one embodiment, based on the emotional substitution data sequence, an emotional substitution dimension score is determined to characterize the degree of inverse correlation between AI interaction and real social activity, including: Based on this emotional substitution data sequence, a first input metric for users interacting with the AI ​​system and a second input metric for users participating in real social activities were determined. Calculate the correlation coefficient between the first input metric and the second input metric within a second preset time period, wherein when the first input metric shows an accelerating growth trend and the second input metric shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches a first threshold. The emotional substitution dimension score is generated based on this correlation coefficient.

[0008] In one embodiment, based on the vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, representing the degree of help-seeking bias of the AI ​​system during emotional low states, including: Identify users' vulnerable moments, which are determined based on users' emotional trajectory data and depression early warning data; The statistics show the percentage of times users initiated requests for help to the AI ​​system during this vulnerable period, out of the total number of requests for help. The vulnerability moment depends on the dimension score, which is generated based on this ratio.

[0009] In one embodiment, based on the withdrawal response data sequence, a withdrawal response dimension score is determined, characterizing the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system, including: In the event that the AI ​​system is detected to be unavailable, the deviation of the user's emotional baseline is obtained through non-contact emotion sensing data. Determine the recovery time required for the magnitude of the emotional baseline deviation to fall back to a preset stable range; The withdrawal response dimension score is generated based on the magnitude of the deviation from the emotional baseline and the recovery time value.

[0010] In one embodiment, the interaction intensity score, emotional substitution score, vulnerability moment dependence score, and withdrawal reaction score are weighted and synthesized to generate a Dependence Risk Index (DRI), including: The Dependency Risk Index (DRI) is calculated using the formula DRI = w1×S1 + w2×S2 + w3×S3 + w4×S4; where S1 is the interaction intensity dimension score, S2 is the emotional substitution dimension score, S3 is the vulnerability moment dependence dimension score, S4 is the withdrawal reaction dimension score, and w1, w2, w3, and w4 are the corresponding preset weight coefficients.

[0011] In one embodiment, when the Dependency Risk Index (DRI) meets preset alarm conditions, a dependency risk alarm message is output, including: Obtain the Dependency Risk Index (DRI) for the current moment; Determine whether the dependency risk index DRI simultaneously meets the first, second, and third conditions; wherein, the first condition is that it exceeds the user's personal baseline DRI range in the historical period; the second condition is that it exceeds the baseline DRI range of the anonymous aggregated population with similar attributes to the user; and the third condition is that it conforms to the clinical baseline DRI pattern corresponding to clinically validated pathological cases. If the first, second, and third conditions are met simultaneously, and the Dependency Risk Index (DRI) shows a continuous upward trend during the continuous monitoring period, then the output of the dependency risk alarm information will be triggered.

[0012] An AI-powered emotional dependence risk detection device based on multi-dimensional comprehensive scoring, the device comprising: The data acquisition module is used to acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the interaction between the user and the artificial intelligence (AI) system within a first preset time period, as well as data sequences of withdrawal reactions during the period when the AI ​​system is unavailable. The first scoring module is used to determine the interaction intensity dimension score, which represents the acceleration of the change in interaction intensity, based on the interaction intensity data sequence. The second scoring module is used to determine the emotional substitution dimension score, which represents the degree of correlation between AI interaction and real social activities, based on the emotional substitution data sequence. The third scoring module is used to determine the vulnerability moment dependence dimension score, which represents the degree of help-seeking bias of the AI ​​system in the emotional low state, based on the vulnerability moment dependence data sequence. The fourth scoring module is used to determine the withdrawal response dimension score based on the withdrawal response data sequence, which represents the degree of deviation of the emotional state and the recovery time caused by the unavailability of the AI ​​system. The index generation module is used to weight and synthesize the interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score and withdrawal reaction dimension score to generate the dependence risk index (DRI). The alarm output module is used to output dependency risk alarm information when the dependency risk index (DRI) meets the preset alarm conditions.

