Artificial intelligence-based online assessment of psychological state

Abstract

This application discloses an online psychological state assessment method based on artificial intelligence, relating to the field of artificial intelligence technology. Based on sample multimodal psychological assessment data and their corresponding psychological state labels, the method sequentially employs a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to train the training parameter encoding multiple times, obtaining the target parameter encoding. The method then deploys a psychological state auxiliary assessment model based on the target parameter encoding. Finally, online psychological state assessment is performed based on the deployed psychological state auxiliary assessment model. This significantly improves the accuracy and convergence speed of the psychological state auxiliary assessment model, solving the problems of traditional methods relying on a single modality, low training efficiency, and susceptibility to local optima. It achieves efficient, accurate, and convenient online psychological state assessment.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an online assessment method for mental states based on artificial intelligence. Background Technology

[0002] With the accelerating pace of society, mental health issues are receiving increasing attention. Traditional mental health assessments mainly rely on face-to-face interviews between psychologists and patients and the completion of questionnaires. This approach has the following drawbacks: 1) It is highly subjective, and the assessment results are easily affected by the doctor's experience level and the patient's subjective expression; 2) It is inefficient and requires a large investment of professional human resources; 3) It has poor accessibility and is difficult to meet the needs of large-scale, routine screening.

[0003] In recent years, AI-based psychological assessment technologies have gradually emerged. These technologies typically infer a user's psychological state by analyzing multimodal data such as text, voice, and facial expressions. However, existing technologies still face challenges: First, the fusion and feature extraction of multimodal data are complex, placing high demands on the model architecture; second, the training process of deep learning models is a high-dimensional, non-linear optimization problem, and traditional optimization algorithms such as gradient descent are prone to getting trapped in local optima, resulting in poor model performance and slow training convergence. Summary of the Invention

[0004] This application aims to provide an online psychological state assessment method based on artificial intelligence, which addresses the problems of low training efficiency, susceptibility to local optima, and insufficient assessment accuracy in existing psychological assessment models.

[0005] This application discloses an online psychological state assessment method based on artificial intelligence, comprising: An artificial intelligence model was used to construct a psychological state auxiliary assessment model, and sample multimodal psychological assessment data and their corresponding psychological state labels were obtained by psychologists through human-computer interaction. The training parameters of the psychological state auxiliary assessment model are randomly initialized and encoded to obtain multiple different training parameter codes; Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding. The psychological state auxiliary assessment model is deployed according to the target parameter encoding, and in response to the psychological state assessment request generated by the user through online interaction, the deployed psychological state auxiliary assessment model is scheduled to evaluate the user's multimodal psychological assessment data in the psychological state assessment request and obtain the psychological state assessment result. The psychological state assessment results and corresponding assessment suggestions from the database are fed back to the user, completing the online psychological state assessment based on artificial intelligence.

[0006] In one possible design approach, an artificial intelligence model is used to construct a psychological state-assisted assessment model, including: The text recognition model is constructed using recurrent neural networks, Transformer networks, and convolutional neural networks; the facial expression recognition model is constructed using convolutional neural networks; and / or the voice recognition neural network is constructed using convolutional neural networks. The text recognition model, facial expression recognition model, and / or voice recognition neural network are used as feature extraction models, and a backpropagation neural network is used to construct a feature recognition model. A psychological state auxiliary assessment model is constructed based on the feature extraction model and the feature recognition model.

[0007] In one possible design approach, the training parameters of the psychological state auxiliary assessment model are randomly initialized and encoded to obtain multiple different training parameter codes, including: randomly initializing and encoding the training parameters of the psychological state auxiliary assessment model into vectors between the corresponding upper and lower limits to obtain training parameter codes, and obtaining multiple different training parameter codes.

[0008] In one possible design approach, based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding, including: Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the fitness of the training parameter encoding is obtained; Based on the fitness of the trained parameter encoding, obtain the optimal parameter encoding; Based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-level training on the training parameter encoding to obtain the training parameter encoding after neighborhood-level training. Based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level expansion training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level expansion training. A fast-movement exploration mechanism is used to perform global-level training on the training parameter encoding after neighborhood-level augmentation training, resulting in the training parameter encoding after global-level training. Determine whether the current number of training iterations is greater than or equal to the preset maximum number of training iterations. If so, determine the target parameter encoding based on the training parameter encoding after global-order training; otherwise, return to the step of obtaining fitness.

