Intelligent Evaluation Method and System for Suicide Ideation Based on Sentence Completion Test

Abstract

The present invention relates to an intelligent evaluation method and system for suicidal ideation based on sentence completion tests, belonging to the cross - technical field of psychometrics and artificial intelligence. This method designs sentences to be completed based on classical suicide theory models, enabling the persons to be evaluated to input completion information according to each sentence to be completed. The completion information is then input into a pre - constructed suicidal ideation evaluation model to obtain the suicidal ideation evaluation result and determine whether the person to be tested has suicidal ideation for early intervention. Among them, the pre - constructed suicidal ideation evaluation model classifies the completion information according to the category and theme features of the completion information by setting a pre - trained language model and an unsupervised topic model, improving the prediction accuracy. Due to the setting of sentences to be completed, the persons to be evaluated can directly complete them according to the sentences to be completed, avoiding the problem of inaccurate data collection caused by social desirability; at the same time, face - to - face interviews are not required, reducing restrictions.

Description

Technical Field

[0001] The present invention relates to the cross - technical field of psychometrics and artificial intelligence, and particularly relates to an intelligent evaluation method and system for suicide ideation based on sentence completion tests. Background Art

[0002] Suicide is a global public health problem. Particularly worrying is the issue of adolescent suicide. Data shows that suicide is the second leading cause of death among young people aged 10 to 24. Compared with childhood, the suicide ideation and behavior of adolescents increase significantly during puberty. Therefore, puberty is a critical period for effective prevention and intervention of suicide. Suicide ideation is the third - strongest predictor of suicide death. Identifying and managing suicide ideation better before it develops into specific behaviors can effectively reduce subsequent suicide attempts and suicide deaths. However, compared with obvious suicide behaviors, suicide ideation is more transient and unstable. Therefore, how to accurately identify suicide ideation is a crucial problem. Early detection and assessment of suicide ideation can effectively prevent suicide.

[0003] In traditional technologies, self - reported self - rating scales and face - to - face interviews are mainly relied on to evaluate the suicide ideation of adolescents. For example, the Columbia Suicide Severity Rating Scale, the Suicide Attitude Questionnaire, the Beck Suicide Ideation Scale, the Suicide Risk Assessment Scale, the Composite International Diagnostic Interview, etc. However, self - rating scales and face - to - face interviews have the following problems: (1) Self - reported tools have the common problem of social desirability. Patients with suicide ideation may conceal information due to concerns about possible negative consequences (such as hospitalization) after reporting truthfully; (2) Compared with self - rating scales, face - to - face interviews are also limited by time, cost, and survey scale; interviews take more time and require higher professional levels of medical staff. Therefore, it is not applicable to large - scale screening scenarios; in addition, the attitude, quality, and experience of interviewers will have a particular impact on interview results.

[0004] Therefore, when traditional technologies are used to assess the suicide intention of adolescents, there are technical problems such as inaccurate data collection and many limitations. For the scenario of active large - scale testing, such as the primary and secondary school groups, there are no other effective means except self - rating scales. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an intelligent evaluation method and system for suicide ideation based on sentence completion tests, so as to overcome the problems of inaccurate data collection and many limitations, resulting in inaccurate evaluation of suicide ideation at present.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] On the one hand, an intelligent evaluation method for suicidal ideation based on a sentence completion test, including:

[0008] Based on the sentence to be completed, obtain the completion information of the sentence to be completed; wherein, the sentence to be completed is obtained according to the classic suicide theory model;

[0009] Input the completion information into the suicidal ideation evaluation model to obtain a suicidal ideation evaluation result corresponding to the completion information;

[0010] Wherein, the suicidal ideation evaluation model is obtained by training based on a completion information sample and a pre-determined suicidal ideation label, and the suicidal ideation label corresponds to the completion information sample one by one;

[0011] The suicidal ideation evaluation model includes a pre-trained language model and an unsupervised topic model, and a connection layer respectively connected to the output ends of the pre-trained language model and the unsupervised topic model; the pre-trained language model is used to obtain the category of the completion information; the unsupervised topic model is used to obtain the topic features of the completion information; the connection layer is used to connect the category and the topic model to predict the suicidal ideation evaluation result of the completion information.

