A method for constructing a network mental health service demand prediction model

Abstract

The application provides a network mental health service demand prediction model construction method, which is used for constructing a network mental health service demand prediction model with better prediction accuracy. The method comprises the following steps: obtaining consultation data of help-seeking behaviors on a mental health service platform and infectious disease case data; extracting mental health topic features of the help-seeking behaviors on the mental health service platform, mental health language features of the help-seeking behaviors on the mental health service platform, and time features of the help-seeking behaviors on the mental health service platform in quantity from the consultation data, and extracting time features of the infectious diseases in case quantity from the infectious disease case data, and taking the four as sample data; training an initial model through the sample data, and taking the trained model as the network mental health service demand prediction model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of prediction, in particular to a method for constructing a network mental health service demand prediction model. BACKGROUND

[0002] Online mental health services (network mental health services) are rated as the best psychological assistance measure, which is beneficial to save time, and more importantly, it avoids face-to-face contact between patients and doctors, which is crucial for successfully containing the spread of infectious diseases.

[0003] Considering the persistent impact of the infectious disease pandemic on the public mental state, it is crucial to establish an interpretable, accurate and early network mental health service demand prediction for the allocation and decision-making of mental health resources, and machine learning techniques have been widely applied in the field of mental health to facilitate the automatic detection of mental illness diagnoses such as depression, and to monitor system trends to predict the outbreak of psychological crises.

[0004] However, in the process of studying the existing related technologies, the inventors found that the existing network mental health service demand prediction schemes have limited prediction accuracy in actual application. SUMMARY

[0005] The present application provides a method for constructing a network mental health service demand prediction model, which can obtain a network mental health service demand prediction model with better prediction accuracy, thereby providing accurate and effective data support for the allocation and decision-making of mental health resources.

[0006] In a first aspect, the present application provides a method for constructing a network mental health service demand prediction model, the method comprising:

[0007] obtaining consultation data of help-seeking behavior on a mental health service platform and infectious disease case data;

[0008] extracting from the consultation data mental health topic features of help-seeking behavior on the mental health service platform, mental health linguistics features of help-seeking behavior on the mental health service platform, and time features of help-seeking behavior on the mental health service platform in terms of quantity, and extracting from the infectious disease case data time features of infectious diseases in terms of case quantity, and taking the four as sample data;

[0009] The initial model is trained by sample data, and the trained model is used as a network mental health service demand prediction model. The network mental health service demand prediction model is used to predict the network mental health service demand in a corresponding time period according to the time characteristics of the number of cases of the input target infectious disease case, the mental health topic characteristics of the help-seeking behavior on the target mental health service platform, the psycholinguistic characteristics of the help-seeking behavior on the target mental health service platform, and the time characteristics of the number of the help-seeking behavior on the target mental health service platform.

[0010] In combination with the first aspect of the present application, in a first possible implementation manner of the first aspect of the present application, in the process of extracting the mental health topic characteristics of the help-seeking behavior on the mental health service platform from the consultation data, the following steps are included:

[0011] The consultation data is converted into vector data mode by neural embedding, and words irrelevant to symptoms and influencing factors are removed by using a domain word library constructed in advance based on the related knowledge of mental diseases and psychological problems.

[0012] In combination with the first aspect of the present application, in a second possible implementation manner of the first aspect of the present application, the network mental health service demand prediction model is specifically used to predict the network mental health service demand within the next two weeks.

[0013] In combination with the first aspect of the present application, in a third possible implementation manner of the first aspect of the present application, the network mental health service demand prediction model adopts an interpretable artificial intelligence framework based on Shapley value, and quantifies the positive and negative influences of different features on the model by Shapley value.

[0014] In combination with the third possible implementation manner of the first aspect of the present application, in a fourth possible implementation manner of the first aspect of the present application, the method further includes:

[0015] According to the Shapley values of different features calculated by the network mental health service demand prediction model, a plurality of feature values having significant influence are determined. The significant influence is determined by the size of the Shapley value or the ranking of the Shapley value within a preset range.

[0016] In combination with the fourth possible implementation manner of the first aspect of the present application, in a fifth possible implementation manner of the first aspect of the present application, the determination of the plurality of feature values having significant influence includes the following features that increase the network mental health service demand:

[0017] High level of trend in the time characteristics of the number of cases of the infectious disease case;

[0018] a high level of biological process, temporal orientation, personal concern, cognitive process, perceptual process, motion, temporal orientation in the psycholinguistic features of the help-seeking behavior on the mental health service platform;

[0019] a high level of distress, depression and anxiety, social phobia in the mental health topic features of the help-seeking behavior on the mental health service platform;

[0020] a high level of trend in the quantity temporal features of the help-seeking behavior on the mental health service platform;

[0021] determining a plurality of feature values that have a significant impact includes the following features that reduce the demand for network mental health services:

[0022] a low level of trend in the quantity temporal features of the infectious disease cases;

[0023] a low level of biological process, temporal orientation, personal concern, cognitive process, perceptual process, motion, temporal orientation in the psycholinguistic features of the help-seeking behavior on the mental health service platform;

[0024] a low level of distress, depression and anxiety, social phobia in the mental health topic features of the help-seeking behavior on the mental health service platform;

[0025] a low level of trend in the quantity temporal features of the help-seeking behavior on the mental health service platform.

[0026] In combination with the first aspect of the application, in a sixth possible implementation manner of the first aspect of the application, after predicting the demand for network mental health services in the corresponding time period through the network mental health service demand prediction model, the method further includes:

[0027] allocating and deciding mental health resources according to the demand for network mental health services in the corresponding time period.

[0028] In a second aspect, the application provides a network mental health service demand prediction model construction device, the device includes:

[0029] an acquisition unit configured to acquire consultation data of help-seeking behavior on a mental health service platform and infectious disease case data;

[0030] an extraction unit configured to extract mental health topic features of help-seeking behavior on the mental health service platform, psycholinguistic features of help-seeking behavior on the mental health service platform, and quantity temporal features of help-seeking behavior on the mental health service platform from the consultation data, and extract quantity temporal features of infectious diseases from the infectious disease case data, and take the four as sample data;

[0031] The training unit is configured to train the initial model by using the sample data, and the model after the training is taken as a network mental health service demand prediction model, which is configured to predict network mental health service demand in a corresponding time period according to time characteristics of a case quantity of an input target infectious disease case, mental health topic characteristics of a help-seeking behavior on a target mental health service platform, mental linguistics characteristics of the help-seeking behavior on the target mental health service platform, and time characteristics of a quantity of the help-seeking behavior on the target mental health service platform.

[0032] In a first possible implementation manner of the second aspect of the present application, the extraction unit is configured to:

[0033] The consultation data is converted into vector data mode by neural embedding, and words irrelevant to symptoms and influencing factors are removed by using a domain word library constructed in advance by using related knowledge of mental diseases and psychological problems.

