Artificial intelligence-based postoperative rehabilitation nursing system for the elderly

Abstract

The application provides an artificial intelligence-based postoperative rehabilitation nursing system for the elderly, and relates to the field of intelligent nursing, which first encrypts and normalizes the collected emotional log data set of the target elderly object to obtain the time sequence distribution of the encrypted emotional log, then uses an artificial intelligence-based data analysis method to summarize the content and embed the semantics of the time sequence distribution of the encrypted emotional log to obtain the time sequence distribution of the emotional log summary semantic features, and obtains the emotion recognition result based on the context causal correlation coding features of the time sequence distribution of the emotional log summary semantic features, and finally judges whether to generate a psychological intervention prompt signal based on the emotion recognition result. In this way, by continuously monitoring and analyzing the emotional fluctuations and laws of the elderly, it is helpful to more accurately capture the emotional state, and thus improve the timeliness, effectiveness and intelligent degree of psychological intervention.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent nursing, and more specifically, to an artificial intelligence-based postoperative rehabilitation nursing system for the elderly. BACKGROUND

[0002] In the postoperative rehabilitation nursing process for the elderly, the importance of psychological intervention cannot be ignored. Surgery and its recovery period often bring physical discomfort and psychological pressure to the elderly, such as worries about the results of surgery, anxiety about future self-care abilities, and other negative emotions. These emotions not only affect the patient's sleep quality and appetite, but also can lead to a decline in immune system function, slow down wound healing, and increase the risk of postoperative complications. Therefore, implementing effective psychological intervention measures in the postoperative rehabilitation nursing of the elderly is of great significance for improving the psychological health of patients, promoting their positive attitude towards the rehabilitation process, and improving the quality of life.

[0003] However, traditional psychological intervention in postoperative rehabilitation nursing centers for the elderly faces many challenges. First, traditional psychological assessment mainly relies on the observation of medical staff and the self-report of patients. This approach may be biased due to individual differences, making it difficult to accurately capture the true emotional changes of each patient. Second, due to the limitation of human resources, it is difficult to continuously monitor and analyze the emotional state of each elderly patient, thus missing some early opportunities for psychological intervention. In addition, previous emotional analysis has been limited to the surface level, failing to deeply understand the underlying emotional fluctuations and their evolution patterns behind the patient's emotional journal. The existence of these problems has greatly reduced the effectiveness of traditional psychological intervention in postoperative rehabilitation nursing for the elderly.

[0004] Therefore, an optimized postoperative rehabilitation nursing scheme is expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide an artificial intelligence-based postoperative rehabilitation nursing system for the elderly.

[0006] According to one aspect of the present application, an artificial intelligence-based postoperative rehabilitation nursing system for the elderly is provided, which comprises:

[0007] a data collection preprocessing module configured to collect an emotional journal of a target elderly subject to obtain a data set of the emotional journal, and to perform encryption processing and data normalization on the data set of the emotional journal to obtain a time series distribution of the encrypted emotional journal;

[0008] An emotion state recognition module is configured to perform emotion analysis on the time sequence distribution of the encrypted emotion log based on text semantics to obtain an emotion state recognition result, wherein the emotion state recognition module comprises: a log time sequence summary embedding unit configured to perform content summary and semantic embedding on the time sequence distribution of the encrypted emotion log to obtain a time sequence distribution of emotion log summary semantic embedding coding features; a log time sequence coding unit configured to perform emotion log context causal correlation coding on the time sequence distribution of the emotion log summary semantic embedding coding features to obtain emotion state time sequence evolution semantic coding features; and a recognition result generation unit configured to obtain the emotion state recognition result based on the emotion state time sequence evolution semantic coding features.

[0009] An intervention signal generation module is configured to determine whether to generate a psychological intervention prompt signal based on the emotion state recognition result.