[0012] Optionally, the step of inputting the completion information into the suicidal ideation evaluation model to obtain a suicidal ideation evaluation result corresponding to the completion information includes:

[0013] Encode the sentence to be completed and the corresponding completion information through the pre-trained language model to obtain the category of the completion information;

[0014] Extract the feature of the sentence to be completed and the corresponding completion information through the unsupervised topic model to obtain the topic features of the completion information;

[0015] Connect the completion information, category, and topic features through the connection layer to predict the suicidal ideation evaluation result of the completion information.

[0016] Optionally, the step of encoding the sentence to be completed and the corresponding completion information through the pre-trained language model to obtain the category of the completion information includes:

[0017] Extract the deep representation features of each sentence to be completed and the corresponding completion information based on the pre-trained language model;

[0018] Encode and predict the category of the completion information according to the deep representation features.

[0019] Optionally, the training method of the suicidal ideation evaluation model includes:

[0020] Obtain sentence samples to be completed, corresponding completion information samples, and corresponding suicide ideation labels; wherein, the suicide ideation labels are pre-annotated;

[0021] Iteratively train the initial language model based on the sentence samples to be completed, the corresponding completion information samples, and the corresponding suicide ideation labels; and, iteratively train the initial unsupervised topic model based on the sentence samples to be completed, the corresponding completion information samples, and the corresponding suicide ideation labels;

[0022] Connect the initial language model and the initial unsupervised topic model to the connection layer, output the suicide ideation evaluation result of the completion information, and train to obtain the pre-trained language model and the unsupervised topic model.

[0023] Optionally, in the iterative training process of the initial language model, the following cross-entropy loss function is adopted:

[0024]

[0025] Among them, C i represents the artificially encoded category information, and C i pred represents the category information predicted by the model.

[0026] Optionally, in the iterative training process of the initial language model, the contrastive learning loss function is applied to the completion information with the same encoding and the completion information with different encodings:

[0027] f PLM (A i ) = H i

[0028]

[0029] Among them, A i represents the corresponding completion information sample of the sentence sample to be completed, and H i is the deep vector representation obtained by the pre-trained language model; for each representation H i , the representation with the same encoding as it is denoted as H i + , and the different ones are denoted as H i -; by defining the distance function f d , the contrastive learning loss function makes the deep vector representations of the answers with the same encoding closer.

[0030] Optionally, the iterative training of the initial topic model based on the sentence samples to be completed, the corresponding completion information samples, and the corresponding suicide ideation labels includes:

[0031] Using the following formula, represent the supplemented information in a low-dimensional vector to obtain the representation result of the supplemented information:

[0032] f topic (A) = T

[0033]

[0034] wherein, T represents the representation of the topic dimension obtained through an unsupervised topic model, that is, a vector with a length of t.

[0035] Optionally, inputting the supplemented information into the suicide ideation assessment model to obtain a suicide ideation assessment result corresponding to the supplemented information includes:

[0036] Inputting the category and the topic feature into the connection layer to perform binary classification prediction of suicide ideation to obtain a suicide ideation evaluation result;

[0037] wherein, the Focal loss function is used for prediction in the process of obtaining the suicide ideation evaluation result:

[0038]

[0039] wherein, y represents an individual's true suicide ideation, measured by a professional scale, y^ represents the result predicted by the model from the encoded feature C and the topic feature T, and γ represents the temperature coefficient of the Focal loss function, which can be set according to the data distribution.

[0040] Optionally, obtaining the supplemented information of the to-be-supplemented sentence based on the to-be-supplemented sentence includes:

[0041] Display the to-be-supplemented sentence, and obtain the input information of the user for each to-be-supplemented sentence as the supplemented information; or,

[0042] Voice broadcast the to-be-supplemented sentence, and obtain the input information of the user for each to-be-supplemented sentence as the supplemented information.

[0043] On the other hand, a suicide ideation intelligent assessment system based on a sentence completion test includes a processor and a memory, and the processor is connected to the memory:

[0044] wherein, the processor is used to call and execute the program stored in the memory;

[0045] The memory is used to store the program, and the program is at least used for the suicide ideation intelligent assessment method based on the sentence completion test described in any one of the above.