[0013] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal reaction data sequences during the period when the AI ​​system is unavailable; Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on this emotional substitution data sequence, an emotional substitution dimension score was determined to represent the degree of correlation between AI interaction and real social activities. Based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, which represents the degree of favoritism towards the AI ​​system in emotionally low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The dependence risk index (DRI) is generated by weighting and combining the scores of the interaction intensity dimension, the emotional substitution dimension, the vulnerability moment dependence dimension, and the withdrawal reaction dimension. When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal reaction data sequences during the period when the AI ​​system is unavailable; Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on this emotional substitution data sequence, an emotional substitution dimension score was determined to represent the degree of correlation between AI interaction and real social activities. Based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, which represents the degree of favoritism towards the AI ​​system in emotionally low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The dependence risk index (DRI) is generated by weighting and combining the scores of the interaction intensity dimension, the emotional substitution dimension, the vulnerability moment dependence dimension, and the withdrawal reaction dimension. When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

[0015] The aforementioned AI emotional dependence risk detection method and device based on multi-dimensional comprehensive scoring comprehensively analyzes user behavior and physiological data from four independent but complementary dimensions: the acceleration of changes in interaction intensity, the substitution relationship between AI and real social interaction, the tendency to seek help during emotional lows, and the withdrawal reaction when AI is unavailable. This generates a composite Dependence Risk Index (DRI). This approach not only quantifies the depth of dependence but also effectively distinguishes between healthy emotional companionship and pathological emotional dependence through cross-validation of multi-dimensional data and a three-layer baseline reference alarm mechanism. This significantly reduces the false alarm rate, thus solving the technical problem of traditional methods being unable to accurately and in real-time assess and warn of AI emotional dependence risks. Attached Figure Description

[0016] Figure 1 This is an application scenario diagram of an AI-based emotional dependence risk detection method based on multi-dimensional comprehensive scoring in one embodiment; Figure 2 This is a flowchart illustrating an AI-based emotional dependence risk detection method based on multi-dimensional comprehensive scoring in one embodiment. Figure 3 This is a flowchart illustrating the steps for determining the interaction intensity dimension score in one embodiment; Figure 4 This is a flowchart illustrating the steps for determining the scoring of the emotion substitution dimension in one embodiment; Figure 5This is a flowchart illustrating the steps for determining the vulnerability moment-dependent dimension scoring process in one embodiment. Figure 6 This is a flowchart illustrating the steps for determining the withdrawal response dimension scoring in one embodiment; Figure 7 This is a flowchart illustrating the steps for outputting dependency risk warning information in one embodiment; Figure 8 This is a structural block diagram of an AI-based emotional dependence risk detection device based on multi-dimensional comprehensive scoring in one embodiment; Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] The AI-based emotional dependence risk detection method based on multi-dimensional comprehensive scoring provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. The user interacts with the AI ​​system deployed on server 104 or locally on terminal 102 via terminal 102. With user authorization, the data acquisition module on terminal 102 or server 104 acquires data sequences of interaction intensity, emotional substitution, vulnerability moment dependence, and withdrawal reactions during AI system unavailability, including data sequences of these interactions. The processor of terminal 102 or server 104 executes the AI ​​emotional dependence risk detection method based on multi-dimensional comprehensive scoring provided in this application, processes the above data sequences, calculates scores for the four dimensions, and weights and synthesizes them to generate a Dependence Risk Index (DRI). When the DRI meets preset alarm conditions, the system outputs alarm information to the user or an authorized emergency contact via terminal 102. To maximize user privacy, all raw data processing is preferably completed locally on terminal 102, with only anonymized information or hash digests used for alarms uploaded to server 104. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets and portable wearable devices, and server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0019] In one embodiment, such as Figure 2 As shown, an AI-based method for detecting the risk of emotional dependence based on multi-dimensional comprehensive scoring is presented, and this method is applied to... Figure 1 Taking a terminal or server as an example, the following steps are included: Step 202: Obtain the data sequence of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as the data sequence of withdrawal reaction during the period when the AI ​​system is unavailable.

[0020] The interaction intensity data sequence refers to a data set recording the frequency, duration, and emotional investment depth of user interactions with the AI ​​system over time. This includes, but is not limited to, interaction frequency, interaction duration, and emotional depth parameters obtained through voice and text sentiment analysis. The emotional substitution data sequence reflects the user's level of investment in real social activities over time, which can be obtained by analyzing the user's authorized mobile phone usage patterns, calendar events, location trajectory diversity, or social media interaction frequency. The vulnerable moment dependence data sequence records the user's behavior of seeking help from the AI ​​system when identified as being in an emotional low state. Identification of emotional low states can be based on the user's emotional trajectory data and depression warning data. The withdrawal reaction data sequence is a data set recording the degree to which the user's emotional state deviates from the baseline and the recovery process during specific events when the AI ​​system is unavailable due to maintenance, malfunction, or other reasons, using non-contact emotional sensing technology.

