An abnormal childbirth parturient psychological resilience intervention method and system

Abstract

The application provides an abnormal delivery parturient psychological resilience intervention method and system, comprising obtaining text corpus data of parturients and structured data containing clinical physiological indicators, family support, economic conditions and previous intervention history; using query and key matrix of psychological stress word library weighted attention mechanism to extract preliminary emotional polarity vector; performing tensor fusion on the preliminary emotional polarity vector and clinical physiological indicators to generate context adjustment matrix; using the adjustment matrix to correct self-attention value matrix to obtain corrected emotional polarity vector; splicing the vector with structured data to generate a comprehensive vector representing psychological state; and matching the optimal psychological intervention scheme from the preset intervention scheme knowledge base according to the comprehensive vector.

Description

Technical Field

[0001] This application belongs to the field of psychological intervention, and in particular relates to a method and system for intervention on the psychological resilience of women with abnormal deliveries. Background Technology

[0002] Abnormal childbirth is a severe stressful event, requiring timely psychological intervention for mothers who experience it. A mother's psychological state is a complex complex. While assessment methods such as clinical interviews and psychometric scales are widely used, they are highly subjective, have limited sensitivity, cannot detect immediate emotional fluctuations, and fail to reflect the mother's true state. Analyzing mothers' text corpora using natural language processing (NLP) technology, at the multimodal data fusion level, only splices or weights text features with structured clinical physiological indicators, failing to achieve interaction and verification between the two. It cannot utilize objective physiological data to correct and adjust emotional features extracted from subjective text, potentially leading to assessment results that deviate from the true level of physiological stress. In terms of text analysis depth, general models struggle to detect subtle emotional changes in mothers under specific stressful situations and are insufficient in identifying the weights of key psychological stress event vocabulary. Furthermore, existing methods, based on classification rule matching or human experience judgment, lack a way to match the mother's multidimensional profile with intervention programs in a knowledge base, making it difficult to achieve optimal intervention recommendations. Summary of the Invention

[0003] This invention proposes a method for intervention on the psychological resilience of women experiencing abnormal childbirth. It addresses the lack of existing technologies that utilize objective physiological data to correct and regulate emotional features extracted from subjective texts, and that provide a means to match multidimensional profiles of mothers with intervention programs in a knowledge base. The method includes: The text corpus data and structured data of the mothers to be analyzed are obtained. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. The text corpus data is processed in a first round. In the first round, the query matrix and key matrix of the self-attention mechanism are weighted using a preset psychological stress event lexicon to extract the preliminary emotional polarity vector representing the preliminary emotional state. The preliminary emotional polarity vector is fused with the clinical physiological indicators using tensors to generate a comprehensive state feature representation, and a context adjustment matrix is ​​constructed from the feature representation. A second processing round is performed on the text corpus data. In the second processing round, the Hadamard product operation is performed using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The modified emotional polarity vector is concatenated with the family support, economic conditions, and previous intervention history in the structured data to generate a comprehensive vector; the optimal psychological intervention plan is determined from a pre-set psychological intervention plan knowledge base based on the comprehensive vector.

[0004] Furthermore, this invention also relates to a psychological resilience intervention system for women experiencing abnormal childbirth, comprising the following modules: The extraction module is used to acquire text corpus data and structured data of the mothers to be analyzed. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. The text corpus data is processed in a first round. In the first round, the query matrix and key matrix of the self-attention mechanism are weighted using a preset psychological stress event lexicon to extract the preliminary emotional polarity vector representing the preliminary emotional state. A construction module is used to perform tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and to construct a context conditioning matrix from the feature representation; The generation module is used to perform a second processing round on the text corpus data. In the second processing round, the Hadamard product operation is performed using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The determination module is used to concatenate the modified emotional polarity vector with the family support, economic conditions and previous intervention history in the structured data to generate a comprehensive vector; and to determine the optimal psychological intervention plan in a pre-set psychological intervention plan knowledge base based on the comprehensive vector.