[0046] The technical solution provided by the present invention at least has the following beneficial effects:

[0047] Based on the classical theoretical model of suicide, incomplete sentences are designed, so that the personnel to be evaluated can input completion information according to each incomplete sentence, and input the completion information into a pre-constructed suicide ideation evaluation model to obtain the suicide ideation evaluation result, and judge whether the person to be tested has suicide ideation, and intervene as early as possible. Among them, the pre-constructed suicide ideation evaluation model is set to include a pre-trained language model and an unsupervised topic model, so as to classify the completion information according to the category and topic characteristics of the completion information, improving the prediction accuracy. Due to the setting of the incomplete sentences, the personnel to be evaluated can directly complete according to the incomplete sentences, avoiding the problem of inaccurate data collection caused by social desirability; at the same time, there is no need for face-to-face interviews, reducing restrictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0049] Figure 1 It is a schematic flowchart of an intelligent suicide ideation evaluation method based on sentence completion test provided by an embodiment of the present invention;

[0050] Figure 2 It is an encoding schematic diagram of a pre-trained language model provided by an embodiment of the present invention;

[0051] Figure 3 It is a working schematic diagram of an unsupervised topic model provided by an embodiment of the present invention;

[0052] Figure 4 It is a working schematic diagram of a suicide ideation evaluation model provided by an embodiment of the present invention;

[0053] Figure 5 It is a training schematic diagram of a suicide ideation evaluation model provided by an embodiment of the present invention;

[0054] Figure 6 It is a structural schematic diagram of an intelligent suicide ideation evaluation system based on sentence completion test provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other implementation manners obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope protected by the present invention.

[0056] As described in the background art, the self-report inventory and face-to-face interview have the following problems: (1) Self-report tools have the general problem of social desirability. Patients with suicidal ideation may conceal information due to concerns about possible negative consequences (such as hospitalization) after reporting truthfully; (2) Compared with self-report inventories, face-to-face interviews are also restricted by time, cost, and survey scale; interviews take more time and require a higher professional level of medical staff, so it is not applicable to large-scale screening scenarios; in addition, the attitude, quality, and experience of the interviewer will have a particular impact on the interview results; (3) Limited by factors such as the instability of suicidal ideation, the data related to suicidal ideation obtained by traditional tools usually come from a single institution or region, making it difficult to obtain a large number of samples and hindering subset analysis based on large-scale demographics, thereby limiting the universality of research results.

[0057] Therefore, traditional technologies have technical problems of inaccurate data collection and many limitations when conducting suicidal intent assessments for teenagers. For the active large-scale testing scenario, such as the primary and secondary school groups, there are no other effective means other than self-report inventories.

[0058] Some researchers have tried to explore the assessment of users' suicidal ideation levels based on social media data, using machine learning methods to automatically analyze the text comment data posted by users on the forum. However, the research cannot reveal the underlying intentions or motives of users' comments, and information such as users' psychological characteristics, offline behaviors, and histories may not be mentioned. In addition, not all patients with suicidal ideation choose to post comments and seek help on forums such as communities, making it difficult to cover a large-scale key population.

[0059] Based on this, the embodiments of the present invention provide a method and system for intelligent assessment of suicidal ideation based on sentence completion tests, which can be applied to the active large-scale testing scenario to test the suicidal ideation of teenagers, such as the primary and secondary school groups.

[0060] Figure 1 For the flowchart of a method for intelligent assessment of suicidal ideation based on sentence completion tests provided by an embodiment of the present invention, please refer to Figure 1 , this embodiment may include the following steps:

[0061] Step S11: Obtain the completion information of the sentence to be completed based on the sentence to be completed; wherein, the sentence to be completed is obtained according to the classical suicide theory model.

[0062] Step S12: Input the completion information into the suicide ideation assessment model to obtain a suicide ideation assessment result corresponding to the completion information.

[0063] Among them, the suicide ideation assessment model is obtained by training based on a completion information sample and a pre-determined suicide ideation label, and the suicide ideation label corresponds to the completion information sample one by one.

[0064] The suicide ideation assessment model includes a pre-trained language model and an unsupervised topic model, and a connection layer connected to the output ends of the pre-trained language model and the unsupervised topic model respectively; the pre-trained language model is used to obtain the category of the completion information; the unsupervised topic model is used to obtain the topic features of the completion information; the connection layer is used to connect the category and the topic model to predict and obtain the suicide ideation evaluation result of the completion information.

[0065] The intelligent suicide ideation assessment method based on sentence completion test provided by the present application can be applied to the primary and secondary school groups for suicide ideation assessment. It should be noted that the intelligent suicide ideation assessment method based on sentence completion test of the present application can be integrated into software, and after each terminal downloads the software, each person to be assessed can use it.