[0021] Step 204: Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity.

[0022] Specifically, the system analyzes the interaction intensity data sequence. The key is not to assess the absolute value of the interaction intensity, but rather to evaluate the second derivative of its change, i.e., the acceleration. In healthy relationships, changes in interaction intensity typically fluctuate smoothly; however, in the formation of pathological dependence, the increase in interaction intensity itself exhibits an accelerating trend. Therefore, the interaction intensity dimension score aims to quantify the degree of this accelerated growth as one of the indicators of dependence risk.

[0023] Step 206: Based on the emotional substitution data sequence, determine the emotional substitution dimension score, which represents the degree of correlation between AI interaction and real social activities.

[0024] Specifically, the system calculates the correlation between users' time and effort invested in AI interaction and their investment in real-world social activities by comparing the trends in these two areas. A key pathological pattern is that while investment in AI interaction is accelerating, investment in real-world social activities is systematically decreasing, showing a significant negative correlation in the data. The emotional substitution dimension score aims to quantify this ebb and flow substitution effect.

[0025] Step 208: Based on the vulnerable moment dependency data sequence, determine the vulnerable moment dependency dimension score, which represents the degree of favoritism towards the AI ​​system in emotional low states.

[0026] Specifically, the system identifies users' vulnerable moments and calculates the proportion of users who choose to engage in help-seeking interactions with the AI ​​system during these moments. In a healthy emotional support system, help is sought from diverse sources, including family and friends. If this proportion approaches 1, meaning that users almost exclusively and completely turn to AI for comfort when feeling down, it constitutes a significant characteristic of pathological dependence. The vulnerability moment dependence dimension score aims to quantify this singular, exclusive tendency to seek help.

[0027] Step 210: Based on the withdrawal reaction data sequence, determine the withdrawal reaction dimension score that represents the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system.

[0028] Specifically, the system utilizes events when the AI ​​system is unavailable as a natural experimental window, monitoring changes in the user's physiological and emotional arousal levels during these periods using non-contact emotion sensors. In a healthy relationship, brief unavailability only causes mild and rapidly recovering emotional fluctuations. Pathological dependence, however, manifests as a dramatic deviation in emotional state, such as fluctuations exceeding 40% of baseline, and significantly prolonged recovery times, such as exceeding 24 hours. The withdrawal reaction dimension score aims to quantify the intensity and duration of this withdrawal reaction.

[0029] Step 212: The interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score, and withdrawal reaction dimension score are weighted and combined to generate the Dependence Risk Index (DRI).

[0030] The Dependence Risk Index (DRI) is a comprehensive quantitative indicator ranging from 0 to 100, used to assess the overall risk of a user developing a pathological emotional dependence on the AI ​​system. A higher value indicates a greater risk. By weighting and summing the scores from the four dimensions mentioned above, the system can comprehensively consider multiple aspects of dependence risk, forming a more robust and accurate assessment result than a single indicator. The weighting coefficients can be preset by the system or configured according to regulatory or clinical needs.

[0031] Step 214: When the Dependency Risk Index (DRI) meets the preset alarm conditions, output the dependency risk alarm information.

[0032] The preset alarm conditions are designed to reduce false alarm rates and ensure the accuracy and seriousness of alarms. An alarm will only be triggered when the DRI (Diagnosis Related Group) simultaneously meets multiple conditions in terms of numerical value, time trend, and reference benchmark. Alarm information can be a gentle notification on the user interface or a notification sent to pre-authorized emergency contacts.

[0033] The aforementioned AI emotional dependence risk detection method based on multi-dimensional comprehensive scoring extracts data features from four dimensions: the acceleration of changes in interaction intensity, the substitution relationship between AI and real social interactions, the tendency to seek help during vulnerable moments, and the intensity of withdrawal reactions. These features are then used to generate a Dependence Risk Index (DRI) through weighted comprehensive scoring, achieving a quantitative assessment of AI emotional dependence risk. This method not only measures the surface behavior of users interacting with AI but also deeply analyzes the underlying trends in behavior, the social function substitution effect, behavioral choices under specific psychological states, and withdrawal reactions at the physiological and psychological levels. This effectively distinguishes between healthy companionship and emotional dependence with pathological tendencies, solving the technical problem that traditional methods cannot accurately assess the risk of emotional dependence.