[0009] In one possible design approach, the fitness of the training parameter encoding is obtained based on the sample multimodal psychological assessment data and its corresponding psychological state labels, including: The sample multimodal psychological assessment data is used as input, and the corresponding psychological state label is used as the expected output to obtain the cross-entropy loss function value. The cross-entropy loss function value is converted into fitness, thus obtaining the fitness encoded by the training parameters.

[0010] In one possible design approach, based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-order training on the training parameter encoding, resulting in the following training parameter encoding after neighborhood-order training:

[0011] in, Indicates the first i The training parameter encoding the first... j One parameter, i =1,2,…,I, where I represents the total number of encoded training parameters. t Indicates the current number of training iterations. j =1,2,…,J, where J represents the total dimension of the parameters encoded during training. The first parameter encoding represents the optimal parameter encoding. j One parameter, This represents the first random number between (0,1).

[0012] In one possible design approach, based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level augmentation training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level augmentation training, including: The exploration space factor is obtained as follows:

[0013] in, Indicates the exploration space factor. Indicates the first j Upper limit of parameters Indicates the first j Lower limit of each parameter; The cluster density factor corresponding to the optimal parameter encoding obtained based on the aforementioned exploration space factor is:

[0014]

[0015] in, This represents the cluster density factor corresponding to the optimal parameter encoding. Indicates the first mThe Euclidean distance between the trained parameter encoding and the optimal parameter encoding after training in neighborhood order 1. This represents an exponential function with base e. A quantification factor representing the degree of influence of the encoding; Based on the cluster density factor, the dynamic search control factor and the current update speed are obtained as follows:

[0016]

[0017] in, Indicates the first m The training parameters encoded after training in the neighborhood order are the ... j One parameter, The first parameter encoding represents the optimal parameter encoding. j One parameter, Indicates the first m The historical best value encoded by the training parameters after training in the nth neighborhood order. j One parameter, express In the t Update speed during each training process; express In the t The update rate during +1 training iterations, i.e., the current update rate; Indicates inertia weight; This represents the dynamic search control factor. Indicates the basic control quantity. represents the control coefficient, T represents the preset maximum number of training iterations, and sin represents the sine function; Based on the dynamic search control factor and the current update rate, the training parameter encoding after neighborhood-level training is further expanded using neighborhood-level training, resulting in the following training parameter encoding after neighborhood-level expansion training:

[0018] in, Indicates the first m The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, This represents the center code corresponding to the training parameter encoding after training with all neighborhood order augmentations. j One parameter, This represents the second random number between (0,1). This represents a third random number between (0, 1).

[0019] In one possible design approach, a fast-movement exploration mechanism is used to perform global-level training on the training parameter encodings after neighborhood-level augmentation training, resulting in global-level training parameter encodings, including:

[0020] in, Indicates the first n The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, Indicates the first n The training parameters encoded after the first global-order training are the first... j One parameter, express Levy The random control factor generated by the motion This represents the fourth random number between (0,1). Indicates the first j The upper limit of each parameter, Indicates the first j The lower bound of each parameter, This represents the fifth random number between (0,1). This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). This represents the eighth random number between (0,1). Represents pi (π). This represents the cosine function.

[0021] In one possible design approach, determining the target parameter encoding based on the training parameter encoding after global-order training includes: obtaining the training parameter encoding after global-order training with the largest fitness value to obtain the target parameter encoding.

[0022] In one possible design approach, the psychological state assessment results and corresponding assessment suggestions from the database are fed back to the user, including: Based on the psychological state assessment results, the system queries the database to find the pre-set association table between psychological state tags and assessment suggestions, obtains the assessment suggestions corresponding to the psychological state assessment results, and feeds back the psychological state assessment results and the obtained assessment suggestions to the user.