[0066] For example, the terminal can prompt each sentence to be completed, and the person to be assessed completes the sentence to be completed. The completion information is input into the suicide ideation assessment model to obtain a suicide ideation assessment result corresponding to the completion information. Among them, the suicide ideation assessment result can be having suicide ideation or not having suicide ideation.

[0067] It should be noted that there can be 12 sentences to be completed, which are obtained according to the classical suicide theory model. The specific obtaining method can be: (1) First, according to the classical suicide theory in psychology, a series of questions are set, the answers of at least 1000 primary and secondary school students are collected, and the suicide ideation levels (that is, labels) of these people are obtained through existing other tools; (2) Process the data of more than 1000 people: (2.1) Determine which questions are valid through expert annotation of the answers, and then select 12; (2.2) Use the answers and labels of these 12 questions to construct a suicide ideation assessment model with the data of more than 1000 people.

[0068] During the evaluation, each person to be evaluated inputs 12 pieces of completion information for 12 sentences to be completed, and then inputs the completion information into a pre-constructed suicide ideation evaluation model, and the recognition result of the suicide ideation of the person to be evaluated can be obtained.

[0069] In some embodiments, obtaining the completion information of the sentence to be completed based on the sentence to be completed includes:

[0070] Displaying the sentence to be completed, and obtaining the input information of each sentence to be completed by the user as the completion information; or,

[0071] Voice-playing the sentence to be completed, and obtaining the input information of each sentence to be completed by the user as the completion information.

[0072] For example, the sentence to be completed can be displayed on the screen, and the person to be evaluated completes it by inputting the completion information; or the sentence to be completed can be voice-played, and the person to be evaluated makes a voice reply, and the terminal collects the voice reply and performs voice recognition, and takes the recognized content as the completion information.

[0073] It can be understood that in the technical solution of this embodiment, the sentence to be completed is designed based on the classic suicide theory model, so that the person to be evaluated can input the completion information according to each sentence to be completed, input the completion information into the pre-constructed suicide ideation evaluation model, conduct the suicide ideation evaluation result, and judge whether the person to be tested has suicide ideation, and intervene as early as possible. Among them, the pre-constructed suicide ideation evaluation model is set to include a pre-trained language model and an unsupervised topic model, so as to classify the completion information according to the category and topic characteristics of the completion information, and improve the prediction accuracy. Since the sentence to be completed is set, the person to be evaluated can directly complete it according to the sentence to be completed, avoiding the problem of inaccurate data collection caused by social desirability; at the same time, there is no need for face-to-face interviews, reducing restrictions.

[0074] In some embodiments, inputting the completion information into the suicide ideation evaluation model to obtain the suicide ideation evaluation result corresponding to the completion information includes:

[0075] Encoding the sentence to be completed and the corresponding completion information through the pre-trained language model to obtain the category of the completion information;

[0076] Extracting the feature of the sentence to be completed and the corresponding completion information through the unsupervised topic model to obtain the topic feature of the completion information;

[0077] Connecting the completion information, category and topic feature through the connection layer to predict the suicide ideation evaluation result of the completion information.

[0078] Specifically, the processing process of the suicide ideation assessment model for the supplemented information can be as follows: Through the pre-trained language model, the to-be-supplemented sentence and the corresponding supplemented information are encoded to obtain the category of the supplemented information. For example, the to-be-supplemented sentence can be "I was (not serious in learning) in the past week", and the content in the parentheses is the corresponding supplemented information, which can be encoded as the category of "learning input", and "(always very happy) in the past week" can be encoded as "positive emotion", etc.

[0079] Through the unsupervised topic model, the to-be-supplemented sentence and the corresponding supplemented information are subjected to feature extraction to obtain the topic features of the supplemented information. Through the connection layer, the category and topic features of the supplemented information are connected to predict the suicide ideation evaluation result of the supplemented information. Among them, the connection layer can be a linear regression layer.

[0080] It can be understood that by adopting the technical solution provided in this embodiment, through the setting of the pre-trained language model, the unsupervised topic model, and the connection layer, the multi-dimensional extraction of the features of the supplemented information is realized, and the prediction accuracy is improved.

[0081] In some embodiments, the encoding of the to-be-supplemented sentence and the corresponding supplemented information through the pre-trained language model to obtain the category of the supplemented information includes:

[0082] Based on the pre-trained language model, the deep representation features of each to-be-supplemented sentence and the corresponding supplemented information are extracted;

[0083] According to the deep representation features, the category of the supplemented information is encoded and predicted.