[0034] In one embodiment, such as Figure 3 As shown, based on this interaction intensity data sequence, an interaction intensity dimension score representing the acceleration of interaction intensity change is determined, including: Step 302: Extract the interaction frequency parameter, interaction duration parameter, and emotional depth parameter from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which the user's emotional state deviates from their personal baseline.

[0035] Among them, the interaction frequency parameter refers to the number of times a user interacts with the AI ​​system per unit time; the interaction duration parameter refers to the total interaction time for each interaction or per unit time; and the emotional depth parameter is the degree of user emotional investment quantified by analyzing the emotional features of text or voice in the interaction content. Specifically, it is the deviation of the user's emotional state from their personal long-term emotional baseline level Va.

[0036] Step 304: Calculate the second derivative values ​​of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter within the first preset time period to quantify the acceleration change trend of each parameter.

[0037] For example, the second derivative of the interaction frequency parameter can reveal whether the frequency of user interaction with AI is accelerating. A positive second derivative exceeding a preset threshold for several consecutive weeks is an important signal of pathological dependence.

[0038] Step 306: Combine the second derivative values ​​of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter to generate the interaction intensity dimension score.

[0039] In this embodiment, by calculating the second derivative values ​​of each parameter, the dynamic characteristic of accelerated rather than simple growth of interactive behavior in pathological dependence is captured, so that the interaction intensity dimension score can more accurately reflect the dependence risk, rather than just measuring the absolute amount of interactive behavior.

[0040] In one embodiment, such as Figure 4As shown, based on this emotional substitution data sequence, an emotional substitution dimension score is determined to characterize the degree of correlation between AI interaction and real social activities, including: Step 402: Based on the emotional substitution data sequence, determine the first input metric for the user's interaction with the AI ​​system and the second input metric for the user's participation in real social activities.

[0041] The first input metric can be derived from the aforementioned interaction intensity data; the second input metric can be calculated by analyzing information such as the user's authorized mobile application usage, schedule, and location movement trajectory.

[0042] Step 404: Calculate the correlation coefficient between the first input metric and the second input metric within a second preset time period, wherein when the first input metric shows an accelerating growth trend and the second input metric shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches the first threshold.

[0043] The correlation coefficient measures the strength and direction of the linear relationship between two variables. When AI interaction investment and real social interaction investment show an inverse relationship, the correlation coefficient will be negative, and the closer its absolute value is to 1, the more significant the substitution relationship. The first threshold can be set to 0.7, meaning that when the absolute value of the correlation coefficient exceeds 0.7, a significant substitution effect can be determined.

[0044] Step 406: Generate the emotional substitution dimension score based on the correlation coefficient.

[0045] In this embodiment, the correlation coefficient between AI interaction and real social engagement is calculated to quantify the substitution effect of AI interaction on social function, which is one of the key dimensions for distinguishing healthy use from pathological dependence.

[0046] In one embodiment, such as Figure 5 As shown, based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined to characterize the degree of favoritism towards the AI ​​system in emotionally low states, including: Step 502: Identify the user's vulnerable moments, which are determined based on the user's emotional trajectory data and depression early warning data.

[0047] Emotional trajectory data records users' long-term emotional state fluctuations; depression early warning data can identify periods when users may be in a low mood or at risk of depression through physiological sensors or behavioral pattern analysis.

[0048] Step 504: Calculate the proportion of times a user initiates a request for help to the AI ​​system during this vulnerable moment, relative to the total number of requests for help.

[0049] Requests for help can be identified by analyzing the user's intent to ask for help in the content of their conversation using natural language processing technology.

[0050] Step 506: Generate a vulnerability moment dependency dimension score based on this ratio.

[0051] In this embodiment, by focusing on behavioral choices in the specific context of vulnerable moments, we can more keenly capture the depth of users' emotional dependence on AI, because the person an individual seeks help from during vulnerable moments often reflects their most core source of emotional support.

[0052] In one embodiment, such as Figure 6 As shown, based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system, including: Step 602: In the event that the AI ​​system is detected to be unavailable, the deviation value of the user's emotional baseline is obtained through non-contact emotion sensing data.

[0053] Non-contact emotion sensing data can be derived from sensors such as millimeter-wave radar, WiFi channel status information (CSI), thermal imaging cameras, or electroencephalography (EEG). These sensors can capture physiological signals such as heart rate variability, breathing patterns, subtle changes in body temperature, or brain activity without contacting the user, thereby inferring their emotional arousal level and valence. The emotional baseline deviation value refers to the maximum percentage deviation of a user's emotional arousal level from their calm state baseline during a withdrawal event.