[0005] This invention constructs a two-round processing framework that integrates preliminary emotional states extracted from text with clinical physiological indicators. The integrated features are then used to correct the interpretation of the text's emotional state, achieving mutual verification between subjective emotional expression and objective physiological data, thus improving the reliability of assessing the postpartum woman's psychological state. Based on this, the corrected emotional vector is integrated with information on family support and economic conditions to form a comprehensive characterization of the postpartum woman's situation. Through a scheme matching strategy combining hard constraint filtering and vector similarity calculation, an intervention plan that is both realistic and psychologically appropriate can be determined for the postpartum woman from a vast knowledge base, enhancing the effectiveness of the intervention. Attached Figure Description

[0006] Figure 1 A flowchart of the first embodiment; Figure 2 A schematic diagram illustrating the generation of the modified attention value matrix; Figure 3 A schematic diagram for generating the optimal psychological intervention plan. Detailed Implementation

[0007] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0008] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent".

[0009] See Figure 1 One method for intervention in the psychological resilience of mothers experiencing abnormal childbirth, specifically: S1, Obtain the text corpus data and structured data of the mother to be analyzed. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. Perform a first processing round on the text corpus data. In the first processing round, use a preset psychological stress event lexicon to weight the query matrix and key matrix of the self-attention mechanism and extract the preliminary emotional polarity vector representing the preliminary emotional state. Textual data was collected from structured clinical interview records of postpartum women and voluntary textual feedback from postpartum women in the hospital's psychological assessment system, and preprocessed by word segmentation and stop word removal. Structured data was obtained from the hospital information system, including postpartum women's physiological monitoring records such as heart rate, blood pressure, and cortisol levels as clinical physiological indicators. Standardized questionnaires, such as the Postpartum Social Support Scale (PSSS), were used to assess family support levels, and monthly family income was recorded as an economic condition. The presence or absence of a history of psychological intervention was also recorded as yes or no.

[0010] The preprocessed text is input into a Transformer-based encoder. A pre-defined lexicon of psychological stress events includes words such as hemorrhage, episiotomy, forceps delivery, helplessness, and fear. During the encoder's self-attention calculation, for each word in the text sequence, if the word belongs to the lexicon, the row vectors of the corresponding query matrix Q and key matrix K are multiplied by a weight coefficient greater than 1, for example, 1.5. This ensures that the model gives high attention to the stress words when calculating the attention score. After weighted attention calculation and a feedforward network layer, the hidden states corresponding to the classification labels in the output layer are taken and passed through a fully connected layer to obtain a three-dimensional vector representing the probabilities of negative, neutral, and positive emotions; this is the initial sentiment polarity vector.

[0011] In an optional embodiment, the step of using a preset psychological stress event lexicon to weight the query matrix and key matrix of the self-attention mechanism to extract the preliminary emotional polarity vector representing the initial emotional state specifically involves: Each word in the text corpus to be analyzed is matched with the psychological stress event lexicon. A first preset weight value is set for words that exist in the lexicon, and a second preset weight value is set for words that do not exist in the lexicon. Thus, a weight vector is constructed for each word. The weight vector is then multiplied element-wise with the query matrix and key matrix of the self-attention mechanism to obtain the weighted query matrix and key matrix.

[0012] Construct a lexicon containing keywords related to postpartum psychological stress, such as insomnia, fatigue, crying, and anxiety. When processing a text input by a new mother, such as "I have been suffering from severe insomnia recently because my baby is crying," scan the text word by word. Identify insomnia and crying in the text as words existing in the psychological stress event lexicon and assign them a higher first preset weight value, such as 1.5. For other words in the text that do not appear in the lexicon, such as "I," "recently," "because," "baby," "and," and "severe," assign a lower second preset weight value, such as 1.0. The text generates a corresponding weight vector [1.0, 1.0, 1.0, 1.0, 1.5, 1.0, 1.0, 1.5].

[0013] At the model structure level, the input text is converted into a vector sequence through a word embedding layer and fed into a standard self-attention module. This module contains three linear transformation layers, used to generate the query matrix Q, key matrix K, and value matrix V, respectively. The generated weight vectors are broadcast to the same dimensions as the query matrix Q and key matrix K, and then multiplied element-wise with them to obtain the weighted query matrix and key matrix. This weighting operation makes the model pay more attention to words related to psychological stress when calculating attention scores between words, thereby enhancing the model's ability to detect key negative emotional information in the text.