[0010] Compared with the prior art, the artificial intelligence-based postoperative rehabilitation nursing system for the elderly provided in the present application first performs encryption processing and data regularization on the collected emotion log data set of the target elderly object to obtain the time sequence distribution of the encrypted emotion log, then adopts an artificial intelligence-based data analysis method to perform content summary and semantic embedding on the time sequence distribution of the encrypted emotion log to obtain the time sequence distribution of emotion log summary semantic features, and obtains an emotion recognition result based on the context causal correlation coding features of the time sequence distribution of the emotion log summary semantic features, and finally determines whether to generate a psychological intervention prompt signal based on the emotion recognition result. In this way, by continuously monitoring and analyzing the emotional fluctuations and rules of the elderly, it is helpful to more accurately capture the emotional state, and thus to improve the timeliness, effectiveness and intelligent degree of psychological intervention. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0012] Figure 1 FIG. 1 is a system block diagram of the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to an embodiment of the present application.

[0013] Figure 2 FIG. 2 is a block diagram of the data acquisition and preprocessing module in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to an embodiment of the present application.

[0014] Figure 3A block diagram of an emotion state recognition module in an artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to an embodiment of the present application.

[0015] Figure 4 A block diagram of a log time summary embedding unit in an artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to an embodiment of the present application.

[0016] Figure 5 A block diagram of a log time coding unit in an artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0018] It should be noted that in the present application, all actions of obtaining signals, information or data are carried out in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the owner of the corresponding device.

[0019] In the postoperative rehabilitation of the elderly, psychological intervention is of great significance. During surgery and recovery, elderly patients often experience negative emotions such as worry about the results of surgery, anxiety about future self-care ability, and so on due to physical discomfort and psychological pressure. These emotions not only affect sleep and appetite, but also can weaken immune function, delay wound healing, and increase the risk of postoperative complications. Therefore, implementing effective psychological intervention is crucial for improving patient mental health, promoting positive rehabilitation, and enhancing quality of life.

[0020] However, traditional psychological intervention in postoperative rehabilitation of the elderly faces many challenges. On the one hand, its evaluation method mainly relies on the observation of medical staff and the self-report of patients, which is easily affected by individual differences and difficult to capture the real emotional changes of patients comprehensively and accurately. On the other hand, due to limited human resources, it is difficult to continuously monitor the emotional state of patients, often missing the best intervention opportunity. In addition, traditional sentiment analysis often stays on the surface and is difficult to deeply analyze the emotional fluctuations and evolution rules behind the emotional logs of patients. The existence of these problems weakens the effect of traditional psychological intervention in postoperative rehabilitation of the elderly.

[0021] To solve the above problems, the technical concept of the present application is to first collect the emotional log data set of the target elderly object, then perform encryption processing and data regularization on it to obtain the time sequence distribution of the encrypted emotional log, then use an artificial intelligence-based data analysis method to summarize the content and embed the semantics of the time sequence distribution of the encrypted emotional log to obtain the time sequence distribution of the emotional log summary semantic features, and based on the context causal correlation coding features of the time sequence distribution of the emotional log summary semantic features, the emotion recognition result is obtained, and finally based on the emotion recognition result, it is judged whether to generate a psychological intervention prompt signal. In this way, through continuous emotion monitoring and analysis and accurate mining of the deep emotional fluctuations and evolution rules behind the emotional log of the elderly in the analysis, the real emotional state of the elderly can be captured more timely, comprehensively and accurately, which is beneficial to improving the timeliness, effectiveness and intelligent degree of psychological intervention.

[0022] Figure 1 The system block diagram of the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to the embodiment of the present application is shown in FIG. 1. Figure 1 As shown in the figure, in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly 100, it includes: a data acquisition and preprocessing module 110, which is used to collect the emotional log of the target elderly object to obtain the data set of the emotional log, and to perform encryption processing and data regularization on the data set of the emotional log to obtain the time sequence distribution of the encrypted emotional log; an emotional state recognition module 120, which is used to perform emotional analysis based on text semantics on the time sequence distribution of the encrypted emotional log to obtain an emotional state recognition result; and an intervention signal generation module 130, which is used to determine whether to generate a psychological intervention prompt signal according to the emotional state recognition result.