[0054] Step 604: Determine the recovery time required for the emotional baseline deviation value to fall back to the preset stable range.

[0055] Step 606: Generate the withdrawal response dimension score based on the magnitude of the emotional baseline deviation and the recovery time value.

[0056] In this embodiment, unavailable events passively generated by the AI ​​system are used as natural experimental conditions, and non-contact emotion sensing technology is combined to achieve objective and non-intrusive measurement of withdrawal reactions in similar substance dependence, providing key evidence for assessing the physiological and psychological intensity of AI-induced emotional dependence.

[0057] In one embodiment, the interaction intensity score, emotional substitution score, vulnerability moment dependence score, and withdrawal reaction score are weighted and combined to generate the Dependence Risk Index (DRI), which includes: The Dependency Risk Index (DRI) is calculated using the formula DRI = w1×S1 + w2×S2 + w3×S3 + w4×S4; where S1 is the interaction intensity dimension score, S2 is the emotional substitution dimension score, S3 is the vulnerability moment dependence dimension score, S4 is the withdrawal reaction dimension score, and w1, w2, w3, and w4 are the corresponding preset weight coefficients.

[0058] In one specific embodiment, the default values ​​for each weighting coefficient w1, w2, w3, and w4 can all be set to 0.25, meaning that the four dimensions are considered equally important. These weighting coefficients can also be configured by users, regulatory agencies, or mental health professionals based on specific application scenarios or clinical experience to improve the system's flexibility and adaptability. The DRI value range is normalized to the interval of 0 to 100 for easy understanding and setting of a unified alarm threshold.

[0059] In one embodiment, such as Figure 7 As shown, when the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output, including: Step 702: Obtain the Dependency Risk Index (DRI) for the current time.

[0060] Step 704: Determine whether the dependency risk index DRI simultaneously meets the first condition, the second condition, and the third condition; wherein, the first condition is that it exceeds the user's personal baseline DRI range in the historical period; the second condition is that it exceeds the population baseline DRI range of the anonymous aggregated population with similar attributes to the user; and the third condition is that it conforms to the clinical baseline DRI pattern corresponding to clinically validated pathological cases.

[0061] An individual baseline refers to the statistical distribution characteristics of a user's DRI values ​​over the past 6 to 12 months, such as the mean and variance. A population baseline is a statistical distribution obtained by aggregating DRI data from a large number of anonymized users with similar attributes such as age, gender, and occupation using differential privacy technology. A clinical baseline is a library of typical patterns established by analyzing the DRI change patterns in AI-induced emotional dependence cases diagnosed by psychiatrists during the dependence formation process.

[0062] Step 706: If the first condition, the second condition, and the third condition are met simultaneously, and the Dependency Risk Index (DRI) shows a continuous upward trend during the continuous monitoring period, then the output of the Dependency Risk Alarm Information is triggered.

[0063] In this embodiment, a three-layer baseline reference mechanism is introduced to cross-validate from three levels: individual longitudinal comparison, group cross-comparison, and clinical gold standard comparison. This greatly reduces false alarms caused by individual differences, measurement noise, or random fluctuations, ensuring the accuracy and clinical reference value of alarm information. The system will only output an alarm when the DRI points to high risk in all three dimensions: numerical value, trend, and reference system.

[0064] It should be understood that, although Figures 2 to 7 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 2 to 7 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0065] In one embodiment, such as Figure 8 As shown, an AI-based emotional dependence risk detection device based on multi-dimensional comprehensive scoring is provided, including: a data acquisition module 801, a first scoring module 802, a second scoring module 803, a third scoring module 804, a fourth scoring module 805, an index generation module 806, and an alarm output module 807, wherein: The data acquisition module 801 is used to acquire the interaction intensity data sequence, emotional substitution data sequence, vulnerable moment dependence data sequence, and withdrawal reaction data sequence during the period when the AI ​​system is unavailable, within a first preset time period during the interaction between the user and the AI ​​system. The first scoring module 802 is used to determine the interaction intensity dimension score, which characterizes the acceleration of the change in interaction intensity, based on the interaction intensity data sequence. The second scoring module 803 is used to determine the emotional substitution dimension score, which represents the degree of correlation between AI interaction and real social activities, based on the emotional substitution data sequence. The third scoring module 804 is used to determine the vulnerability moment dependence dimension score, which represents the degree of favoritism towards the AI ​​system in the emotional low state, based on the vulnerability moment dependence data sequence. The fourth scoring module 805 is used to determine the withdrawal response dimension score based on the withdrawal response data sequence, which represents the degree of deviation of the emotional state and the recovery time caused by the unavailability of the AI ​​system. The index generation module 806 is used to weight and synthesize the interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score and withdrawal reaction dimension score to generate the dependence risk index DRI. The alarm output module 807 is used to output dependency risk alarm information when the dependency risk index DRI meets the preset alarm conditions.