[0014] The hidden state of the classification label is extracted from the encoder output layer. This hidden state is then input into a fully connected layer with 128 neurons and the activation function is ReLU. The output layer is then passed through, and the activation function is Softmax to obtain a three-dimensional vector, such as [0.82, 0.15, 0.03]. This vector is the initial sentiment polarity vector, which represents the probability of negative, neutral, and positive emotions, respectively.

[0015] In an optional embodiment, the clinical physiological indicators include: The clinical physiological indicators in the structured data include resting heart rate, systolic blood pressure, diastolic blood pressure, and serum cortisol concentration.

[0016] Specifically, the clinical physiological indicators in this embodiment constitute a multi-dimensional assessment of the physiological and psychological state of postpartum women. Resting heart rate, systolic blood pressure, and diastolic blood pressure are typically obtained through wearable devices or regular clinical measurements. For example, a postpartum woman's resting heart rate might be recorded as 88 beats per minute, her systolic blood pressure as 140 mmHg, and her diastolic blood pressure as 95 mmHg. Elevated values ​​in these values ​​may reflect a state of sustained stress in the autonomic nervous system. Serum cortisol concentration, as a stress hormone, is obtained through laboratory analysis of blood samples collected from postpartum women. For example, a measurement of 22 micrograms per deciliter, which is higher than the normal range, may indicate long-term physiological stress.

[0017] S2, the preliminary emotional polarity vector and the clinical physiological indicators are fused using tensors to generate a comprehensive state feature representation, and a context adjustment matrix is ​​constructed from the feature representation; Clinical physiological indicators such as heart rate, blood pressure, and cortisol levels are normalized and used as clinical physiological vectors. The outer product of the preliminary emotional polarity vector and the clinical physiological vector is calculated to generate a fusion matrix. Each element of this matrix reflects the interaction between a specific emotional tendency and a specific physiological indicator; this is the comprehensive state feature representation. This fusion matrix is ​​then input into a fully connected layer for feature extraction and dimensionality transformation, matching the matrix shape to the value matrix V in the attention mechanism, thereby generating the context adjustment matrix.

[0018] In an optional embodiment, the step of tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and constructing a context conditioning matrix from the feature representation, specifically involves: The values ​​of each clinical physiological indicator were normalized and then concatenated into a clinical physiological vector. For the preliminary emotional polarity vector With the clinical physiological vector Performing an outer product operation generates a comprehensive state tensor T as the feature representation of the comprehensive state. The calculation formula is as follows: The integrated state tensor T is then input into a fully connected neural network layer for transformation to generate the context adjustment matrix.

[0019] The collected clinical physiological indicators were standardized to eliminate dimensional differences. For example, a postpartum woman's resting heart rate of 88 beats per minute, systolic blood pressure of 140 mmHg, diastolic blood pressure of 95 mmHg, and serum cortisol concentration of 22 μg / dL were converted into values ​​between 0 and 1 using min-max normalization, such as [0.8, 0.85, 0.8, 0.9]. These normalized values ​​were then concatenated into a clinical physiological vector. Meanwhile, assuming the preliminary sentiment polarity vector obtained from text analysis... For a dimension The vector.

[0020] Perform the outer product operation to generate a dimension of A comprehensive state tensor T is generated. Each element of this tensor represents the interaction strength between a certain emotional dimension and a certain physiological indicator, thereby detecting the correlation between mental and physical states. The comprehensive state tensor T is flattened and input into a fully connected neural network layer. This network layer adaptively captures the correlation strength between physiological indicators and emotional features by learning a set of weights and bias parameters, mapping and reshaping the high-dimensional interaction features into a matrix with the same dimensions as the attention mechanism value matrix V. This matrix is ​​the context adjustment matrix.