[0023] Beneficial effects: This application provides an artificial intelligence-based online psychological state assessment method. Based on multimodal psychological assessment data and their corresponding psychological state labels, the method sequentially employs a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to train the training parameter encoding multiple times, thereby obtaining the target parameter encoding. The method then deploys a psychological state auxiliary assessment model based on the target parameter encoding. Finally, online psychological state assessment is performed based on the deployed psychological state auxiliary assessment model. This method significantly improves the accuracy and convergence speed of the psychological state auxiliary assessment model, and solves the problems of traditional methods relying on a single modality, low training efficiency, and easy getting trapped in local optima. It achieves efficient, accurate, and convenient online psychological state assessment. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of an online psychological state assessment method based on artificial intelligence, proposed in one embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] like Figure 1 As shown in the embodiment of this application, an online psychological state assessment method based on artificial intelligence is provided, including: S101. A psychological state auxiliary assessment model is constructed using an artificial intelligence model, and sample multimodal psychological assessment data and their corresponding psychological state labels are obtained by a psychologist through human-computer interaction.

[0028] A multimodal fusion artificial intelligence model is used to construct a psychological state auxiliary assessment model, which can process various data such as text, voice, and / or images. Simultaneously, a large amount of sample data is collected through a human-computer interaction platform for professional psychologists, including users' multimodal psychological assessment data (such as chat logs, facial expression videos, and voice clips) and authoritative psychological state labels provided by doctors (such as "anxiety," "depression," and "normal"). Sample multimodal psychological assessment data can also include users' audio information of reading test statements aloud, answers to psychological test questionnaires, and facial expression images.

[0029] By integrating multi-dimensional information such as text, facial expressions, and voice, the model can capture complex psychological cues that cannot be reflected by a single modality. For example, a user may appear positive in their text, but their micro-expressions and tone of voice may reveal anxiety. This complementarity significantly reduces the false positive rate and improves the accuracy of the assessment.

[0030] S102. Randomly initialize and encode the training parameters of the psychological state auxiliary assessment model to obtain multiple different training parameter codes.

[0031] All parameters to be trained in the psychological state-assisted assessment model (such as weights and biases in a neural network) are randomly initialized, and each complete set of parameters is encoded into a high-dimensional vector to form an initial population containing multiple different training parameter encodings.

[0032] S103. Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding.

[0033] Traditional algorithms rely on local gradients, making them prone to getting trapped in local optima or saddle points, resulting in slow and unstable convergence. This application's embodiments employ three collaborative mechanisms for optimization: a global guidance mechanism provides macroscopic direction, improving convergence stability; a dual-cluster density neighborhood exploration mechanism actively identifies and escapes local optimum traps, addressing a core pain point; and a fast-movement exploration mechanism enables large-step jumps within safe regions, significantly accelerating training. The combination of these three mechanisms transforms the optimization process from local nearsightedness to global awareness, leading not only to faster convergence but also a higher probability of finding superior solutions. This is particularly suitable for handling complex non-convex problems such as multimodal psychological evaluation, ultimately resulting in more powerful target parameter encoding.

[0034] S104. Deploy the psychological state auxiliary assessment model according to the target parameter encoding, and in response to the psychological state assessment request generated by the user through online interaction, schedule the deployed psychological state auxiliary assessment model to evaluate the user's multimodal psychological assessment data in the psychological state assessment request, and obtain the psychological state assessment result.

[0035] The target parameters are encoded and embedded into the model, which is then deployed on a cloud server. Users assess themselves via a mobile app, which guides them to complete a text description, record a video of facial expressions, and provide audio. This data is sent to the server, where the deployed model analyzes it and returns an assessment result, such as: "Assessment Result: Moderate depressive tendency, confidence level 78%".

[0036] The user multimodal psychological assessment data and the sample multimodal psychological assessment data have the same data structure. It is worth noting that during the testing process, multiple facial expression images may exist; features corresponding to each facial expression image can be extracted and used in the evaluation.

[0037] S105. Feed back the psychological state assessment results and the corresponding assessment suggestions in the database to the user to complete the online psychological state assessment based on artificial intelligence.

[0038] Based on the result of "moderate depressive tendencies," the server retrieved corresponding suggestions from its database: "Your assessment results indicate moderate depressive tendencies, which may affect your daily life. We recommend that you: 1. Contact a professional psychologist for further diagnosis as soon as possible; 2. Try to do 30 minutes of aerobic exercise every day; 3. Communicate your feelings with trusted family and friends." These suggestions were pushed to the user along with the assessment results.

[0039] The output incorporates suggestions from a professional psychological database. This model, combining AI assessment with expert knowledge, leverages the efficiency and objectivity of AI while ensuring the scientific rigor and guidance of the feedback, thereby enhancing the overall system's reliability and user trust.