[0066] In one embodiment, the first scoring module 802 is further configured to extract an interaction frequency parameter, an interaction duration parameter, and an emotional depth parameter from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which a user's emotional state deviates from their personal baseline. The second derivative values ​​of the interaction frequency parameter, the interaction duration parameter, and the emotional depth parameter are calculated within the first preset time period to quantify the acceleration trend of each parameter. The second derivative values ​​of the interaction frequency parameter, the interaction duration parameter, and the emotional depth parameter are combined to generate the interaction intensity dimension score.

[0067] In one embodiment, the second scoring module 803 is further configured to determine, based on the emotional substitution data sequence, a first input metric value for the user's interaction with the AI ​​system and a second input metric value for the user's participation in real social activities; calculate the correlation coefficient between the first input metric value and the second input metric value within a second preset time period, wherein when the first input metric value shows an accelerating growth trend and the second input metric value shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches a first threshold; and generate the emotional substitution dimension score based on the correlation coefficient.

[0068] In one embodiment, the third scoring module 804 is also used to identify the user's vulnerable moments, which are determined based on the user's emotional trajectory data and depression warning data; to count the proportion of the number of times the user initiates help-seeking interactions with the AI ​​system during the vulnerable moment to the total number of all help-seeking interactions; and to generate a vulnerability moment-dependent dimension score based on the proportion.

[0069] In one embodiment, the fourth scoring module 805 is further configured to, in the event that the AI ​​system is detected to be in an unavailable state, acquire the user's emotional baseline deviation value through non-contact emotional sensing data; determine the recovery time value required for the emotional baseline deviation value to fall back to a preset stable range; and generate the withdrawal response dimension score based on the emotional baseline deviation value and the recovery time value.

[0070] In one embodiment, the index generation module 806 is further configured to calculate the dependency risk index DRI according to the formula DRI = w1×S1 + w2×S2 + w3×S3 + w4×S4; where S1 is the interaction intensity dimension score, S2 is the emotional substitution dimension score, S3 is the vulnerability moment dependency dimension score, S4 is the withdrawal reaction dimension score, and w1, w2, w3, and w4 are the corresponding preset weight coefficients.

[0071] In one embodiment, the alarm output module 807 is further configured to obtain the Dependency Risk Index (DRI) at the current moment; determine whether the Dependency Risk Index (DRI) simultaneously meets the first condition, the second condition, and the third condition; wherein, the first condition is that it exceeds the user's personal baseline DRI range in the historical period; the second condition is that it exceeds the population baseline DRI range of an anonymous aggregated population with similar attributes to the user; the third condition is that it conforms to the clinical baseline DRI pattern corresponding to a clinically validated pathological case; if the first condition, the second condition, and the third condition are simultaneously met, and the Dependency Risk Index (DRI) shows a continuous upward trend in the continuous monitoring period, then the output of the Dependency Risk Alarm Information is triggered.

[0072] Specific limitations regarding the AI-based emotional dependence risk detection device based on multi-dimensional comprehensive scoring can be found in the limitations of the AI-based emotional dependence risk detection method based on multi-dimensional comprehensive scoring mentioned above, and will not be repeated here. Each module in the aforementioned AI-based emotional dependence risk detection device based on multi-dimensional comprehensive scoring can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0073] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements an AI-based emotional dependence risk detection method based on multi-dimensional comprehensive scoring. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0074] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0075] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal reaction data sequences during the period when the AI ​​system is unavailable; Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on this emotional substitution data sequence, an emotional substitution dimension score was determined to represent the degree of correlation between AI interaction and real social activities. Based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, which represents the degree of favoritism towards the AI ​​system in emotionally low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The dependence risk index (DRI) is generated by weighting and combining the scores of the interaction intensity dimension, the emotional substitution dimension, the vulnerability moment dependence dimension, and the withdrawal reaction dimension. When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

[0076] In one embodiment, the processor, when executing a computer program, also performs the following steps: Interaction frequency, interaction duration, and emotional depth parameters are extracted from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which a user's emotional state deviates from their personal baseline. The second derivative values ​​of the interaction frequency, interaction duration, and emotional depth parameters are calculated within the first preset time period to quantify the acceleration trend of each parameter. The interaction intensity dimension score is generated by combining the second derivative values ​​of the interaction frequency, interaction duration, and emotional depth parameters.