[0021] S3, perform a second processing round on the text corpus data. In the second processing round, perform a Hadamard product operation using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The preprocessed text corpus is input again into a Transformer encoder with the same structure; however, those skilled in the art will recognize that while the structure is the same, the parameters are different. In the final step of the self-attention mechanism computation, before the weighted summation, the value matrix V generated in the first processing round is combined with the context adjustment matrix obtained in the previous step using an element-wise Hadamard product, resulting in a new value matrix V'. This operation adjusts the content representation of each lexical unit using physiological-emotional interaction information. For example, if physiological indicators show high tension, the representation of fear-related lexical units in V' will be enhanced. Attention computation is then performed using the modified value matrix V', similarly generating an emotion polarity vector modified under the supervision of objective physiological data through the hidden states of the classification labels and fully connected layers.

[0022] In an optional embodiment, the Hadamard product operation using the context adjustment matrix and the value matrix of the self-attention mechanism specifically involves: The context adjustment matrix is ​​multiplied element-wise with the value matrix to generate a modified attention value matrix that carries information on the regulation of clinical physiological indicators.

[0023] Specifically, in the self-attention mechanism, in addition to the query matrix and key matrix, there is also a value matrix V. The dimension of the value matrix is ​​related to the length of the input text sequence and the word vector dimension. For example, for a sentence containing 10 words with a word vector dimension of 64, V is a 10×64 matrix. Each row of the value matrix can be regarded as the semantic content representation of the word at that position. The context conditioning matrix generated above has the same dimension as the value matrix V, i.e., 10×64. Each element value of this conditioning matrix is ​​determined by the comprehensive clinical physiological indicators of the parturient, reflecting the appropriate conditioning strength of the physiological state on the semantic content of each word.

[0024] The context adjustment matrix and the value matrix V are subjected to a Hadamard product operation, that is, element-wise multiplication at corresponding positions, to generate a modified attention value matrix V', as shown below. Figure 2 As shown. For example, if the word "happy" appears in the text, but the mother's physiological indicators show high cortisol levels, then in the context conditioning matrix, the element values ​​of the row corresponding to the word "happy" may generally be small, such as 0.7. After element-wise multiplication, the semantic representation strength of "happy" in the modified value matrix V' will be weakened. The model uses attention weights to perform a weighted summation on the modified value matrix V', and the result is the modified sentiment polarity vector, which includes both textual information and the objective conditioning of physiological state.

[0025] S4. The modified emotional polarity vector is concatenated with the family support, economic conditions and previous intervention history in the structured data to generate a comprehensive vector; the optimal psychological intervention plan is determined from the pre-set psychological intervention plan knowledge base based on the comprehensive vector.

[0026] Family support scores, economic conditions, and previous intervention history are encoded using one-hot encoding. These one-hot encodings are then concatenated into a vector, which is then combined with the modified affective polarity vector obtained in the previous step to obtain a comprehensive vector representing the mother's psychological, physiological, and social background.

[0027] In an optional embodiment, determining the optimal psychological intervention plan from a pre-set psychological intervention plan knowledge base based on the comprehensive vector specifically involves: The candidate psychological intervention program is retrieved from the knowledge base and filtered based on the hard constraints of family support, economic conditions, and previous intervention history to obtain a set of candidate programs. For each candidate solution in the candidate solution set, the cosine similarity between the solution feature vector corresponding to the candidate solution and the comprehensive vector is calculated as the matching score, and the candidate solution with the maximum matching score is determined as the psychological intervention solution.

[0028] Specifically, the psychological intervention program knowledge base pre-stores various intervention programs, such as cognitive behavioral therapy, interpersonal therapy, and supportive psychotherapy. Each program is marked with applicable conditions; for example, cognitive behavioral therapy may be marked as requiring moderate to high economic conditions, and interpersonal therapy may be marked as requiring strong family support. Based on the hard conditions that the mother's actual economic conditions are moderate and her family support score is 7, the knowledge base is queried, and all programs that do not meet the conditions are filtered out. For example, programs marked as requiring high economic conditions are removed, and the remaining programs constitute the candidate program set.

[0029] Each solution in the knowledge base is pre-generated with a feature vector of the same dimension as the comprehensive vector through expert annotation or model training. This feature vector represents the ideal maternal profile to which the solution is applicable. For example, for cognitive behavioral therapy for post-traumatic stress, the feature vector has high values ​​in the dimensions of negative emotions and physiological stress. The cosine similarity between the maternal's comprehensive vector and the feature vector of each solution in the candidate solution set is calculated one by one. For example, the similarity of solution A is 0.92, solution B is 0.78, and solution C is 0.85. Solution A, with the highest similarity of 0.92, is selected as the optimal psychological intervention solution recommended to the maternal patient. Figure 3 .