[0023] In the embodiment of the present application, the data acquisition and preprocessing module 110 is used to collect the emotional log of the target elderly object to obtain the data set of the emotional log, and to perform encryption processing and data regularization on the data set of the emotional log to obtain the time sequence distribution of the encrypted emotional log. Specifically, Figure 2 The block diagram of the data acquisition and preprocessing module in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to the embodiment of the present application is shown in FIG. 2. Figure 2 As shown in the figure, the data acquisition and preprocessing module 110 includes: an emotional log acquisition unit 111, which is used to collect the emotional log of the target elderly object to obtain the data set of the emotional log; an emotional log encryption processing unit 112, which is used to perform encryption processing on each emotional log in the data set of the emotional log to obtain the data set of the encrypted emotional log; and an emotional log time sequence distribution adjustment unit 113, which is used to adjust the data distribution pattern based on the time dimension according to the time stamp to obtain the time sequence distribution of the encrypted emotional log.

[0024] In the embodiment of the present application, the emotion log collection unit 111 is configured to collect the emotion log of the target elderly person to obtain a data set of the emotion log. It should be understood that the collected data set of the emotion log mainly includes the record information of the emotional experience of the target elderly person at different time points, and can also include the event description information triggering the emotion, in addition to the association record information between the physical condition and the emotion, etc. Specifically, the emotional experience record in the emotion log is the most direct basis for judging the emotional state. If the proportion of negative emotions in the record is high and the intensity is large, it indicates that the elderly person can be in a negative emotional state. For example, if there are high-intensity anxiety or sadness emotion records for several consecutive days, attention should be paid. The event description triggering the emotion can help the model understand the background of the emotion generation. When the event triggering negative emotions frequently occurs or the intensity increases, such as persistent pain of a wound or frequent unpleasant communication with others, it can indicate that the emotion will continue to develop negatively. The physical condition and emotion association record can dynamically help the model evaluate the emotional state. For example, if the physical condition is continuously poor, leading to a persistent low mood, or the improvement of the physical condition does not bring about the improvement of the emotion, it can indicate that there is a potential emotional problem. In summary, through in-depth analysis of the data set of the emotion log, the emotional fluctuations of the elderly person that need to be paid attention to can be accurately identified, and the psychological intervention prompt signal can be automatically generated when necessary, so that medical staff can intervene in time to provide necessary support and help for the elderly patients.

[0025] In the embodiment of the present application, the emotion log encryption processing unit 112 is configured to encrypt each emotion log in the data set of the emotion log to obtain a data set of the encrypted emotion log. Accordingly, considering that the emotion log contains a large amount of personal sensitive information, such as the emotional state and inner feelings of the target elderly person, if these information is leaked, it can infringe the personal privacy of the elderly person. Therefore, the present application encrypts each emotion log in the data set of the emotion log to effectively protect the privacy and safety of the elderly person. It is worth mentioning that since the encryption process guarantees the integrity of the data, the subsequent detailed data processing is based on the original data that has not been tampered with, which can make the final emotional state recognition result more reliable, and thus ensure the accuracy of the generated psychological intervention prompt signal.

[0026] In the embodiment of the present application, the emotion log time sequence distribution adjustment unit 113 is configured to adjust the data distribution mode of the encrypted emotion log dataset based on the time dimension according to the timestamps to obtain the time sequence distribution of the encrypted emotion log. It should be understood that the emotional state of the elderly is not fixed, but fluctuates over time. By adjusting the data distribution mode of the encrypted emotion log based on the time dimension according to the timestamps, the emotion log can be arranged in chronological order, so that the emotional changes of the elderly at different stages of postoperative rehabilitation, at different times of the day, and at other time scales can be systematically observed. For example, some elderly people may have greater emotional fluctuations in the early postoperative period, and as their bodies gradually recover, their emotions tend to stabilize; or at different times of the day, anxiety is more likely to occur at night. That is, arranging the data in chronological order can help the model clearly capture the emotional change pattern, and thus better judge the development direction of the emotional state of the elderly subject.