[0034] In a second possible implementation manner of the second aspect of the present application, the network mental health service demand prediction model is specifically configured to predict network mental health service demand in the next two weeks.

[0035] In a third possible implementation manner of the second aspect of the present application, the network mental health service demand prediction model adopts an explainable artificial intelligence framework based on Shapley values, and quantifies positive and negative influences of different features on the model by using Shapley values.

[0036] In a fourth possible implementation manner of the third possible implementation manner of the second aspect of the present application, the apparatus further includes a determination unit configured to:

[0037] The Shapley values of different features calculated by the network mental health service demand prediction model are used to determine a plurality of feature values having a significant influence, which is determined by the size of the Shapley value or the ranking of the Shapley value within a preset range.

[0038] In a fifth possible implementation manner of the fourth possible implementation manner of the second aspect of the present application, the determination of the plurality of feature values having a significant influence includes the following features that increase network mental health service demand:

[0039] A high-level trend in the time characteristics of the case quantity of the infectious disease case;

[0040] The high level of biological process, time orientation, personal concern, cognitive process, perceptual process, motion, time orientation in the psychological language feature of the help-seeking behavior on the mental health service platform;

[0041] The high level of pain, depression and anxiety, social phobia in the psychological language feature of the help-seeking behavior on the mental health service platform;

[0042] The high level of trend in the quantity time feature of the help-seeking behavior on the mental health service platform;

[0043] The multiple feature values that have a significant impact include the following features that reduce the demand for network mental health services:

[0044] The low level of trend in the case quantity time feature of the infectious disease case;

[0045] The low level of biological process, time orientation, personal concern, cognitive process, perceptual process, motion, time orientation in the psychological language feature of the help-seeking behavior on the mental health service platform;

[0046] The low level of pain, depression and anxiety, social phobia in the psychological language feature of the help-seeking behavior on the mental health service platform;

[0047] The low level of trend in the quantity time feature of the help-seeking behavior on the mental health service platform.

[0048] In combination with the second aspect of the application, in a sixth possible implementation manner of the second aspect of the application, the device further includes a processing unit configured to:

[0049] After predicting the demand for network mental health services in the corresponding time period through the network mental health service demand prediction model, the mental health resources are allocated and decided according to the demand for network mental health services in the corresponding time period.

[0050] In a third aspect, the application provides a processing device including a processor and a memory, the memory storing a computer program, and the processor invoking the computer program in the memory to execute the method provided in the first aspect of the application or any possible implementation manner of the first aspect of the application.

[0051] In a fourth aspect, the application provides a computer readable storage medium, which stores a plurality of instructions, and the instructions are adapted to be loaded by a processor to execute the method provided in the first aspect of the application or any possible implementation manner of the first aspect of the application.

[0052] From the above, the application has the following beneficial effects:

[0053] For the prediction of network mental health service demand, in the configuration of sample data for training the model, in addition to the time characteristics of the number of help-seeking behaviors on the mental health service platform, the time characteristics of the number of cases of infectious diseases, the mental health topic characteristics of help-seeking behaviors on the mental health service platform, and the psychological linguistics characteristics of help-seeking behaviors on the mental health service platform are also configured. Not only the macro trend of mental help-seeking behavior since this time and the influence of the epidemic situation on mental health are considered from the external environment, that is, the time characteristics of the number of help-seeking behaviors and the time characteristics of the number of cases of infectious diseases, but also the mental health topic characteristics and the linguistic characteristics of help-seeking behaviors are considered from the help-seeker. Therefore, a network mental health service demand prediction model with better prediction accuracy and better pertinence for network mental service demand can be trained, so as to provide accurate and effective data support for the allocation and decision of mental health resources. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0055] Figure 1 A flowchart of the construction method of the network mental health service demand prediction model of the present application;

[0056] Figure 2 A scene diagram of the silhouette coefficient of the K-Means model with different cluster numbers of the present application;

[0057] Figure 3 A scene diagram of the construction method of the network mental health service demand prediction model of the present application;

[0058] Figure 4 A structure diagram of the construction device of the network mental health service demand prediction model of the present application;

[0059] Figure 5 A structure diagram of the processing device of the present application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0061] The terms "first", "second", and the like in the description and in the claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged, where appropriate, so that the embodiments described herein can be carried out in sequences other than those illustrated or described herein. Furthermore, the terms "comprise" and "have", and any variations thereof, are intended to cover non-exclusive inclusion, for example, processes, methods, systems, products, or devices that include a list of steps or modules as non- limiting to those steps or modules clearly listed, but can include other steps or modules not clearly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in the present application does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering, and the named or numbered flow steps can be executed in a different order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0062] The division of modules appearing in the present application is a logical division, and in actual application, there can be another division manner, for example, a plurality of modules can be combined or integrated in another system, or some features can be ignored or not executed, in addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, the indirect coupling or communication connection between the modules can be electrical or other similar forms, which are not limited in the present application. Moreover, the modules or sub-modules described as separate components can or can not be physically separate, can or can not be physical modules, or can be distributed into a plurality of circuit modules, and part or all of the modules can be selected according to actual needs to achieve the purpose of the present application scheme.

[0063] Before introducing the construction method of the network mental health service demand prediction model provided by the present application, the background content involved in the present application is first introduced.

[0064] The construction method, device and computer readable storage medium of the network mental health service demand prediction model provided by the present application can be applied to a processing device, for constructing a network mental health service demand prediction model that can obtain better prediction accuracy, so as to provide accurate and effective data support for the allocation and decision of mental health resources.

[0065] The network mental health service demand prediction model construction method mentioned in the present application can be executed by a network mental health service demand prediction model construction device, or a server, a physical host, a user equipment (UE) or other types of processing devices integrated with the network mental health service demand prediction model construction device. The network mental health service demand prediction model construction device can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, a tablet computer, a notebook computer, a desktop computer or a personal digital assistant (PDA). The processing device can be set in a device cluster.

[0066] For example, the processing device can be a background server on a network mental health service platform, which can predict the network mental health service demand of the platform itself or multiple platforms in the future time period behind the network mental health service platform. This can provide accurate and effective data support for the allocation and decision of mental health resources required by the platform to provide network mental health services, and can more perfectly and efficiently utilize mental health resources such as counselor resources, or can perform corresponding operations such as demand warning.

[0067] Next, the network mental health service demand prediction model construction method provided by the present application will be introduced.

[0068] First, referring to Figure 1 , Figure 1 a flowchart of the network mental health service demand prediction model construction method of the present application is shown. The network mental health service demand prediction model construction method provided by the present application can specifically include the following steps S101 to S103:

[0069] Step S101, obtaining consultation data of help-seeking behavior and infectious disease case data on a mental health service platform;

[0070] It can be understood that in order to construct an evaluation model for predicting network mental health service demand, sample data for constructing and training the model needs to be prepared.