[0030] In an optional embodiment, the filtering of candidate psychological intervention programs based on the hard constraints of family support, economic conditions, and previous intervention history includes: When the maternal family support rating is lower than the first preset threshold, intervention programs that require deep involvement of family members are excluded from the candidate programs; When the mother's economic condition rating is lower than the second preset threshold, intervention programs with costs exceeding the preset cost threshold are excluded from the candidate programs. When a mother's past intervention history shows no response to a specific therapy, all interventions based on that specific therapy are excluded from the candidate pool.

[0031] Specifically, assume an initial pool of candidate intervention options, including Option A: Couple-involved family cognitive therapy, Option B: High-intensity one-on-one counseling (high cost), Option C: Online cognitive behavioral therapy (CBT) courses, and Option D: Community peer support groups. Simultaneously, a first pre-set threshold for family support and a second pre-set threshold for economic conditions are set as moderate.

[0032] For a specific new mother, assessment data shows a low family support rating, a poor economic condition rating, and a history indicating she previously tried CBT with poor results. Filtering rules are applied sequentially. Option A, requiring deep family involvement, is excluded due to below-moderate family support. Option B, with its higher cost due to below-moderate economic conditions, is excluded. Option C is also excluded based on her history of ineffective CBT responses. After filtering through these hard constraints, only Option D—a community peer support group—remains for the next step of personalized ranking and recommendation. This process ensures that the recommended options are realistic and feasible, and meet the individual's specific constraints.

[0033] In an optional embodiment, the modified emotional polarity vector is concatenated with family support, economic conditions, and previous intervention history from the structured data to generate a comprehensive vector, specifically in the following manner: The family support level, economic conditions, and previous intervention history (with or without) are divided into multiple preset levels and converted into corresponding numerical vectors using one-hot encoding. The modified emotional polarity vector and the numerical vector are then connected end-to-end in the dimension to form the comprehensive vector.

[0034] Family support levels are categorized into high, medium, and low; economic conditions are categorized into good, medium, and poor; and previous intervention history is categorized into yes or no. One-hot coding transforms this non-numerical classification information into numerical vectors that can be processed by machine learning models. For a mother with medium family support, poor economic conditions, and no previous intervention history, the corresponding one-hot codes are [0,1,0], [0,0,1], and [0,1], respectively.

[0035] The one-hot encoded vectors are concatenated to form a single social context vector, which in the above example is [0,1,0,0,0,1,0,1]. The modified sentiment polarity vector generated in the previous step is then concatenated with the above-mentioned social context vector of length 8 in terms of dimension. The concatenated result is a comprehensive vector that integrates the mother's textual emotions, physiological state, family support, economic status, and medical history.

[0036] In a second embodiment, the present invention also proposes a psychological resilience intervention system for women with abnormal deliveries, comprising the following modules: The extraction module is used to acquire text corpus data and structured data of the mothers to be analyzed. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. The text corpus data is processed in a first round. In the first round, the query matrix and key matrix of the self-attention mechanism are weighted using a preset psychological stress event lexicon to extract the preliminary emotional polarity vector representing the preliminary emotional state. A construction module is used to perform tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and to construct a context conditioning matrix from the feature representation; The generation module is used to perform a second processing round on the text corpus data. In the second processing round, the Hadamard product operation is performed using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The determination module is used to concatenate the modified emotional polarity vector with the family support, economic conditions and previous intervention history in the structured data to generate a comprehensive vector; and to determine the optimal psychological intervention plan in a pre-set psychological intervention plan knowledge base based on the comprehensive vector.

[0037] In an optional embodiment, determining the optimal psychological intervention plan from a pre-set psychological intervention plan knowledge base based on the comprehensive vector specifically involves: The candidate psychological intervention program is retrieved from the knowledge base and filtered based on the hard constraints of family support, economic conditions, and previous intervention history to obtain a set of candidate programs. For each candidate solution in the candidate solution set, the cosine similarity between the solution feature vector corresponding to the candidate solution and the comprehensive vector is calculated as the matching score, and the candidate solution with the maximum matching score is determined as the psychological intervention solution.