[0040] Users can conduct self-assessments anytime, anywhere via the internet, without the need for appointments or visits to offline institutions, greatly improving the accessibility of mental health services. The automated assessment process can replace a significant amount of initial screening work, allowing psychologists to focus more on complex cases requiring in-depth intervention, thus optimizing the allocation of medical resources. The online interactive approach provides users with a more private and secure environment for self-expression, helping them to more authentically portray their state, thereby improving the effectiveness of the assessment.

[0041] This application provides an artificial intelligence-based online psychological state assessment method. Based on multimodal psychological assessment data and their corresponding psychological state labels, the method sequentially employs a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to train the training parameter encoding multiple times, thereby obtaining the target parameter encoding. The method then deploys a psychological state auxiliary assessment model based on the target parameter encoding, and finally performs online psychological state assessment based on the deployed model. This significantly improves the accuracy and convergence speed of the psychological state auxiliary assessment model, and solves the problems of traditional methods relying on a single modality, low training efficiency, and easy getting trapped in local optima. It achieves efficient, accurate, and convenient online psychological state assessment.

[0042] In one possible design approach, an artificial intelligence model is used to construct a psychological state-assisted assessment model, including: Text recognition models are constructed using recurrent neural networks, Transformer networks, and convolutional neural networks; facial expression recognition models are constructed using convolutional neural networks; and / or voice recognition neural networks are constructed using convolutional neural networks.

[0043] The text recognition model, facial expression recognition model, and / or voice recognition neural network are used as feature extraction models, and a backpropagation neural network is used to construct a feature recognition model. A psychological state auxiliary assessment model is constructed based on the feature extraction model and the feature recognition model.

[0044] For the features output by the feature extraction model, the fused features are obtained by weighted fusion, fusion into a vector or fusion into a data matrix, and then input into the feature recognition model for recognition. Finally, the feature recognition model can output the evaluation result (i.e., output the predicted label).

[0045] For example, text recognition models use BERT (a type of Transformer network) to analyze the text entered by users in the chat window and extract deep semantic features such as sentiment and theme.

[0046] Facial expression recognition model: EfficientNet (an efficient convolutional neural network) is used to analyze video frames of 10-second facial expression videos uploaded by users to identify micro-expressions (such as downturned corners of the mouth and furrowed brows).

[0047] Voice recognition model: VGGish (a convolutional neural network for audio) is used to analyze the speech of users when answering questions and extract acoustic features such as pitch, speech rate, and energy. The input can be the time domain features or frequency domain features corresponding to the speech.

[0048] Feature recognition and fusion model: A three-layer BP neural network is used. Its input layer receives the feature vectors extracted by the three models mentioned above, performs weighted fusion and advanced feature learning, and the output layer uses the Softmax function to output the probability corresponding to different psychological states (such as "normal", "mild depression", "moderate depression" and "severe depression").

[0049] For example, the text recognition model, facial expression recognition model, and / or voice recognition neural network can all be constructed using convolutional neural networks, and the fixed dimension of the output of the fully connected layer can be used as the output feature.

[0050] In one possible design approach, the training parameters of the psychological state auxiliary assessment model are randomly initialized and encoded to obtain multiple different training parameter codes, including: randomly initializing and encoding the training parameters of the psychological state auxiliary assessment model into vectors between the corresponding upper and lower limits to obtain training parameter codes, and obtaining multiple different training parameter codes.

[0051] In one possible design approach, based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding, including: Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the fitness of the training parameter encoding is obtained; Based on the fitness of the trained parameter encoding, obtain the optimal parameter encoding; Based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-level training on the training parameter encoding to obtain the training parameter encoding after neighborhood-level training. Based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level expansion training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level expansion training. A fast-movement exploration mechanism is used to perform global-level training on the training parameter encoding after neighborhood-level augmentation training, resulting in the training parameter encoding after global-level training. Determine whether the current number of training iterations is greater than or equal to the preset maximum number of training iterations. If so, determine the target parameter encoding based on the training parameter encoding after global-order training; otherwise, return to the step of obtaining fitness.