[0027] In the embodiment of the present application, the emotion state recognition module 120 is configured to perform emotion analysis on the time sequence distribution of the encrypted emotion log based on the text semantics to obtain an emotion state recognition result. Specifically, Figure 3 A block diagram of the emotion state recognition module in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to the embodiment of the present application is shown in FIG. 12. As shown in FIG. 12, the emotion state recognition module 120 includes a log time sequence summary embedding unit 121, a log time sequence encoding unit 122, and a recognition result generation unit 123. Figure 3 The log time sequence summary embedding unit 121 is configured to perform content summary and semantic embedding on the time sequence distribution of the encrypted emotion log to obtain a time sequence distribution of emotion log summary semantic embedding coding features. The log time sequence encoding unit 122 is configured to perform emotion log context causal correlation coding on the time sequence distribution of the emotion log summary semantic embedding coding features to obtain emotion state time sequence evolution semantic coding features. The recognition result generation unit 123 is configured to obtain the emotion state recognition result based on the emotion state time sequence evolution semantic coding features.

[0028] In the embodiment of the present application, the log time sequence summary embedding unit 121 is configured to perform content summary and semantic embedding on the time sequence distribution of the encrypted emotion log to obtain a time sequence distribution of emotion log summary semantic embedding coding features. Specifically, Figure 4 A block diagram of the log time sequence summary embedding unit in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to the embodiment of the present application is shown in FIG. 13. As shown in FIG. 13, the log time sequence summary embedding unit 121 includes a content summary unit 131 and a semantic embedding unit 132. Figure 4As shown, the log time sequence summary embedding unit 121 comprises: an emotional log summary generation subunit 1211 configured to input each encrypted emotional log in the time sequence distribution of the encrypted emotional logs into a content summary encoder based on a large language model to obtain a time sequence distribution of emotional log summaries; and an emotional log summary embedding coding subunit 1212 configured to perform semantic embedding coding on each emotional log summary in the time sequence distribution of the emotional log summaries to obtain a time sequence distribution of emotional log summary semantic embedding coding vectors as the time sequence distribution of emotional log summary semantic embedding coding features.

[0029] In the embodiment of the present application, the emotional log summary generation subunit 1211 is configured to input each encrypted emotional log in the time sequence distribution of the encrypted emotional logs into a content summary encoder based on a large language model to obtain a time sequence distribution of emotional log summaries. It should be understood that the time sequence distribution of encrypted emotional logs contains a large amount of detailed information, which may complicate and be redundant for subsequent analysis. Considering that a large language model is an artificial intelligence model based on deep learning, which is trained by a large amount of text data and can understand and generate natural language. It can not only generate coherent and meaningful text, but also perform context-aware tasks such as extracting key information or summarizing main content from long documents. Based on this, the present application inputs each encrypted emotional log in the time sequence distribution of the encrypted emotional logs into a content summary encoder based on a large language model to simplify these detailed emotional logs, filter out events and emotional expressions closely related to emotional changes, and remove irrelevant information, making the data easier to process and analyze. The time sequence distribution of emotional log summaries obtained by processing is a concise but key emotional information containing data representation form, which is more refined than the original encrypted emotional logs, and is convenient for the model to quickly understand and analyze the emotional state of the elderly at different time points.

[0030] The following is a detailed description of one specific implementation process of "inputting each encrypted emotional log in the time sequence distribution of the encrypted emotional logs into a content summary encoder based on a large language model to obtain a time sequence distribution of emotional log summaries":

[0031] First, preliminary work needs to be done. This includes selecting a suitable large language model, of which there are many available on the market, such as OpenAI's GPT series, Google's BERT, etc. When choosing, multiple factors need to be considered, such as the model's language understanding and generation capabilities, performance on sentiment analysis tasks. At the same time, the encrypted sentiment log data needs to be obtained from the time series distribution of the encrypted sentiment log, and the corresponding encryption key needs to be used for decryption to restore it to its original text format. After that, the decrypted text needs to be cleaned to remove irrelevant special characters, garbled characters, etc., and the text format needs to be unified to ensure the standardization and consistency of the text, laying a good foundation for subsequent processing.