[0038] In an optional embodiment, the weighting of the query matrix and key matrix of the self-attention mechanism using a preset psychological stress event lexicon specifically involves: Each word in the text corpus to be analyzed is matched with the psychological stress event lexicon. A first preset weight value is set for words that exist in the lexicon, and a second preset weight value is set for words that do not exist in the lexicon. Thus, a weight vector is constructed for each word. The weight vector is then multiplied element-wise with the query matrix and key matrix of the self-attention mechanism to obtain the weighted query matrix and key matrix.

[0039] In an optional embodiment, the clinical physiological indicators include: The clinical physiological indicators in the structured data include resting heart rate, systolic blood pressure, diastolic blood pressure, and serum cortisol concentration.

[0040] In an optional embodiment, the step of tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and constructing a context conditioning matrix from the feature representation, specifically involves: The values ​​of each clinical physiological indicator were normalized and then concatenated into a clinical physiological vector. For the preliminary emotional polarity vector With the clinical physiological vector Performing an outer product operation generates a comprehensive state tensor T as the feature representation of the comprehensive state. The calculation formula is as follows: The integrated state tensor T is then input into a fully connected neural network layer for transformation to generate the context adjustment matrix.

[0041] In an optional embodiment, the Hadamard product operation using the context adjustment matrix and the value matrix of the self-attention mechanism specifically involves: The context adjustment matrix is ​​multiplied element-wise with the value matrix to generate a modified attention value matrix that carries information on the regulation of clinical physiological indicators.

[0042] In an optional embodiment, the modified emotional polarity vector is concatenated with family support, economic conditions, and previous intervention history from the structured data to generate a comprehensive vector, specifically in the following manner: The family support level, economic conditions, and previous intervention history (with or without) are divided into multiple preset levels and converted into corresponding numerical vectors using one-hot encoding. The modified emotional polarity vector and the numerical vector are then connected end-to-end in the dimension to form the comprehensive vector.

[0043] In an optional embodiment, the filtering of candidate psychological intervention programs based on the hard constraints of family support, economic conditions, and previous intervention history includes: When the maternal family support rating is lower than the first preset threshold, intervention programs that require deep involvement of family members are excluded from the candidate programs; When the mother's economic condition rating is lower than the second preset threshold, intervention programs with costs exceeding the preset cost threshold are excluded from the candidate programs. When a mother's past intervention history shows no response to a specific therapy, all interventions based on that specific therapy are excluded from the candidate pool.

[0044] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0045] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for intervening in the psychological resilience of women experiencing abnormal childbirth, characterized in that, Includes the following steps: The text corpus data and structured data of the mothers to be analyzed are obtained. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. The text corpus data is processed in a first round. In the first round, the query matrix and key matrix of the self-attention mechanism are weighted using a preset psychological stress event lexicon to extract the preliminary emotional polarity vector representing the preliminary emotional state. The preliminary emotional polarity vector is fused with the clinical physiological indicators using tensors to generate a comprehensive state feature representation, and a context adjustment matrix is ​​constructed from the feature representation. A second processing round is performed on the text corpus data. In the second processing round, the Hadamard product operation is performed using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The modified emotional polarity vector is concatenated with the family support, economic conditions, and previous intervention history in the structured data to generate a comprehensive vector; the optimal psychological intervention plan is determined from a pre-set psychological intervention plan knowledge base based on the comprehensive vector.

2. The method according to claim 1, characterized in that, The step of determining the optimal psychological intervention plan from a pre-set psychological intervention plan knowledge base based on the comprehensive vector specifically involves: The candidate psychological intervention program is retrieved from the knowledge base and filtered based on the hard constraints of family support, economic conditions, and previous intervention history to obtain a set of candidate programs. For each candidate solution in the candidate solution set, the cosine similarity between the solution feature vector corresponding to the candidate solution and the comprehensive vector is calculated as the matching score, and the candidate solution with the maximum matching score is determined as the psychological intervention solution.