[0084] It can be understood that by adopting the technical solution of this embodiment, the prediction accuracy is improved by extracting the deep representation features and categories of the supplemented information.

[0085] Among them, each to-be-supplemented text can be a question, and each supplemented information can be an answer. Taking questions and answers as examples, the technical solution of this embodiment is further described. Specifically, Figure 2 is an encoding schematic diagram of a pre-trained language model provided by an embodiment of the present invention. Refer to Figure 2 , each question corresponds to an answer. Through the pre-trained language model, the vectors of each question and answer are extracted, and each question is independently encoded. Each encoding corresponds to a category, and the encoding is predicted through a supervised prediction model.

[0086] Figure 3 is a working schematic diagram of an unsupervised topic model provided by an embodiment of the present invention. Refer to Figure 3 , for each question and answer, the topic features can be extracted through the unsupervised topic model to represent the answer text.

[0087] Figure 4 The working schematic diagram of a suicide ideation assessment model provided by an embodiment of the present invention is shown in Figure 4 , and the category and theme features of the answers to multiple questions of a respondent are connected to obtain the suicide ideation evaluation result.

[0088] In some embodiments, the training method of the suicide ideation assessment model includes:

[0089] Obtaining a sentence sample to be completed, a corresponding completion information sample, and a corresponding suicide ideation label; wherein, the suicide ideation label is pre-annotated;

[0090] Based on the sentence sample to be completed, the corresponding completion information sample, and the corresponding suicide ideation label, iteratively train the initial language model; and, based on the sentence sample to be completed, the corresponding completion information sample, and the corresponding suicide ideation label, iteratively train the initial topic model;

[0091] Connect the initial language model and the initial unsupervised topic model to the connection layer, output the suicide ideation evaluation result of the completion information, and train to obtain the pre-trained language model and the unsupervised topic model.

[0092] For example, the above 12 questions can be used as the sentence samples to be completed, the corresponding answers can be used as the completion information samples, and each completion information sample can be annotated, for example, annotating whether there is suicide ideation. After preparing these sample information, iteratively train the initial language model and, iteratively train the initial topic model; after encoding by the initial language training model, the category is obtained, and the theme feature is obtained by the initial topic model. Both the category and the theme feature enter the connection layer, and according to the marked suicide ideation evaluation result, the final pre-trained language model and the unsupervised topic model are trained.

[0093] Specifically, Figure 5 The training schematic diagram of a suicide ideation assessment model provided by an embodiment of the present invention is shown in Figure 5 , after the initial language model extracts the deep representations in the questions and answers, through manual coding, continuous comparative learning is carried out for iterative training.

[0094] The input of the model is n question topics and the corresponding sentence completion answer content, denoted as:

[0095] Q = {Q1, Q2,..., Q n}},

[0096] A = {A1, A2,..., A n}},

[0097] wherein, Qn is the nth question; A n is the response corresponding to the nth question.

[0098] To represent the individual response results, based on two assumptions: (1) In each question of the test, the responses of the test taker may involve different response patterns, and (2) The same test taker may express similar content in different questions. Two text representation techniques are used in the same model:

[0099] On the one hand, the text is encoded manually. For example, "(I was not serious in my studies in the past week)" can be encoded into the "learning input" category, and "(I was very happy in the past week)" can be encoded into the "positive emotion" category. The BERT pre-trained language model (PLM) and the linear regression layer are used to represent the encoded text, that is:

[0100]

[0101] Among them, the parameters in the pre-trained language model can be fine-tuned according to the data, and the linear regression layer needs to learn according to the data. To achieve accurate prediction of the encoding, the cross-entropy loss function of the encoding result is introduced in the model training, that is: In the iterative training process of the initial language model, the following cross-entropy loss function is adopted:

[0102]

[0103] Among them, C i represents the category information of the manual encoding, and C i pred represents the category information predicted by the model.

[0104] In some embodiments, in the iterative training process of the initial language model, the contrastive learning loss function is applied to the completion information of the same encoding and the completion information of different encodings:

[0105] f PLM (A i ) = H i

[0106]

[0107] Among them, A i represents the completion information sample, and H i is the deep vector representation obtained from the pre-trained language model. For each representation H i , the representation with the same encoding as it is denoted as H i + , and the different one is denoted as H i - . By defining the distance function f d, the contrastive learning loss function makes the deep vector representations of responses with the same encoding closer to each other.