[0032] Next is the model parameter setting stage. If calling an API, the parameters need to be set according to the API documentation of the model. Taking GPT-3.5 as an example, the temperature parameter is used to control the randomness of the generated text, with a value range of 0-1, and the closer the value is to 0, the more deterministic the generated result will be; the maximum generation token number (max_tokens) is used to limit the length of the generated text. If the model is deployed locally, similar parameters need to be set according to the model's configuration file to optimize the generation effect.

[0033] After completing the preparation and parameter setting, the stage of interacting with the large language model begins. Use programming languages such as Python and corresponding libraries such as OpenAI's openai library and HuggingFace's transformers library to write code to implement interaction with the large language model. In the code, build a request body and pass the cleaned sentiment log text as input to the model.

[0034] When the emotional log text enters the large language model, a series of complex operations are performed inside the model to realize content summary coding. First, text coding, the large language model will split the input emotional log text into a token, and then map each token to a vector through an embedding layer. These vectors contain semantic information about the token, and the sequence formed becomes the basis for subsequent processing by the model. Next, the model uses its own attention mechanism to process the token vector sequence. The attention mechanism allows the model to focus on parts of the text related to emotions and key events, such as words that describe emotions (like "anxiety" "happy"), events that trigger emotions (such as "surgical wound pain intensifies"), and related information such as time, location, etc. Through a multi-layer Transformer architecture, the model performs deep semantic understanding of the text, mining implicit information and semantic associations. Then, the model generates an emotional log summary based on the extracted key information and semantic understanding of the text. Based on its learned language generation patterns, it predicts the next most likely token to appear, gradually building the summary text. During generation, it will fully consider the context information to ensure that the generated summary is coherent and logical. In addition, the generated summary may not be accurate or concise enough, and the model will use its own optimization mechanism to adjust it, such as using reinforcement learning methods to optimize the generated summary based on certain reward strategies (such as the matching degree of the generated summary with the key information of the original text, the fluency of the language, etc.), to improve the quality of the summary.

[0035] Finally, the emotional log summary result returned by the large language model is received and arranged. The generated summary is arranged in chronological order according to the encrypted emotional log, forming the time sequence distribution of the emotional log summary. To ensure the accuracy of the summary, the generated summary needs to be verified to check whether the summary accurately reflects the key emotions and events of the original emotional log, and whether there are logical errors or information missing. Automatic evaluation indicators such as ROUGE can be used to calculate the similarity between the generated summary and the reference summary (if any) to evaluate the quality of the summary. Through a series of implementation operations, the time sequence distribution of the emotional log summary can be obtained from the time sequence distribution of the encrypted emotional log, which can provide strong support for subsequent emotional analysis and intervention.