[0052] In one possible design approach, the fitness of the training parameter encoding is obtained based on the sample multimodal psychological assessment data and its corresponding psychological state labels, including: The sample multimodal psychological assessment data is used as input, and the corresponding psychological state label is used as the expected output to obtain the cross-entropy loss function value. The cross-entropy loss function value is converted into fitness, thus obtaining the fitness encoded by the training parameters.

[0053] For example, you can use 1,000 samples of multimodal psychological assessment data and their corresponding psychological state labels to test, calculate the cross-entropy loss function value, and convert it into fitness (e.g., fitness = 1 / (1 + loss value)).

[0054] In one possible design approach, based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-order training on the training parameter encoding, resulting in the following training parameter encoding after neighborhood-order training:

[0055] in, Indicates the first i The training parameter encoding the first... j One parameter, i =1,2,…,I, where I represents the total number of encoded training parameters. t Indicates the current number of training iterations. j =1,2,…,J, where J represents the total dimension of the parameters encoded during training. The first parameter encoding represents the optimal parameter encoding. j One parameter, This represents the first random number between (0,1).

[0056] Find the optimal parameter encoding with the highest fitness in the current population. Guide all other parameter encodings to move closer to the optimal encoding through a small-scale, fine-grained search. This stage ensures the convergence of the algorithm, enabling the population to quickly converge towards the region of high-quality solutions.

[0057] In one possible design approach, based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level augmentation training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level augmentation training, including: The exploration space factor is obtained as follows:

[0058] in, Indicates the exploration space factor. Indicates the first j Upper limit of parameters Indicates the first j Lower limit of each parameter; The cluster density factor corresponding to the optimal parameter encoding obtained based on the aforementioned exploration space factor is:

[0059]

[0060] in, This represents the cluster density factor corresponding to the optimal parameter encoding. Indicates the first m The Euclidean distance between the trained parameter encoding and the optimal parameter encoding after training in neighborhood order 1. This represents an exponential function with base e. A quantification factor representing the degree of influence of the encoding; Based on the cluster density factor, the dynamic search control factor and the current update speed are obtained as follows:

[0061]

[0062] in, Indicates the first m The training parameters encoded after training in the neighborhood order are the ... j One parameter, The first parameter encoding represents the optimal parameter encoding. j One parameter, Indicates the first m The historical best value encoded by the training parameters after training in the nth neighborhood order. j One parameter, express In the t Update speed during each training process; express In the t The update rate during +1 training iterations, i.e., the current update rate; Indicates inertia weight; This represents the dynamic search control factor. Indicates the basic control quantity. represents the control coefficient, T represents the preset maximum number of training iterations, and sin represents the sine function; Based on the dynamic search control factor and the current update rate, the training parameter encoding after neighborhood-level training is further expanded using neighborhood-level training, resulting in the following training parameter encoding after neighborhood-level expansion training:

[0063] in, Indicates the first m The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, This represents the center code corresponding to the training parameter encoding after training with all neighborhood order augmentations. j One parameter, This represents the second random number between (0,1). This represents a third random number between (0, 1).

[0064] To prevent premature entrapment in local optima after population aggregation, this mechanism is introduced. First, the cluster density around the optimal encoding is calculated; this density reflects whether the population is trapped in a local optimum. Then, the search step size and direction are dynamically adjusted based on the density, and combined with individual historical optimal information and population center information, a broader neighborhood exploration is conducted. This mechanism balances mining and exploration, enhancing the algorithm's ability to escape local optima.

[0065] In one possible design approach, a fast-movement exploration mechanism is used to perform global-level training on the training parameter encodings after neighborhood-level augmentation training, resulting in global-level training parameter encodings, including:

[0066] in, Indicates the first n The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, Indicates the first n The training parameters encoded after the first global-order training are the first... j One parameter, express Levy The random control factor generated by the motion This represents the fourth random number between (0,1). Indicates the first j The upper limit of each parameter, Indicates the first j The lower bound of each parameter, This represents the fifth random number between (0,1). This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). This represents the eighth random number between (0,1). Represents pi (π). This represents the cosine function.

[0067] To further enhance global search capabilities, this mechanism introduces significant random perturbations into some parameter codes. Utilizing stochastic strategies such as Levy flight and cosine functions, the parameter codes can leap to distant, unknown regions in the parameter space for exploration, increasing the probability of discovering the global optimum.