[0108] On the other hand, the LDA topic model is used to represent the text as a low-dimensional vector. That is, based on the sentence completion sample, the corresponding completion information sample, and the corresponding suicide ideation label, the initial topic model is iteratively trained, including:

[0109] Using the following formula, the completion information is represented as a low-dimensional vector to obtain the representation result of the completion information:

[0110] f topic (A) = T

[0111]

[0112] where T represents the representation of the topic dimension obtained through the unsupervised topic model, that is, a vector of length t.

[0113] Based on the representation results of the input data, the numerical values of the encoding and the topic are uniformly input into the linear regression layer for binary classification prediction of suicide ideation. Based on the constructed dataset, the cross-entropy loss function is also used at the suicide ideation level. That is, connecting the pre-trained language model and the unsupervised topic model through the connection layer and outputting the suicide ideation evaluation result of the completion information, including:

[0114] Inputting the category and the topic features into the connection layer for binary classification prediction of suicide ideation to obtain the suicide ideation evaluation result;

[0115] Among them, the Focal loss function is used for prediction in the process of obtaining the suicide ideation evaluation result:

[0116]

[0117] where y represents the individual's true suicide ideation, measured by a professional scale, y^ represents the result predicted by the model from the encoding feature C and the topic feature T, and γ represents the temperature coefficient of the Focal loss function, which can be set according to the data distribution.

[0118] It can be understood that for the problem of automated scoring of suicide ideation, based on the text data collected in the sentence completion test, this system realizes an automated scoring model based on text data through two types of text representation methods of encoding and topic. Experiments show that this model can identify more than 40% of individuals with suicide ideation.

[0119] An embodiment of the present invention provides a large-scale automated assessment system for suicidal ideation among primary and secondary school students and adolescents. The greatest value of this method lies in its ability to avoid the problem of social desirability in traditional assessment tools and enable rapid large-scale testing and assessment of the target group without the need for expert scoring. Traditional self-report tools are difficult to avoid the problem of patients concealing information. In particular, the interview method is also limited by the professional level of the interviewer and cannot achieve large-scale testing. Research based on text data from social media cannot explain the underlying motivations behind user comments, and the data obtained is only from a part of the target group. Compared with existing solutions, the open-ended questions of the sentence completion test used in this system can reflect the deeper inner reactions and thoughts of patients, and combined with machine learning, it realizes automated scoring of text responses, solving the limitation of test scoring on the professional level of scorers, and finally achieving large-scale testing for this specific group of adolescents.

[0120] Based on a general inventive concept, an embodiment of the present invention also provides an intelligent evaluation system for suicidal ideation based on a sentence completion test to implement the above method embodiment.

[0121] Figure 6 is a structural schematic diagram of an intelligent evaluation system for suicidal ideation based on a sentence completion test provided by an embodiment of the present invention. Refer to Figure 6 , the intelligent evaluation system for suicidal ideation based on a sentence completion test in this embodiment includes a processor 61 and a memory 62, and the processor 61 is connected to the memory 62. Among them, the processor 61 is used to call and execute the program stored in the memory 62; the memory 62 is used to store the program, and the program is at least used to execute the intelligent evaluation method for suicidal ideation based on a sentence completion test in the above embodiments.

[0122] The specific implementation of the intelligent evaluation system for suicidal ideation based on a sentence completion test provided by the embodiments of the present application can refer to the implementation manner of the intelligent evaluation method for suicidal ideation based on a sentence completion test in any of the above embodiments, and will not be elaborated here.

[0123] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and the content not detailed in some embodiments can be referred to the same or similar content in other embodiments.

[0124] It should be noted that in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "a plurality" is at least two.

[0125] Any process or method description depicted in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations where functions may be performed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0126] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0127] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried out in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program. The said program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0128] In addition, in each embodiment of the present invention, the functional units can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0129] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disc, or the like.