[0071] The sample data, which can also be understood as feature data, is the data obtained by processing for training the model. In order to obtain the sample data, the corresponding initial data, i.e. the consultation data of help-seeking behavior and the infectious disease case data on the mental health service platform, needs to be obtained.

[0072] Step S102, extracting the psychological health topic features of the help-seeking behavior on the psychological health service platform, the psychological linguistics features of the help-seeking behavior on the psychological health service platform, and the time features of the help-seeking behavior on the psychological health service platform in quantity from the consultation data, and extracting the time features of the infectious disease cases in quantity from the infectious disease case data, and taking the four as sample data;

[0073] In the present application, the sample data for training the model can be extracted from the consultation data of the help-seeking behavior on the psychological health service platform and the infectious disease case data obtained in the foregoing, and the sample data specifically includes four types of data, 1-time features of infectious disease cases in quantity, 2-psychological health topic features of help-seeking behavior on the psychological health service platform, 3-psychological linguistics features of help-seeking behavior on the psychological health service platform, and 4-time features of help-seeking behavior on the psychological health service platform in quantity.

[0074] To predict the network psychological health service demand in the future period of time, it is easy to understand that the time features are used to complete the prediction function, and it is generally easy to think that the time features of the help-seeking behavior on the psychological health service platform in data, i.e., the time sequence features of the service demand, are used as the basis for training the model.

[0075] As can be found from the above, for predicting the network psychological health service demand in the future period of time, the sample data includes not only the time features of the help-seeking behavior on the psychological health service platform in data, but also the time features of the infectious disease cases in quantity, the psychological health topic features of the help-seeking behavior on the psychological health service platform, and the psychological linguistics features of the help-seeking behavior on the psychological health service platform.

[0076] For the time features of the infectious disease cases in quantity (time sequence features), it can be understood that the present application considers that the external environment and the social environment can affect the network psychological health service demand.

[0077] Previous work found that infectious disease cases can affect the investment and trust behavior of the public, as well as sports activities, and the present application also investigated the network psychological help-seeking behavior of public workers and college students affected by infectious disease cases. However, how the network psychological help-seeking behavior of the public is affected by infectious disease cases has not been understood, that is, there is no method to investigate the change in the quantity of network psychological help-seeking behavior under the influence of infectious disease cases from the perspective of the public.

[0078] For the psychological health topic features and the psychological linguistics features of the help-seeking behavior on the psychological health service platform, it can be understood that the present application considers that the network psychological help-seeking behavior itself can reflect the corresponding service demand features in addition to the time features in quantity.

[0079] Obviously, the topic features and linguistic features are new features that are introduced and further processed based on the help-seeking behavior on the mental health service platform. Different levels of online mental health service demand may have different manifestations in these two aspects.

[0080] Taking the five stages of grief proposed by Kubler-Ross as an example, a person experiencing grief will go through a series of five emotions, including denial, anger, bargaining, depression, and acceptance. With the support of this model, people may experience these emotions in turn, and mental problems related to these emotions may occur at different stages of the pandemic. When seeking help based on a mental health service platform, the mental health topic features of their help-seeking behavior may correspond to the emotional types of the five stages.

[0081] In addition, the present application found that online psychological help-seekers seek help for different psychological problems at different stages of the pandemic, and their expression and words also change accordingly, which indicates that the psychological language features are likely to be an important variable in predicting online mental health service demand.

[0082] As for the above four types of sample data, the time characteristics of the number of cases of infectious diseases can be obtained from official statistical channels, while the mental health topic features, linguistic features, and time characteristics of the number of help-seeking behaviors on the mental health service platform can be obtained from the mental health service platform.

[0083] As an example, the first data source is platform A. On this platform, about 200 million people have consulted mental health services. People can anonymously post their psychological problems in the Q&A community of the platform and seek psychological help and support from psychological consultants. Problem posts can include the following optional content: the title of the problem, the age and gender of the help-seeker, the process of the psychological problem, the inner feelings, the duration of the problem, and the tags (such as occupation, marriage, romantic relationship, family, etc.). The present application scraped 54,797 psychological help-seeking problems from January 31, 2018 to January 8, 2021, of which 3,263 posts mentioned the infectious disease pandemic, with an average of 29.93 online psychological help-seeking per day. Each post contains three parts: title description, psychological problem description, and inquiry time. A report by the platform in 2020 showed that in 2020, the report showed that among the visitors to the online mental health service users, female visitors were more than three times the number of male visitors; visitors in early adulthood (21-35 years old) accounted for 77.57%.

[0084] The second data source is the infectious disease case statistics, including new and cumulative case data.

[0085] The third data source is the B platform, which can collect time series data of the number of daily network psychological help. Since January 31, 2020, the platform has been open to psychological help seekers. From January 31, 2020 to January 8, 2021, time series data of the number of daily network psychological help behaviors were collected, a total of 37,698 network psychological help behaviors.

[0086] Among them, it can be understood that the sample data obtained here can be obtained in real time, or the data obtained before can be called.

[0087] In addition, in the process of obtaining the four types of sample data, the process of screening the four types of sample data from big data can also be involved, which can be adjusted according to the data acquisition method, and is not limited here.

[0088] In the process of extracting the four types of sample data, the corresponding features can be obtained through clustering and other data processing methods. In the specific process of extracting the corresponding feature data, the present application can also perform content screening to filter out meaningless content existing in the rough data acquisition process. For example, the present application can pre-construct a domain word library, which is constructed by using knowledge related to psychological help, that is, constructed by using knowledge related to mental illness and psychological problems. In this way, the content irrelevant to psychological help behavior can be filtered out, and the effective content can be retained. In the process of screening the content, it can also be in the form of vector data mode, so as to facilitate the processing of content screening and promote the conversion of data types, which is convenient for subsequent data processing.

[0089] Taking the process of extracting the psychological health topic features of the help-seeking behavior on the psychological health service platform from the counseling data as an example, as an exemplary implementation manner, in the extraction process, it can include:

[0090] The counseling data is converted into vector data mode by neural embedding, and the domain word library constructed by using knowledge related to mental illness and psychological problems in advance is used to eliminate words irrelevant to symptoms and influencing factors.

[0091] Specifically, in practice, the present application can be used for neural word embeddings, which is a family of techniques for obtaining compact, dense, and continuous vector space representations of entities that can effectively encode the multifaceted relationships between these entities, has become a core component of modern machine learning, and recently has provided new opportunities and solutions for challenging problems, such as language evolution, gender, and stereotypes. In the present application, in order to analyze the psycholinguistic cues (i.e., mental problems and influencing factors) in online psychological help-seeking behavior, the present application proposes a neural embedding method named Word2vec to learn the dense and compact vector space representations of mental health-related terms in online psychological help-seeking problem texts.

[0092] Specifically, first, the present application constructs a predefined word library about mental problems and influencing factors of mental problems, extracts two types of seed words from materials directly related to mental health, and classifies them, such as Kessler 10 and the Patient Health Questionnaire, an emotional lexicon, and a label system in the online mental health service platform.