[0068] The three mechanisms described above form a complete closed loop of fine mining, dynamic exploration, and bold exploration, which enables model training to find the global optimal solution with a higher probability, thereby achieving higher evaluation accuracy and stronger generalization ability. This allows for accurate data identification and more accurate assistance to psychologists in conducting automated assessments.

[0069] In one possible design approach, determining the target parameter encoding based on the training parameter encoding after global-order training includes: obtaining the training parameter encoding after global-order training with the largest fitness value to obtain the target parameter encoding.

[0070] In one possible design approach, the psychological state assessment results and corresponding assessment suggestions from the database are fed back to the user, including: Based on the psychological state assessment results, the system queries the database to find the pre-set association table between psychological state tags and assessment suggestions, obtains the assessment suggestions corresponding to the psychological state assessment results, and feeds back the psychological state assessment results and the obtained assessment suggestions to the user.

[0071] For example, based on the result of "moderate depressive tendencies," the server retrieves corresponding suggestions from the database: "Your assessment results show moderate depressive tendencies, which may affect your daily life. We suggest you: 1. Contact a professional psychologist as soon as possible for in-depth diagnosis; 2. Try to do 30 minutes of aerobic exercise every day; 3. Communicate your feelings with trusted family and friends." These suggestions are pushed to the user along with the assessment results.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0073] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0077] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0078] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An online psychological state assessment method based on artificial intelligence, characterized in that, include: An artificial intelligence model was used to construct a psychological state auxiliary assessment model, and sample multimodal psychological assessment data and their corresponding psychological state labels were obtained by psychologists through human-computer interaction. The training parameters of the psychological state auxiliary assessment model are randomly initialized and encoded to obtain multiple different training parameter codes; Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding. The psychological state auxiliary assessment model is deployed according to the target parameter encoding, and in response to the psychological state assessment request generated by the user through online interaction, the deployed psychological state auxiliary assessment model is scheduled to evaluate the user's multimodal psychological assessment data in the psychological state assessment request and obtain the psychological state assessment result. The psychological state assessment results and corresponding assessment suggestions from the database are fed back to the user, completing the online psychological state assessment based on artificial intelligence.

2. The online psychological state assessment method based on artificial intelligence according to claim 1, characterized in that, A psychological state-assisted assessment model is constructed using artificial intelligence models, including: The text recognition model is constructed using recurrent neural networks, Transformer networks, and convolutional neural networks; the facial expression recognition model is constructed using convolutional neural networks; and / or the voice recognition neural network is constructed using convolutional neural networks. The text recognition model, facial expression recognition model, and / or voice recognition neural network are used as feature extraction models, and a backpropagation neural network is used to construct a feature recognition model. A psychological state auxiliary assessment model is constructed based on the feature extraction model and the feature recognition model.

3. The online psychological state assessment method based on artificial intelligence according to claim 1, characterized in that, The training parameters of the psychological state auxiliary assessment model are randomly initialized and encoded to obtain multiple different training parameter codes, including: randomly initializing and encoding the training parameters of the psychological state auxiliary assessment model into vectors between the upper and lower limits to obtain training parameter codes, and obtaining multiple different training parameter codes.

4. The online psychological state assessment method based on artificial intelligence according to claim 1, characterized in that, Based on the sample multimodal psychological assessment data and their corresponding psychological state labels, the training parameter encoding is trained multiple times using a global guidance mechanism, a dual-cluster density neighborhood exploration mechanism, and a fast-moving exploration mechanism to obtain the target parameter encoding, including: Based on the sample multimodal psychological assessment data and its corresponding psychological state labels, the fitness of the training parameter encoding is obtained; Based on the fitness of the trained parameter encoding, obtain the optimal parameter encoding; Based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-level training on the training parameter encoding to obtain the training parameter encoding after neighborhood-level training. Based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level expansion training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level expansion training. A fast-movement exploration mechanism is used to perform global-level training on the training parameter encoding after neighborhood-level augmentation training, resulting in the training parameter encoding after global-level training. Determine whether the current number of training iterations is greater than or equal to the preset maximum number of training iterations. If so, determine the target parameter encoding based on the training parameter encoding after global-order training; otherwise, return to the step of obtaining fitness.