[0130] In the description of this specification, descriptions with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0131] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent evaluation method for suicidal ideation based on sentence completion tests, characterized in that, Including: Based on the sentence to be completed, obtain the completion information of the sentence to be completed; wherein, the sentence to be completed is obtained according to the classic suicide theory model; Input the completion information into the suicide ideation assessment model to obtain a suicide ideation assessment result corresponding to the completion information; Wherein, the suicide ideation assessment model is obtained by training based on a completion information sample and a pre-determined suicide ideation label, and the suicide ideation label corresponds one-to-one with the completion information sample; The suicide ideation assessment model includes a pre-trained language model and an unsupervised topic model, and a connection layer respectively connected to the output ends of the pre-trained language model and the unsupervised topic model; the pre-trained language model is used to obtain the category of the completion information; the unsupervised topic model is used to obtain the topic features of the completion information; the connection layer is used to connect the category and the topic features and predict the suicide ideation evaluation result of the completion information; Wherein, the training method of the suicide ideation assessment model includes: Obtain a sample of the sentence to be completed, the corresponding completion information sample and the corresponding suicide ideation label; wherein, the suicide ideation label is pre-annotated; Based on the sample of the sentence to be completed, the corresponding completion information sample and the corresponding suicide ideation label, perform iterative training on the initial language model; and, based on the sample of the sentence to be completed, the corresponding completion information sample and the corresponding suicide ideation label, perform iterative training on the initial unsupervised topic model; Connect the initial language model and the initial unsupervised topic model to the connection layer, output the suicide ideation evaluation result of the completion information, and train to obtain the pre-trained language model and the unsupervised topic model; During the iterative training process of the initial language model, apply a contrastive learning loss function to the completion information with the same encoding and the completion information with different encodings: Among them, A i represents the corresponding completion information sample of the sentence sample to be completed, and H i is the deep vector representation obtained by the pre-trained language model; for each representation H i , the representation with the same encoding as it is denoted as H i + , and the different ones are denoted as H i - ; by defining the distance function f d , the contrastive learning loss function makes the deep vector representations of the answers with the same encoding closer to each other.

2. The method according to claim 1, characterized in that, The step of inputting the completion information into the suicide ideation assessment model to obtain a suicide ideation assessment result corresponding to the completion information includes: Encode the sentence to be completed and the corresponding completion information through the pre-trained language model to obtain the category of the completion information; Extract the feature of the sentence to be completed and the corresponding completion information through the unsupervised topic model to obtain the topic feature of the completion information; Connect the completion information, the category and the topic feature through the connection layer, and predict the suicide ideation evaluation result of the completion information.

3. The method according to claim 2, wherein The step of encoding the sentence to be completed and the corresponding completion information through the pre-trained language model to obtain the category of the completion information includes: Based on the pre-trained language model, extract the deep representation features of each sentence to be completed and the corresponding completion information; Encode and predict the category of the completion information according to the deep representation features.

4. The method according to claim 1, characterized in that, During the iterative training process of the initial language model, adopt the following cross-entropy loss function: ， Among them, C i represents the manually encoded category information, and C i pred represents the category information predicted by the model.

5. The method according to claim 1, characterized in that, The step of performing iterative training on the initial topic model based on the sample of the sentence to be completed, the corresponding completion information sample and the corresponding suicide ideation label includes: Using the following formula, the supplementary information is represented as a low-dimensional vector to obtain the representation result of the supplementary information: where T represents the representation of the topic dimension obtained by the unsupervised topic model, that is, a vector of length t.

6. The method according to claim 5, wherein The step of inputting the supplementary information into the suicide ideation assessment model to obtain the suicide ideation assessment result corresponding to the supplementary information includes: Inputting the category and the topic feature into the connection layer to perform binary classification prediction of suicide ideation, and obtaining the suicide ideation evaluation result; wherein, the Focal loss function is used for prediction in the process of obtaining the suicide ideation evaluation result: where y represents the individual's true suicide ideation, measured by a professional scale, y^ represents the result predicted by the model from the encoded feature C and the topic feature T, and γ represents the temperature coefficient of the Focal loss function, which can be set according to the data distribution.

7. The method according to claim 1, characterized in that, The step of obtaining the supplementary information of the to-be-completed sentence based on the to-be-completed sentence includes: displaying the to-be-completed sentence, and obtaining the input information of the user for each to-be-completed sentence as the supplementary information; or, audio-playing the to-be-completed sentence, and obtaining the input information of the user for each to-be-completed sentence as the supplementary information.

8. An intelligent evaluation system for suicide ideation based on sentence completion tests, characterized in that, It includes a processor and a memory, and the processor is connected to the memory: wherein, the processor is configured to call and execute the program stored in the memory; the memory is configured to store the program, and the program is at least used to execute the suicide ideation intelligent assessment method based on sentence completion test according to any one of claims 1-7.

Images