[0093] Second, the present application constructs a domain word library of the online mental health service community. The present application cuts the text of mental health problems by using the Jieba tool (i.e., a Python Chinese word segmentation package) and the Baidu pause word list, and deletes the pause words. According to the word embedding algorithm, these texts are used as training corpus. The Word2vec in Gensim software constructs a vector model of the pre-trained words of mental health to obtain the domain word library of mental problems and related influencing factors. On the basis of word vectors, the cosine similarity between the words in the vector model and the pre-defined words is calculated to establish the domain dictionary of mental problems and influencing factors. Specifically, the mental health word library contains two parts. (1) 2567 words related to mental problems in online psychological help-seeking. The semantic similarity between these words and the predetermined seed words is greater than 0.3260. (2) 1077 words related to influencing factors of online psychological help-seeking. The semantic similarity between these words and the predetermined seed words is greater than 0.3556.

[0094] Third, the present application also obtains the topics of mental problems and influencing factors of help-seekers to remove the words in the online psychological help-seeking text that are not related to the mental problem or influencing factor dictionary, facilitate the induction of the characteristics of the mental health topic to which they belong, and improve the accuracy and interpretability of topic identification, thereby avoiding mutual interference between the two types of semantics. The word vector representation of mental problems and influencing factors is obtained by the average word embedding method. Based on these text vector representations, the present application uses the k-means clustering algorithm (Python implementation of the K-Means method in scikit-learn) and its evaluation indicators (i.e., the silhouette coefficient) to obtain and evaluate the clustering performance under different numbers of cluster centers.

[0095] As another example, the application attempts to select the number of 4-20 cluster centers, and finally selects the best k-means clustering model with 7 cluster centers for topic detection, and selects the cluster number with the best silhouette coefficient to construct the clusters of psychological problems and influencing factors. The model of different clusters and its silhouette coefficient can be seen in Figure 2 A scenario diagram of the silhouette coefficient of the K-Means model of different clusters of the application is shown. The value range of the silhouette coefficient is -1 to 1. The higher the value, the better the clustering performance. Then, according to the high-frequency keywords related to several clusters, the topics of similar psychological problems and influencing factors are classified, and the content and quantity of the psychological problems and influencing factors of the help seekers are determined.

[0096] Step S103, training the initial model through sample data, and taking the trained model as the network mental health service demand prediction model. The network mental health service demand prediction model is used to predict the network mental health service demand of the corresponding time period according to the time characteristics of the case quantity of the input target infectious disease case, the psychological health topic characteristics of the help-seeking behavior on the target mental health service platform, the psychological linguistics characteristics of the help-seeking behavior on the target mental health service platform, and the time characteristics of the quantity of the help-seeking behavior on the target mental health service platform.

[0097] After obtaining the processed sample data, the construction of the network mental health service demand prediction model can be carried out.

[0098] In the process of constructing or training the model, the principle of taking four types of sample data as independent variables and network mental health service demand as dependent variable is followed, that is, taking the four types of sample data labeled with corresponding network mental health service demand as model training data to train the model to predict the network mental health service demand corresponding to the input data. Thus, when the optimization of model parameters meets the training requirements such as training times, training time, recognition accuracy, etc., the training of the model can be completed. At this time, the model can be put into practical application to predict the network mental health service demand of the input data, and achieve the effect of network mental health service demand prediction.

[0099] Wherein, for the network mental health service demand prediction model, the application adopts an interpretable learning regression method, such as linear regression (LR), ridge regression (RR), least absolute shrinkage and selection operator (LASSO), support vector regression (SVR) and random forest (RF), to establish the network mental health service demand prediction model. In the loss function involved in the model training process, the application can specifically use the mean absolute error (Mean Absolute Error, MAE) and the Pearson correlation coefficient (Pearson Coef) to evaluate the performance of different algorithms and feature sets in the model, and use ten-fold cross-validation method to select the best prediction model.

[0100] From Figure 1 As can be seen from the embodiments shown, for the prediction of network mental health service demand, the application configures not only the time characteristics of the number of help-seeking behaviors on the mental health service platform, but also the time characteristics of the number of cases of infectious diseases, the psychological health topic characteristics of help-seeking behaviors on the mental health service platform, and the psychological linguistics characteristics of help-seeking behaviors on the mental health service platform when configuring sample data for training the model. Not only does it consider the macro trend of psychological help-seeking behavior and the impact of infectious diseases on mental health from the external environment since this period of time, i.e. the time characteristics of the number of help-seeking behaviors in a macro quantity and the time characteristics of the number of cases of infectious diseases, but it also considers the psychological health topic characteristics and linguistic characteristics of help-seeking behaviors from the help-seeker itself. Therefore, a network mental health service demand prediction model with better prediction accuracy and more pertinence to network mental health service demand can be trained, so as to provide accurate and effective data support for the allocation and decision-making of mental health resources.

[0101] In addition, for practical application, the network mental health service demand prediction model of the application needs to predict the number of network psychological help-seeking in a relatively long period of time (one to two weeks), rather than in a short period of time (for example, on the same day or the next day). The application specifically adopts a machine learning regression method to establish network psychological help-seeking quantity prediction models with a lag of 1 day, 3 days, 7 days and 14 days.

[0102] That is, in the application, the network mental health service demand prediction model is specifically used to predict the network mental health service demand within the next two weeks.

[0103] In order to verify the impact of the infectious disease pandemic on the number of network psychological help-seeking, the application uses two network mental health service platforms related to network psychological help-seeking time series data to understand the trend of daily network psychological help-seeking quantity of infectious diseases. The network psychological help-seeking trends of two network mental health service platforms in different lag days.

[0104] The results show that the network psychological help-seeking behavior of the platform increased dramatically after the beginning of the infectious disease pandemic. Specifically, the network psychological help-seeking behavior of the network mental health service community reached a peak in early March, compared with the network psychological help-seeking behavior of the B platform, which reached a peak in mid-March. In addition, as shown in Table 1, the correlation between the time series of the number of network psychological help-seeking in the network mental health service community and the platform was calculated. The number of network psychological help-seeking on the B platform has the strongest correlation with the number of network psychological help-seeking in the network mental health service community, with a lead time of 13 days, reaching 0.585 (N = 343, P < 0.05). The relationship between the daily number of network psychological help-seeking of the two network mental health service platforms during the infectious disease pandemic is shown in Table 1 below. The trend between the two daily network psychological help-seeking numbers also reached a very strong correlation, reaching 0.911 (N = 343, P < 0.05) in 13 days of preparation time.

[0105] Table 1 - Correlation between time series of number of network psychological help-seeking in network mental health service community and platform.

[0106]

[0107] Note: ** P < 0.001.

[0108] By topic modeling of network psychological help-seeking texts, the present application extracts seven psychological problems, seven influencing factors and corresponding keywords. The topics of psychological problems include depression and anxiety, distress, social phobia, lack of interest, worry (fear) and anger. The topics of influencing factors involve love, marriage, psychotherapy, work, interpersonal relationships, personal characteristics and family.