[0036] In the embodiment of the present application, the affective log summary embedding and encoding subunit 1212 is configured to perform semantic embedding and encoding on each affective log summary in the time sequence distribution of the affective log summary to obtain a time sequence distribution of affective log summary semantic embedding and encoding vectors as the time sequence distribution of affective log summary semantic embedding and encoding features. Specifically, in the embodiment of the present application, the affective log summary embedding and encoding subunit is configured to: perform semantic embedding and encoding on each affective log summary in the time sequence distribution of the affective log summary by using a semantic encoder based on a bidirectional recurrent neural network model to obtain the time sequence distribution of affective log summary semantic embedding and encoding vectors. Accordingly, considering that the affective log summary is text information, direct text-based analysis is difficult to capture deep emotional and semantic associations. In order to enable the computer to more effectively process and understand the complex semantic information contained therein, thereby more accurately grasping the emotional state of the elderly, in the present application, semantic embedding and encoding is performed on each affective log summary in the time sequence distribution of the affective log summary to obtain a time sequence distribution of affective log summary semantic embedding and encoding vectors as the time sequence distribution of affective log summary semantic embedding and encoding features. Semantic embedding and encoding operation can mine semantic information and potential features in the affective log summary. For example, it can identify the semantic similarity and association between different words and phrases. For emotion-related words such as "anxiety", "worry", "fear", etc., their close relationship can be found in the vector space through encoding, thereby more accurately understanding the semantic connotation behind the emotional expression of the elderly. In particular, in one specific embodiment of the present application, semantic embedding and encoding is performed on each affective log summary in the time sequence distribution of the affective log summary by using a semantic encoder based on a bidirectional recurrent neural network model to obtain the time sequence distribution of affective log summary semantic embedding and encoding vectors. Those skilled in the art should know that a bidirectional recurrent neural network is composed of two unidirectional recurrent neural networks, one of which is forward and processes data from front to back in the time sequence of the input sequence; the other is backward and processes data in the order from back to front. In the time sequence distribution of the affective log summary, it may contain records of emotional changes over a long period of time. Using a bidirectional recurrent neural network model can better mine the emotional evolution law therein through bidirectional information processing, avoiding information loss or forgetting. For example, the trigger factor of emotion (which may be in the front part of the affective log summary) and the subsequent impact caused by the emotion (which may be in the back part of the affective log summary) can be effectively encoded into the affective log summary semantic embedding and encoding vector through the bidirectional recurrent neural network model, thereby improving the accuracy of affective log summary semantic understanding.

[0037] The following is a detailed description of one specific implementation process of "using a semantic encoder based on a bidirectional recurrent neural network model to semantically embed and encode each of the emotion log summaries in the temporal distribution of the emotion log summary to obtain a temporal distribution of emotion log summary semantic embedding encoding vectors":

[0038] First, data preprocessing. This step needs to clean each emotion log summary comprehensively, remove special characters, punctuation, and unify the case form of the text, reducing the interference of word diversity. After cleaning, perform word segmentation on the text. For Chinese text, you can use tools such as Jieba, and for English text, you can use libraries such as NLTK and SpaCy to split the text into individual words or tokens. Based on the segmentation results, count all the tokens to build a vocabulary, assign each token a unique integer index, and set a special index to represent words not in the vocabulary. Then, convert the tokens in each emotion log summary to integer index sequences based on the vocabulary. If the sequence lengths are not consistent, pad the sequence with a specific index value 0 before and after the sequence or truncate it to make it equal in length for subsequent processing.

[0039] After data preprocessing, start building the bidirectional recurrent neural network (BiRNN) semantic encoder model. First, design the model architecture, which generally includes an input layer, a bidirectional recurrent layer, and an output layer. The input layer receives the preprocessed text sequence and converts it to a word vector representation, which can use pre-trained word embeddings (such as Word2Vec, GloVe) or randomly initialized word embedding matrices. The bidirectional recurrent layer processes the input sequence in both forward and reverse directions to capture context information in both directions, and concatenates the forward and reverse outputs to obtain richer semantic representations. The output layer maps the output of the bidirectional recurrent layer to the desired dimension to obtain the semantic embedding encoding vector of the emotion log summary. Finally, initialize the weights and biases of the model, commonly using methods such as Xavier initialization and He initialization to help the model converge faster.