5. The online psychological state assessment method based on artificial intelligence according to claim 4, characterized in that, Based on the sample multimodal psychological assessment data and their corresponding psychological state labels, the fitness of the training parameter encoding is obtained, including: The sample multimodal psychological assessment data is used as input, and the corresponding psychological state label is used as the expected output to obtain the cross-entropy loss function value. The cross-entropy loss function value is converted into fitness, thus obtaining the fitness encoded by the training parameters.

6. The online psychological state assessment method based on artificial intelligence according to claim 4, characterized in that, Based on the optimal parameter encoding, a global guidance mechanism is used to perform neighborhood-order training on the training parameter encoding, resulting in the following training parameter encoding after neighborhood-order training: in, Indicates the first i The training parameter encoding the first... j One parameter, i =1,2,…,I, where I represents the total number of encoded training parameters. t Indicates the current number of training iterations. j =1,2,…,J, where J represents the total dimension of the parameters encoded during training. The first parameter encoding represents the optimal parameter encoding. j One parameter, This represents the first random number between (0, 1).

7. The online psychological state assessment method based on artificial intelligence according to claim 4, characterized in that, Based on the optimal parameter encoding, a dual-cluster density neighborhood exploration mechanism is used to perform neighborhood-level augmentation training on the training parameter encoding after neighborhood-level training, resulting in the training parameter encoding after neighborhood-level augmentation training, including: The exploration space factor is obtained as follows: in, Indicates the exploration space factor. Indicates the first j Upper limit of parameters Indicates the first j Lower limit of each parameter; The cluster density factor corresponding to the optimal parameter encoding obtained based on the aforementioned exploration space factor is: in, This represents the cluster density factor corresponding to the optimal parameter encoding. Indicates the first m The Euclidean distance between the trained parameter encoding and the optimal parameter encoding after training in neighborhood order 1. This represents an exponential function with base e. A factor representing the quantification of the degree of influence of the encoding; Based on the cluster density factor, the dynamic search control factor and the current update speed are obtained as follows: in, Indicates the first m The training parameters encoded after training in the neighborhood order are the ... j One parameter, The first parameter encoding represents the optimal parameter encoding. j One parameter, Indicates the first m The historical best value encoded by the training parameters after training in the nth neighborhood order. j One parameter, express In the t Update speed during each training process; express In the t The update rate during +1 training iterations, i.e., the current update rate; Indicates inertia weight; This represents the dynamic search control factor. Indicates the basic control quantity. represents the control coefficient, T represents the preset maximum number of training iterations, and sin represents the sine function; Based on the dynamic search control factor and the current update rate, the training parameter encoding after neighborhood-level training is further expanded using neighborhood-level training, resulting in the following training parameter encoding after neighborhood-level expansion training: in, Indicates the first m The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, This represents the center code corresponding to the training parameter encoding after training with all neighborhood order augmentations. j One parameter, This represents the second random number between (0,1). This represents a third random number between (0, 1).

8. The online psychological state assessment method based on artificial intelligence according to claim 4, characterized in that, A fast-movement exploration mechanism is used to perform global-order training on the training parameter encodings after neighborhood-order augmentation training, resulting in global-order training parameter encodings, including: in, Indicates the first n The training parameters encoded after the neighborhood order augmentation training are of the th order. j One parameter, Indicates the first n The training parameters encoded after the first global-order training are the first... j One parameter, express Levy The random control factor generated by the motion This represents the fourth random number between (0,1). Indicates the first j The upper limit of each parameter, Indicates the first j The lower bound of each parameter, This represents the fifth random number between (0,1). This represents the sixth random number between (0,1). This represents the seventh random number between (0,1). This represents the eighth random number between (0,1). Represents pi (π). This represents the cosine function.

9. The online psychological state assessment method based on artificial intelligence according to claim 7, characterized in that, Determining the target parameter code based on the training parameter code after global-order training includes: obtaining the training parameter code after global-order training with the largest fitness value to obtain the target parameter code.

10. The online psychological state assessment method based on artificial intelligence according to claim 1, characterized in that, The psychological state assessment results and corresponding assessment suggestions from the database are fed back to the user, including: Based on the psychological state assessment results, the system queries the database to find the pre-set association table between psychological state tags and assessment suggestions, obtains the assessment suggestions corresponding to the psychological state assessment results, and feeds back the psychological state assessment results and the obtained assessment suggestions to the user.

Images