[0109] Table 2 - Average prediction performance of network psychological help-seeking number by different algorithms.

[0110]

[0111] In order to predict the number of network psychological help-seeking on different lag days, the importance of different features and feature sets was studied, and the present application also tried to obtain a regression model with the best performance on the basis of a perfect feature set.

[0112] As shown in Tables 2 and 3, RF achieved the best performance at lag 3 days, with the ratio of MAE to the average number of network psychological help-seeking being 20.03% (5.99 / 29.93*100%). SVR (linear kernel function) achieved the best performance at lag 1 day, 7 days and 14 days. The ratios of MAE to the average number of network psychological help-seeking were 20.11%, 21.14% and 22.84%, respectively. Overall, RF and SVR performed better than other typical regression algorithms.

[0113] Table 3 - Predictive performance of feature set combinations.

[0114]

[0115] Note: P** < 0.001.

[0116] Then, the present application compares the performance of different combinations of the four feature sets based on the RF regressor. As shown in Table 4 below, the temporal feature of network mental health seeking number performs better than other features in the prediction of a single feature set. It is worth noting that the performance of a single feature set decreases as the lag days increase. At any lag day, the combination of all four feature sets shows better performance than any single feature set. However, the combination of the four feature sets does not always show the best performance, for example, although the combination of the four feature sets achieves the best performance at lag 14 days, its performance is not superior to that of the combination of topics, time series, and contagion at lag 1 day, 3 days, and 7 days. In addition, compared with the advanced time series prediction method named "prophet", the prediction model with the four feature sets achieves better performance at lag 3 days and 7 days. In addition, the results between the correlation coefficient and MAE are also similar. It can be seen that the prediction with a long lead time has a high correlation between its predicted value and the true value, although its MAE is high.

[0117] Table 4 - Impact of different feature sets on network mental health seeking behavior at different lag days.

[0118]

[0119] Further, in order to improve the use value of the network mental health service demand prediction model, the present application configures it to use the explainable artificial intelligence framework based on Shapley value, and quantifies the positive and negative effects of different features (different features that the above four types of features can involve) on the model through Shapley value.

[0120] For this Shapley value, which is a widely used method from cooperative game theory, it has desirable properties. The Shapley value represents a feature's responsibility for the change in the model's output, and the Shapley value provides two important benefits. First, global interpretability, that is, the Shapley value can show how much each predictor contributes to the target variable, whether positively or negatively. Second, local interpretability, that is, each observation gets its own set of Shapley values. Traditional variable importance algorithms only show the results for the entire population, not for each individual. Local interpretability allows the present application to accurately point out and compare the impact of various factors. Shapley values greatly increase the transparency of machine learning and have been implemented in many research and industrial scenarios.

[0121] After introducing the Shapley value-based interpretable artificial intelligence framework, the present application can quantify the positive and negative effects of different features on the model using the cumulative positive and negative Shapley values of different feature sets.

[0122] Let the data amount be M, the number of features in the feature set be {1, 2,..., p}, and the total number of features be N, so the positive Shapley value of the feature set is:

[0123] ,

[0124] In the same way, the negative SHAP value can be calculated.

[0125] Thus, the positive or negative effects of different features on the model prediction network mental health service demand discovered in the model construction process can be quantitatively visualized by Shapley value, so as to determine the features that can produce positive or negative effects, and provide data guidance, suggestions, corrections for better mental health resource allocation and decision-making, and the model more clearly explains the importance and impact of the prediction variables.

[0126] Specifically, for the above content, reference can also be made to Figure 3 A scenario schematic diagram of the construction method of the network mental health service demand prediction model of the present application is shown for understanding.

[0127] And after introducing the explainable artificial intelligence framework based on Shapley value, the Shapley values calculated according to different features of the network mental health service demand prediction model can be used to determine the significant influence of multiple feature values, which is determined by the size of the Shapley value or the ranking name within the preset range, so as to provide more specific data guidance, prompt and correction for better mental health resource allocation and decision-making, and the model can more clearly explain the importance and influence of the prediction variable.

[0128] For example, the present application selects effective features for the model from four pre-defined feature sets based on the RF regression and Shapley value of the explainable machine learning framework.

[0129] In order to study the influence of feature sets on the number of online psychological help, the present application calculates the cumulative SHAP value of different feature sets. The results show that the positive and negative prediction of the time feature set of the number of online psychological help is the largest. The prediction of language features is greater than that of the topic as a whole. The prediction of infectious disease cases is greater than that of the topic, but less than that of the language features, but its positive and negative prediction is stronger than that of the language features and the topic feature set at lag 14 days.

[0130] To quantify the cumulative contribution of different features in different predictions, the cumulative Shapley values of the top 20 features in the 1-day, 3-day, 7-day, and 14-day lagged prediction models were calculated. Among these features, the temporal features of the number of online psychological help-seeking (i.e., trend, additive term, year, Yhat), and the infectious disease case-related feature (i.e., PEOPLE POSITIVE CASESCOUNT) were included in the top 20 features of the four models with different lag days. Except for one day, the language features (i.e., Love) were included in the top 20 features of all four lag days. Other top 20 features of different lag days included some features in the language category, such as personal pronouns (i.e., I, SheHe, They), numbers, informal language (i.e., Swear), time orientation (i.e., TenseM, FutureM), social processes (i.e., Friend, Humans), emotional processes (i.e., NegEmo, Anx, Sad), cognitive processes (i.e., certain, Inhibition, Inclusive, Exclusive), perceptual processes (i.e., See, Hear, Bio), biological processes (i.e., Body, Sexual, Ingest), relative (i.e., Relative, Motion), personal concerns (i.e., Work), drives (i.e., Achieve), personal concerns (i.e., Leisure), personal concerns (i.e., Home, Love), time orientations (i.e., tPast, tNow). Some features in the factors influencing mental problems and mental health themes were also among the top 20 features in the specific lag day models, such as depression and anxiety, distress, social phobia, lack of interest, lnLove, work, social interaction, personal characteristics, and family.

[0131] To understand the contribution of the way of features to the prediction performance, the way of the top 20 features to the number of online psychological help-seeking was summarized.

[0132] Results show that the temporal feature of daily online psychological help-seeking quantity (i.e., trend and yhat) positively predicts online psychological help-seeking quantity for all lag days. When the linguistic feature (i.e., Love) is at a low level, it positively predicts online psychological help-seeking quantity, while it negatively predicts online psychological help-seeking quantity at lag 3 days, 7 days, and 14 days. The additive terms, the annual trend of the temporal feature of online psychological help-seeking quantity, the infectious disease cases (i.e., PEOPLE POSITIVE CASES COUNT), the quantity, the biological process (i.e., Body, Ingest), the temporal orientation (i.e., tNow), the personal focus, the cognitive process (i.e., certain), the perceptual process (i.e., Hear), the motion (Motion), the temporal orientation (i.e., In the linguistic feature, the social process (i.e., human), the affective process (i.e., Anx, NegEmo), the perceptual process (i.e., Bio), and in the thematic feature, the distress, depression and anxiety, and social phobia, positively predict online psychological help-seeking quantity when these features are at a high level, while they negatively predict the quantity when they are at a low level.