[0040] After the model is built, it enters the training phase. First, divide the data set, and divide the preprocessed data into training set, validation set and test set according to the common proportion (70%-80% for training set, 10%-15% for validation set, and 10%-15% for test set). The training set is used for model training, the validation set is used for adjusting hyperparameters (such as learning rate, batch size), and the test set is used for evaluating the final performance. Then define the loss function and the optimizer, select the appropriate loss function (such as mean square error loss, cross entropy loss) and optimizer (such as stochastic gradient descent, Adam) according to the task, and the optimizer is used to minimize the loss function value in training. Then perform multiple rounds of iterative training on the training set, each iteration will input the data batch into the model, calculate the output result and calculate the loss value according to the loss function, and use the optimizer to update the model parameters according to the loss value. During the training process, the model performance is evaluated on the validation set regularly, the loss value and other evaluation indicators (such as accuracy, recall rate) are monitored to prevent overfitting, and if the performance on the validation set no longer improves, the training is stopped in advance. After training is completed, the performance of the trained model is evaluated using the test set, and relevant indicators such as accuracy, recall rate, and F1 value are calculated to determine whether the model performance meets the requirements. If the model performance does not meet the expectations, try adjusting the hyperparameters (such as learning rate, hidden layer dimension), increasing the amount of training data, and continuously optimizing the model performance and stability to ensure that the semantic encoder based on the bidirectional recurrent neural network model can efficiently and accurately encode the semantic embedding of the emotional log summary.

[0041] Finally, load the trained model with saved parameters, and input each emotional log summary in the time series distribution of the emotional log summary after preprocessing according to the preprocessing process. The model processes the input text sequence, extracts context information through the bidirectional recurrent layer, and finally obtains the corresponding semantic embedding coding vector in the output layer. Arrange these vectors in chronological order to form a time series distribution of emotional log summary semantic embedding coding vectors.

[0042] In the embodiments of the present application, the log time coding unit 122 is configured to perform emotional log context causal correlation coding on the time series distribution of the emotional log summary semantic embedding coding features to obtain emotional state time evolution semantic coding features. Specifically, Figure 5 A block diagram of the log time coding unit in the artificial intelligence-based postoperative rehabilitation nursing system for the elderly according to the embodiments of the present application. As shown in Figure 5As shown, the log timing encoding unit 122 comprises: an emotional log summary implicit semantic feature mining subunit 1221, configured to perform emotional log summary implicit semantic feature mining on the timing distribution of the emotional log summary semantic embedding encoding vector to obtain a timing distribution of an emotional log summary semantic deep implicit embedding encoding vector; an emotional log summary semantic causal analysis subunit 1222, configured to perform emotional log summary semantic causal correlation factor calculation and emotional log summary semantic causal trigger on the timing distribution of the emotional log summary semantic deep implicit embedding encoding vector to obtain an emotional log summary semantic causal correlation topological feature matrix; and an emotional state timing evolution semantic encoding feature generation subunit 1223, configured to perform log semantic context dynamic walk encoding fusion on the timing distribution of the emotional log summary semantic deep implicit embedding encoding vector and the timing distribution of the emotional log summary semantic embedding encoding vector based on the emotional log summary semantic causal correlation topological feature matrix to obtain an emotional state timing evolution semantic encoding vector as the emotional state timing evolution semantic encoding feature.

[0043] It should be understood that the bidirectional recurrent neural network model has certain limitations in processing time series data. Specifically, taking the association between wound healing and emotions as an example, there may be records of “wound infection aggravation, pain intensification, and resulting in extremely low emotions”. Although the traditional processing method can perceive the sequence of wound condition changes and emotional changes over time, it is difficult to explicitly model the causal relationship between wound infection aggravation and emotional decline. Moreover, the context of the timing distribution of the emotional log summary semantic embedding encoding feature presents multi-level associations, including not only explicit semantic information (such as directly expressed emotions), but also complex background information and potential psychological states. The traditional single-layer analysis method cannot effectively distinguish these different levels of information, resulting in incomplete and one-sided integration of emotional log summary context semantics. Based on this, the timing distribution of the emotional log summary semantic embedding encoding feature is encoded for emotional log context causal correlation in the present application to obtain emotional state timing evolution semantic encoding features. In this way, through the emotional log context causal correlation encoding operation, not only can the causal relationship be accurately sorted out, but also the multi-level semantics of the original features can be deeply mined, which is conducive to more accurately capturing the potential trend of emotional changes and thus improving the understanding of the emotional state of the elderly.