[0077] In one embodiment, the processor, when executing a computer program, also performs the following steps: Based on the emotional substitution data sequence, a first input metric for user interaction with the AI ​​system and a second input metric for user participation in real social activities are determined; the correlation coefficient between the first input metric and the second input metric is calculated within a second preset time period, wherein when the first input metric shows an accelerating growth trend and the second input metric shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches a first threshold; and an emotional substitution dimension score is generated based on the correlation coefficient.

[0078] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal reaction data sequences during the period when the AI ​​system is unavailable; Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on this emotional substitution data sequence, an emotional substitution dimension score was determined to represent the degree of correlation between AI interaction and real social activities. Based on this vulnerable moment-dependent data sequence, a vulnerable moment-dependent dimension score is determined, which represents the degree of favoritism towards the AI ​​system in emotionally low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score was determined to characterize the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The dependence risk index (DRI) is generated by weighting and combining the scores of the interaction intensity dimension, the emotional substitution dimension, the vulnerability moment dependence dimension, and the withdrawal reaction dimension. When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

[0079] In one embodiment, when the computer program is executed by a processor, it also performs the following steps: Interaction frequency, interaction duration, and emotional depth parameters are extracted from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which a user's emotional state deviates from their personal baseline. The second derivative values ​​of the interaction frequency, interaction duration, and emotional depth parameters are calculated within the first preset time period to quantify the acceleration trend of each parameter. The interaction intensity dimension score is generated by combining the second derivative values ​​of the interaction frequency, interaction duration, and emotional depth parameters.

[0080] In one embodiment, when the computer program is executed by a processor, it also performs the following steps: Based on the emotional substitution data sequence, a first input metric for user interaction with the AI ​​system and a second input metric for user participation in real social activities are determined; the correlation coefficient between the first input metric and the second input metric is calculated within a second preset time period, wherein when the first input metric shows an accelerating growth trend and the second input metric shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches a first threshold; and an emotional substitution dimension score is generated based on the correlation coefficient.

[0081] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM, electrically programmable ROM, electrically erasable programmable ROM, or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM, dynamic RAM, synchronous DRAM, dual data rate SDRAM, enhanced SDRAM, synchronous link DRAM, memory bus direct RAM, direct memory bus dynamic RAM, and memory bus dynamic RAM, etc.

[0082] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An AI-based method for detecting the risk of emotional dependence based on multi-dimensional comprehensive scoring, characterized in that, The method includes: Acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during a first preset time period in the process of user interaction with the artificial intelligence (AI) system, as well as withdrawal response data sequences during periods when the AI ​​system is unavailable. Based on the interaction intensity data sequence, determine the interaction intensity dimension score that characterizes the acceleration of the change in interaction intensity; Based on the emotional substitution data sequence, an emotional substitution dimension score is determined to represent the degree of correlation between AI interaction and real social activities. Based on the vulnerability moment-dependent data sequence, a vulnerability moment-dependent dimension score is determined, which represents the degree of bias towards seeking help from the AI ​​system during emotional low states. Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score is determined, which characterizes the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score, and withdrawal reaction dimension score are weighted and combined to generate the Dependence Risk Index (DRI). When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output.

2. The method according to claim 1, characterized in that, Based on the interaction intensity data sequence, an interaction intensity dimension score characterizing the acceleration of interaction intensity change is determined, including: Interaction frequency parameter, interaction duration parameter, and emotional depth parameter are extracted from the interaction intensity data sequence. The emotional depth parameter is used to characterize the degree to which a user's emotional state deviates from their personal baseline. The second derivatives of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter are calculated within the first preset time period to quantify the acceleration change trend of each parameter. The interaction intensity dimension score is generated by combining the second derivative values ​​of the interaction frequency parameter, interaction duration parameter, and emotional depth parameter.