[0133] Alternatively, the plurality of feature values identified above that have a significant impact include the following features that increase online psychological health service demand:

[0134] a high level of trend in the temporal feature of the number of infectious disease cases;

[0135] a high level of biological process, temporal orientation, personal focus, cognitive process, perceptual process, motion, temporal orientation in the psychological health topic feature of the help-seeking behavior on the psychological health service platform;

[0136] a high level of distress, depression and anxiety, and social phobia in the psycholinguistic feature of the help-seeking behavior on the psychological health service platform;

[0137] a high level of trend in the temporal feature of the number of help-seeking behavior on the psychological health service platform;

[0138] In addition, the plurality of feature values identified above that have a significant impact include the following features that decrease online psychological health service demand:

[0139] a low level of trend in the temporal feature of the number of infectious disease cases;

[0140] a low level of biological process, temporal orientation, personal focus, cognitive process, perceptual process, motion, temporal orientation in the psychological health topic feature of the help-seeking behavior on the psychological health service platform;

[0141] a low level of distress, depression and anxiety, and social phobia in the psycholinguistic feature of the help-seeking behavior on the psychological health service platform;

[0142] The low level of the trend in the number of help-seeking behaviors on the mental health service platform in the temporal characteristics.

[0143] Overall, for the above scheme, this paper established four types of feature sets (i.e., language features, topic features, temporal features of the number of online mental health help-seeking, and temporal features of infectious disease cases), and used machine learning methods (i.e., LR, RR, LASSO, SVR, and RF) to predict and explain the daily number of online mental health help-seeking during the pandemic, and found the following findings:

[0144] First, the daily number of online mental health help-seeking in both the online mental health service community and the C platform increased significantly, and the number of help-seeking in the online mental health service community peaked 13 days earlier than that in the online mental health service platform. In addition, there was a strong positive correlation between the daily number of online mental health help-seeking in the C platform and the daily number of online mental health help-seeking in the B platform, which indicated that the dynamic changes in online mental health help-seeking behavior of the online public were not an exception.

[0145] Second, for the prediction performance of different feature sets, this paper found that under the RF or SVR regression, the model containing the number of online mental health help-seeking, topic features, language features, and temporal features of infectious disease cases achieved the best performance. (1) Although the feature set containing all four types of features performed better overall than any single feature set, it could not always perform the best. For example, when predicting the number of online mental health help-seeking 14 days later, using all four types of features achieved the best performance. However, when predicting the number of online mental health help-seeking 3 days or 7 days later, using only two types of features (i.e., topic and temporal features) could achieve the best performance. This finding can be supported by the principle of feature selection, that is, more features do not necessarily bring better performance due to feature redundancy and irrelevance. (2) This paper found that the temporal features of the number of online mental health help-seeking had an advantage in prediction over other features. For example, models with a lag of 1 day, 3 days, and 14 days showed that the trend of daily online mental health help-seeking was the most important feature, followed by the predicted value and the annual trend generated by the prophet. A possible explanation is that temporal features contain more information, such as periodic and trend changes affected by the environment and events. (3) Compared with language features, topic features were more important and had an increasing impact on the overall performance of models with different lag days, which indicated that language features related to mental health were more targeted for predicting online mental health help-seeking behavior. This can be supported by a previous study that found that language models performed better in documents of about 22 sentences, while topic models performed better in documents of about 2 sentences. Help-seeking posts are usually short and express their mental problems, which means that topic models perform better is reasonable.

[0146] Third, the prediction performance of different lag days, the prediction of network psychological help-seeking for lag days up to two weeks. Compared with the advanced classical prediction method "prophet", the model has an advantage in lag days of 3 days and 7 days, and has the interpretability that the prophet does not have. The prediction model may help to make early, rapid and accurate prediction and explanation of the number of daily network psychological help-seeking in major public health emergencies. At the same time, it can help to reasonably arrange the number of on-duty psychological consultants, and take targeted intervention measures and public policies to prevent potential network public psychological crisis.

[0147] In particular, in the interpretation of the model established in this application, the application finds some meaningful results.

[0148] First of all, in all four models, the first 20 features include trend, additive term, year, yhat, which indicates that these features may be the most important features to predict the number of network psychological help-seeking.

[0149] Second, the results of Shapley value provide possible explanations for black box models, which break the stereotype that machine learning methods are difficult to interpret and understand, and it is crucial to better understand the way features contribute to the performance of the prediction model. For example, when these features are at a high level, the cumulative confirmed cases positively predict the number of network psychological help-seeking, while when they are at a low level, they negatively predict the number of network psychological help-seeking, which indicates that the increase in the number of network psychological help-seeking is affected by the cumulative confirmed cases, but not sensitive to the newly confirmed cases. When predicting the number of network psychological help-seeking with a longer lag day, the effect size of these two features related to infectious diseases becomes larger. Considering the continuous change of individual mental health status, sporadic new cases of infectious diseases may not have a great impact on the network psychological help-seeking behavior of the public. However, the significant change in the social environment has a profound and lasting impact on the public's mental health, and this application shows that this phenomenon is also reflected in the growth of network psychological help-seeking behavior. Therefore, network mental health services should continue to be supported, and the primary problems of network psychological help-seekers such as depression and anxiety, distress, social phobia, lack of interest, worry and fear, anger, etc. should be focused on, and then network psychological assistance forces related to these problems should be cultivated to take targeted intervention measures for network psychological help-seekers in different stages of the infectious disease pandemic.

[0150] Third, other factors with less or moderate influence are also worth attention. (1) The results show that the language cues of biological processes related to physical and interest are associated with the increase of public online psychological help-seeking behavior. This is consistent with previous studies that chronic diseases can lead to poor mental health, so during the epidemic of infectious diseases, online mental health services can be an option for hospitals to deal with mental illness related to traditional physical diseases. (2) The results show that the increase of language cues of perceptual processes related to hearing, and cognitive processes related to certainty are associated with the increase of public online psychological help-seeking behavior. Previous work pointed out that mental health problems are accompanied by abnormal states of individual perception and cognition, and these abnormal problems may be related to the increase of public online psychological help-seeking behavior. (3) The results show that the language cues of topics related to social processes and social phobia are associated with the increase of public online psychological help-seeking behavior. For example, previous studies on adolescents found that individuals with stronger ties to school are less likely to have mental health problems such as depression and anxiety, and this application found that the problem of the connection between individuals and the social environment is related to the increase of public online psychological help-seeking behavior. (4) The results show that the language cues of emotional processes related to anxiety and negative emotions, and the topics related to pain, depression and anxiety are all associated with the increase of online psychological help-seeking behavior. Previous work found that negative emotions significantly affect an individual's mental health, leading to the occurrence of depression, and this application found that these emotional problems are related to the increase of public online psychological help-seeking behavior.