3. The method according to claim 1, characterized in that, The method of weighting the query matrix and key matrix of the self-attention mechanism using a pre-set psychological stress event lexicon is as follows: Each word in the text corpus to be analyzed is matched with the psychological stress event lexicon. A first preset weight value is set for words that exist in the lexicon, and a second preset weight value is set for words that do not exist in the lexicon. Thus, a weight vector is constructed for each word. The weight vector is then multiplied element-wise with the query matrix and key matrix of the self-attention mechanism to obtain the weighted query matrix and key matrix.

4. The method according to any one of claims 1-3, characterized in that, The clinical physiological indicators include: The clinical physiological indicators in the structured data include resting heart rate, systolic blood pressure, diastolic blood pressure, and serum cortisol concentration.

5. The method according to claim 1, characterized in that, The step of tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and constructing a context conditioning matrix from the feature representation, specifically involves: The values ​​of each clinical physiological indicator were normalized and then concatenated into a clinical physiological vector. For the preliminary emotional polarity vector With the clinical physiological vector An outer product operation is performed to generate a comprehensive state tensor T as the comprehensive state feature representation, and the comprehensive state tensor T is input into a fully connected neural network layer for transformation to generate the context adjustment matrix.

6. The method according to claim 1, characterized in that, The Hadamard product operation using the context adjustment matrix and the value matrix of the self-attention mechanism is specifically as follows: The context adjustment matrix is ​​multiplied element-wise with the value matrix to generate a modified attention value matrix that carries information on the regulation of clinical physiological indicators.

7. The method according to claim 1, characterized in that, The modified emotional polarity vector is concatenated with family support, economic conditions, and past intervention history from the structured data to generate a comprehensive vector. The specific method is as follows: The family support level, economic conditions, and previous intervention history (with or without) are divided into multiple preset levels and converted into corresponding numerical vectors using one-hot encoding. The modified emotional polarity vector and the numerical vector are then connected end-to-end in the dimension to form the comprehensive vector.

8. The method according to claim 2, characterized in that, The filtering of candidate psychological intervention programs based on the hard constraints of family support, economic conditions, and previous intervention history includes: When the maternal family support rating is lower than the first preset threshold, intervention programs that require deep involvement of family members are excluded from the candidate programs; When the mother's economic condition rating is lower than the second preset threshold, intervention programs with costs exceeding the preset cost threshold are excluded from the candidate programs. When a mother's past intervention history shows no response to a specific therapy, all interventions based on that specific therapy are excluded from the candidate pool.

9. A psychological resilience intervention system for women experiencing abnormal childbirth, characterized in that, Includes the following modules: The extraction module is used to acquire text corpus data and structured data of the mothers to be analyzed. The structured data includes clinical physiological indicators, family support, economic conditions and previous intervention history. The text corpus data is processed in a first round. In the first round, the query matrix and key matrix of the self-attention mechanism are weighted using a preset psychological stress event lexicon to extract the preliminary emotional polarity vector representing the preliminary emotional state. A construction module is used to perform tensor fusion of the preliminary emotional polarity vector and the clinical physiological indicators to generate a comprehensive state feature representation, and to construct a context conditioning matrix from the feature representation; The generation module is used to perform a second processing round on the text corpus data. In the second processing round, the Hadamard product operation is performed using the context adjustment matrix and the value matrix of the self-attention mechanism to generate a modified sentiment polarity vector corrected by clinical data. The determination module is used to concatenate the modified emotional polarity vector with the family support, economic conditions and previous intervention history in the structured data to generate a comprehensive vector; and to determine the optimal psychological intervention plan in a pre-set psychological intervention plan knowledge base based on the comprehensive vector.

10. The system according to claim 9, characterized in that, The step of determining the optimal psychological intervention plan from a pre-set psychological intervention plan knowledge base based on the comprehensive vector specifically involves: The candidate psychological intervention program is retrieved from the knowledge base and filtered based on the hard constraints of family support, economic conditions, and previous intervention history to obtain a set of candidate programs. For each candidate solution in the candidate solution set, the cosine similarity between the solution feature vector corresponding to the candidate solution and the comprehensive vector is calculated as the matching score, and the candidate solution with the maximum matching score is determined as the psychological intervention solution.

Images