3. The method according to claim 1, characterized in that, Based on the aforementioned emotional substitution data sequence, an emotional substitution dimension score is determined to characterize the degree of correlation between AI interaction and real social activities, including: Based on the emotional substitution data sequence, a first input metric for the user's interaction with the AI ​​system and a second input metric for the user's participation in real social activities are determined. Calculate the correlation coefficient between the first input metric and the second input metric within a second preset time period, wherein when the first input metric shows an accelerating growth trend and the second input metric shows a systematic decreasing trend, the absolute value of the correlation coefficient approaches a first threshold. The emotional substitution dimension score is generated based on the correlation coefficient.

4. The method according to claim 1, characterized in that, Based on the vulnerability moment-dependent data sequence, a vulnerability moment-dependent dimension score is determined, representing the degree of help-seeking bias towards the AI ​​system during emotional low states, including: Identify users' vulnerable moments, which are determined based on users' emotional trajectory data and depression early warning data; The statistics show the percentage of times users initiated help requests to the AI ​​system during the aforementioned vulnerable moments, relative to the total number of help requests. The vulnerability moment-dependent dimension score is generated based on the stated ratio.

5. The method according to claim 1, characterized in that, Based on the withdrawal reaction data sequence, a withdrawal reaction dimension score is determined, representing the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system, including: In the event that the AI ​​system is detected to be unavailable, the deviation of the user's emotional baseline is obtained through non-contact emotion sensing data; Determine the recovery time required for the magnitude of the emotional baseline deviation to fall back to a preset stable range; The withdrawal response dimension score is generated based on the deviation magnitude of the emotional baseline and the recovery time value.

6. The method according to claim 1, characterized in that, The interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score, and withdrawal reaction dimension score are weighted and synthesized to generate the Dependence Risk Index (DRI), which includes: The Dependency Risk Index (DRI) is calculated according to the formula DRI = w1×S1 + w2×S2 + w3×S3 + w4×S4; where S1 is the interaction intensity dimension score, S2 is the emotional substitution dimension score, S3 is the vulnerability moment dependence dimension score, S4 is the withdrawal reaction dimension score, and w1, w2, w3, and w4 are the corresponding preset weight coefficients.

7. The method according to claim 1, characterized in that, When the Dependency Risk Index (DRI) meets the preset alarm conditions, a dependency risk alarm message is output, including: Obtain the Dependency Risk Index (DRI) at the current moment; Determine whether the Dependency Risk Index (DRI) simultaneously meets the first, second, and third conditions; wherein, the first condition is that it exceeds the user's personal baseline DRI range in the historical period; the second condition is that it exceeds the population baseline DRI range of an anonymous aggregated population with similar attributes to the user; and the third condition is that it conforms to the clinical baseline DRI pattern corresponding to a clinically validated pathological case. If the first, second, and third conditions are met simultaneously, and the Dependency Risk Index (DRI) shows a continuous upward trend during the continuous monitoring period, then the output of the Dependency Risk Alarm Information will be triggered.

8. An AI-based emotional dependence risk detection device based on multi-dimensional comprehensive scoring, characterized in that, The device includes: The data acquisition module is used to acquire data sequences of interaction intensity, emotional substitution, and vulnerability moment dependence during the interaction between the user and the artificial intelligence (AI) system within a first preset time period, as well as data sequences of withdrawal reactions during the period when the AI ​​system is unavailable. The first scoring module is used to determine the interaction intensity dimension score, which characterizes the acceleration of the change in interaction intensity, based on the interaction intensity data sequence. The second scoring module is used to determine the emotional substitution dimension score, which represents the degree of correlation between AI interaction and real social activities, based on the emotional substitution data sequence. The third scoring module is used to determine a vulnerability moment dependence dimension score, which represents the degree of bias towards seeking help from the AI ​​system in an emotionally low state, based on the vulnerability moment dependence data sequence. The fourth scoring module is used to determine a withdrawal response dimension score based on the withdrawal response data sequence, which represents the degree of emotional state deviation and recovery time caused by the unavailability of the AI ​​system. The index generation module is used to weight and synthesize the interaction intensity dimension score, emotional substitution dimension score, vulnerability moment dependence dimension score and withdrawal reaction dimension score to generate the dependence risk index (DRI). The alarm output module is used to output dependency risk alarm information when the dependency risk index (DRI) meets the preset alarm conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images