[0151] The above is the introduction of the construction method of the network mental health service demand prediction model provided by the present application. In order to better implement the construction method of the network mental health service demand prediction model provided by the present application, the present application also provides a network mental health service demand prediction model construction device from the functional module angle.

[0152] Reference Figure 4 , Figure 4 is a structural schematic diagram of the network mental health service demand prediction model construction device of the present application. In the present application, the network mental health service demand prediction model construction device 400 can specifically include the following structure:

[0153] The acquisition unit 401 is configured to acquire the consultation data of the help-seeking behavior on the mental health service platform and the infectious disease case data.

[0154] The extraction unit 402 is configured to extract the mental health topic features of the help-seeking behavior on the mental health service platform, the psychological linguistics features of the help-seeking behavior on the mental health service platform, and the time features of the help-seeking behavior on the mental health service platform in quantity from the consultation data, and extract the time features of the infectious disease cases in the number of cases from the infectious disease case data, and take the four as sample data.

[0155] The training unit 403 is configured to train the initial model by using the sample data, and output the trained model as a network mental health service demand prediction model, which is configured to predict the network mental health service demand in a corresponding time period according to the time feature of the number of cases of the target infectious disease case, the mental health topic feature of the help-seeking behavior on the target mental health service platform, the psycholinguistic feature of the help-seeking behavior on the target mental health service platform, and the time feature of the number of the help-seeking behavior on the target mental health service platform.

[0156] In an exemplary implementation, the extraction unit 402 is configured to, in the process of extracting the mental health topic feature of the help-seeking behavior on the mental health service platform from the counseling data, specifically:

[0157] The counseling data is converted into vector data mode by neural embedding, and irrelevant words to symptoms and influencing factors are removed by using a domain word library constructed in advance based on the related knowledge of mental illness and psychological problems.

[0158] In another exemplary implementation, the network mental health service demand prediction model is specifically configured to predict the network mental health service demand in the next two weeks.

[0159] In another exemplary implementation, the network mental health service demand prediction model adopts a Shapley value-based explainable artificial intelligence framework, and uses Shapley values to quantify the positive and negative effects of different features on the model.

[0160] In another exemplary implementation, the apparatus further comprises a determination unit 404 configured to:

[0161] According to the Shapley values calculated by the network mental health service demand prediction model for different features, a plurality of feature values having a significant impact are determined, and the significant impact is determined by the size of the Shapley value or the ranking of the Shapley value within a preset range.

[0162] In another exemplary implementation, the plurality of feature values having a significant impact include the following features that increase the network mental health service demand:

[0163] High level of trend in the time feature of the number of cases of the infectious disease case;

[0164] High level of biological process, time orientation, personal concern, cognitive process, perceptual process, motion, and time orientation in the psycholinguistic feature of the help-seeking behavior on the mental health service platform;

[0165] The mental health themes of help-seeking behavior on mental health service platforms include high levels of distress, depression and anxiety, and social phobia.

[0166] The number of requests for help on mental health service platforms shows a high-level trend in terms of temporal characteristics.

[0167] Several characteristics that identified a significant impact included the following features that reduced the demand for online mental health services:

[0168] The trend of low to medium levels in the number of infectious disease cases over a period of time;

[0169] The psycholinguistic characteristics of help-seeking behavior on mental health service platforms include low-level biological processes, time orientation, personal attention, cognitive processes, perceptual processes, motor processes, and time orientation.

[0170] The psychological health themes of help-seeking behavior on mental health service platforms include low-level distress, depression and anxiety, and social phobia.

[0171] The number of people seeking help on mental health service platforms tends to be at a low level over time.

[0172] In yet another exemplary implementation, the apparatus further includes a processing unit 405, configured to:

[0173] After predicting the demand for online mental health services for a corresponding time period using an online mental health service demand prediction model, mental health resources are allocated and decisions are made based on the demand for online mental health services for that time period.

[0174] This application also provides a processing device from a hardware architecture perspective, see [link / reference]. Figure 5 , Figure 5 This diagram illustrates a structural schematic of the processing device of this application. Specifically, the processing device may include a processor 501, a memory 502, and an input / output device 503. The processor 501 executes the computer program stored in the memory 502 to implement, for example... Figure 1 The corresponding steps of the method for constructing the online mental health service demand prediction model in the embodiment; or, when the processor 501 executes the computer program stored in the memory 502, it implements as follows: Figure 4 Corresponding to the functions of each unit in the embodiment, the memory 502 is used to store the functions executed by the processor 501 as described above. Figure 1 The computer program required for constructing the online mental health service demand prediction model in the corresponding embodiment.

[0175] For example, the computer program can be divided into one or more modules / units, one or more modules / units are stored in the memory 502 and executed by the processor 501 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device.

[0176] The processing device can include, but not limited to, the processor 501, the memory 502, the input and output device 503. Those skilled in the art can understand that the schematic is only an example of the processing device, and does not constitute a limitation on the processing device, which can include more or less components than the schematic, or combine certain components, or different components, for example, the processing device can also include a network access device, a bus, etc., and the processor 501, the memory 502, the input and output device 503 are connected through the bus.

[0177] The processor 501 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the processing device, which connects various parts of the whole device through various interfaces and lines.

[0178] The memory 502 can be used to store computer programs and / or modules, and the processor 501 realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 502, and calling the data stored in the memory 502. The memory 502 can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, at least one application required by a function, etc.; the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device.

[0179] The processor 501 is used to execute the computer program stored in the memory 502, and can specifically implement the following functions:

[0180] Obtain counseling data of the help-seeking behavior on the mental health service platform and infectious disease case data;

[0181] Extract the mental health topic features of the help-seeking behavior on the mental health service platform, the mental health language features of the help-seeking behavior on the mental health service platform, and the time features of the help-seeking behavior on the mental health service platform in quantity from the counseling data, and extract the time features of the infectious disease cases in quantity from the infectious disease case data, and take the four as sample data;

[0182] Train the initial model through the sample data, and take the trained model as the network mental health service demand prediction model, which is used to predict the network mental health service demand in the corresponding time period according to the input time features of the target infectious disease cases in quantity, the mental health topic features of the target help-seeking behavior on the mental health service platform, the mental health language features of the target help-seeking behavior on the mental health service platform, and the time features of the target help-seeking behavior on the mental health service platform in quantity.

[0183] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the construction device of the network mental health service demand prediction model, the processing device and the corresponding units thereof described above can be referred to as Figure 1 The construction method of the network mental health service demand prediction model in the corresponding embodiment will not be repeated here.

[0184] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling related hardware by instructions, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0185] Therefore, the present application provides a computer readable storage medium, which stores a plurality of instructions capable of being loaded by a processor to execute the present application as Figure 1 The steps of the construction method of the network mental health service demand prediction model in the corresponding embodiment can be specifically referred to as Figure 1 The construction method of the network mental health service demand prediction model in the corresponding embodiment will not be repeated here.

[0186] The computer readable storage medium can include a read only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0187] Due to the instructions stored in the computer readable storage medium, the application can be implemented as Figure 1 According to the steps of the method for constructing the network mental health service demand prediction model in the corresponding embodiment, the application can be implemented as Figure 1 The beneficial effects that can be achieved by the method for constructing the network mental health service demand prediction model in the corresponding embodiment are described in detail in the foregoing description, which will not be repeated here.

[0188] The method for constructing the network mental health service demand prediction model, the device, the processing equipment and the computer readable storage medium provided by the application are described in detail above, and the principles and implementation manners of the application are described by applying specific examples in this paper. The above description of the embodiments is only used to help understand the method of the application and its core idea; at the same time, for those skilled in the art, according to the idea of the application, the specific implementation manner and application range will be changed; in view of the above, the content of the specification should not be understood as a limitation of the application.

Claims

1. A method for constructing a demand prediction model for online mental health services, characterized in that, The method includes: Obtain consultation data on help-seeking behavior and infectious disease case data from mental health service platforms; The mental health topic characteristics, psycholinguistic characteristics, and temporal characteristics of the help-seeking behavior on the mental health service platform are extracted from the consultation data. The temporal characteristics of the number of cases of infectious diseases are extracted from the infectious disease case data. All four are used as sample data. The initial model is trained using the sample data, and the trained model is used as a predictive model for the demand for online mental health services. The predictive model is used to predict the demand for online mental health services in the corresponding time period based on the temporal characteristics of the number of target infectious disease cases, the mental health topic characteristics of the help-seeking behavior on the target mental health service platform, the psycholinguistic characteristics of the help-seeking behavior on the target mental health service platform, and the temporal characteristics of the number of help-seeking behavior on the target mental health service platform. The process of extracting mental health topic characteristics of help-seeking behavior on the mental health service platform from the consultation data includes: The consultation data is transformed into vector data patterns through neural embedding, and words that are irrelevant to symptoms and influencing factors are removed by using a domain lexicon built in advance with relevant knowledge of mental illness and psychological problems. The online mental health service demand prediction model is specifically used to predict the demand for online mental health services within the next two weeks. The online mental health service demand prediction model adopts an interpretable artificial intelligence framework based on Shapley values, which quantifies the positive and negative impacts of different features on the model. The method further includes: Based on the Shapley values ​​calculated from the different features using the online mental health service demand prediction model, multiple feature values ​​that have a significant impact are determined. The significant impact is determined within a preset range by the magnitude of the Shapley value or the ranking of the Shapley value. The identified features that significantly increase service demand include the following: The number of infectious disease cases showed a high-level trend over time in terms of case count. The psycholinguistic characteristics of help-seeking behavior on the mental health service platform include high levels of biological processes, time orientation, personal attention, cognitive processes, perceptual processes, and motor activities. The mental health topics identified in the help-seeking behavior on the mental health service platform included high levels of distress, depression and anxiety, and social phobia. The number of requests for help on the mental health service platform shows a high-level trend in terms of temporal characteristics. The identified features that significantly reduce the demand for online mental health services include the following: The number of infectious disease cases showed a low to medium trend over time. The psychological health topic characteristics of the help-seeking behavior on the psychological health service platform include low-level biological processes, time orientation, personal attention, cognitive processes, perceptual processes, and motor activities. The psycholinguistic characteristics of help-seeking behavior on the mental health service platform include low levels of distress, depression and anxiety, and social phobia. The number of people seeking help on the mental health service platform shows a low-level trend in terms of time characteristics.

2. The method according to claim 1, characterized in that, After predicting the demand for online mental health services for a corresponding time period using the online mental health service demand prediction model, the method further includes: Based on the online mental health service needs of the corresponding time period, mental health resources are allocated and decisions are made.

3. A device for constructing a demand prediction model for online mental health services, characterized in that, The device includes: The acquisition unit is used to acquire consultation data on help-seeking behavior and infectious disease case data from the mental health service platform; The extraction unit is used to extract the psychological health topic features of the help-seeking behavior, the psycholinguistic features of the help-seeking behavior, and the temporal features of the help-seeking behavior in terms of quantity from the consultation data, and to extract the temporal features of the number of infectious diseases from the infectious disease case data, and to use the four as sample data. The training unit is used to train an initial model using the sample data and use the trained model as a predictive model for online mental health service demand. The predictive model for online mental health service demand is used to predict the demand for online mental health services in the corresponding time period based on the temporal characteristics of the number of target infectious disease cases, the mental health topic characteristics of the help-seeking behavior on the target mental health service platform, the psycholinguistic characteristics of the help-seeking behavior on the target mental health service platform, and the temporal characteristics of the number of help-seeking behavior on the target mental health service platform. In the process of extracting the mental health topic features of help-seeking behavior on the mental health service platform from the consultation data, the extraction unit is specifically used for: The consultation data is transformed into vector data patterns through neural embedding, and words that are irrelevant to symptoms and influencing factors are removed by using a domain lexicon built in advance with relevant knowledge of mental illness and psychological problems. The online mental health service demand prediction model is specifically used to predict the demand for online mental health services within the next two weeks. The online mental health service demand prediction model adopts an interpretable artificial intelligence framework based on Shapley values, which quantifies the positive and negative impacts of different features on the model. The device further includes a determining unit for: Based on the Shapley values ​​calculated from the different features using the online mental health service demand prediction model, multiple feature values ​​that have a significant impact are determined. The significant impact is determined within a preset range by the magnitude of the Shapley value or the ranking of the Shapley value. The identified features that significantly increase service demand include the following: The number of infectious disease cases showed a high-level trend over time in terms of case count. The psycholinguistic characteristics of help-seeking behavior on the mental health service platform include high levels of biological processes, time orientation, personal attention, cognitive processes, perceptual processes, and motor activities. The mental health topics identified in the help-seeking behavior on the mental health service platform included high levels of distress, depression and anxiety, and social phobia. The number of requests for help on the mental health service platform shows a high-level trend in terms of temporal characteristics. The identified features that significantly reduce the demand for online mental health services include the following: The number of infectious disease cases showed a low to medium trend over time. The psychological health topic characteristics of the help-seeking behavior on the psychological health service platform include low-level biological processes, time orientation, personal attention, cognitive processes, perceptual processes, and motor activities. The psycholinguistic characteristics of help-seeking behavior on the mental health service platform include low levels of distress, depression and anxiety, and social phobia. The number of people seeking help on the mental health service platform shows a low-level trend in terms of time characteristics.

4. A processing device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method as described in claim 1 or 2 when it invokes the computer program in the memory.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the method of claim 1 or 2.

Images