[0044] In detail, first, the timing distribution of the emotional log summary semantic embedding encoding vector is subjected to emotional log summary implicit semantic feature mining to obtain a timing distribution of an emotional log summary semantic deep implicit embedding encoding vector. The above process can be represented by the formula:

[0045] O = {x1, x2,..., x i ,...,xn}

[0046] v i =Sigmoid[Conv 1×1 (x i )]

[0047] D = {v1, v2, ..., v} i ,...,v n}

[0048] Where O represents the temporal distribution of the semantic embedding encoding vector of the sentiment log summary, x1, x2, x... i and x n These are the 1st, 2nd, ith, and nth sentiment log summary semantic embedding encoding vectors in the temporal distribution of the sentiment log summary semantic embedding encoding vectors, respectively. 1×1 It is a pointwise convolutional coding, and Sigmoid is the activation function for convolutional coding. v1, v2, v i and v n These are the 1st, 2nd, 1st, and 2nd implicit embedding encoding vectors of the semantic deep embedding encoding vector of the sentiment log summary, respectively, and D is the temporal distribution of the semantic deep implicit embedding encoding vector of the sentiment log summary.

[0049] It is understandable that the original emotional log summary semantic embedding encoding vector may contain a large amount of redundant information and noise, which can interfere with subsequent causal association analysis. By mining the implicit semantic features of the emotional log summary, deep and meaningful semantic features can be extracted from these vectors, reducing unnecessary information interference. In other words, the generated emotional log summary semantic deep implicit embedding encoding vector can more accurately reflect the core features of the emotional state of elderly individuals, providing richer clues for the model to comprehensively understand the emotional state of the elderly.

[0050] Specifically, in this embodiment, the sentiment log summary semantic causal analysis subunit includes: a first-level subunit for calculating sentiment log summary semantic causal association factors, used to calculate the sentiment log summary semantic causal association factors between any two sentiment log summary semantic deep implicit embedding encoding vectors in the temporal distribution of the sentiment log summary semantic deep implicit embedding encoding vectors to obtain a sentiment log summary semantic causal association topology matrix composed of multiple sentiment log summary semantic causal association factors; and a first-level subunit for triggering sentiment log summary semantic causal events based on a gating function on the sentiment log summary semantic causal association topology matrix to obtain the sentiment log summary semantic causal association topology feature matrix.

[0051] More specifically, in the embodiments of the present application, the emotion log summary semantic causal correlation factor calculation first-level subunit is used for: calculating the correlation matrix between any two emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vector to obtain the time sequence distribution of the emotion log summary semantic deep correlation embedding coding matrix; calculating the emotion log summary semantic causal correlation factor of each emotion log summary semantic deep correlation embedding coding matrix in the time sequence distribution of the emotion log summary semantic deep correlation embedding coding matrix to obtain the emotion log summary semantic causal correlation topology matrix composed of a plurality of emotion log summary semantic causal correlation factors, wherein the emotion log summary semantic causal correlation factor is calculated by the maximum value, the mean value, the variance and the causal correlation bias term of the emotion log summary semantic deep correlation embedding coding matrix; in response to the variance of the emotion log summary semantic deep correlation embedding coding matrix being greater than or equal to a preset threshold, the weighted average value of the distance between any two emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vector is taken as the causal correlation bias term, and in response to the variance of the emotion log summary semantic deep correlation embedding coding matrix being less than the preset threshold, the weighted average value of the emotion log summary semantic deep correlation embedding coding matrix is taken as the causal correlation bias term. The above process can be expressed by the following formula:

[0052]

[0053]

[0054] wherein v i and v j are the i th and j th emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vector, is matrix multiplication, v j T is the transposed vector of v j , M i-j is the emotion log summary semantic deep correlation embedding coding matrix between v i and v j , σ 2 (M i-j ) is the variance of M i-j , max(M i-j ) is the maximum value in M i-j , μ(M i-j ) is the mean value of M i-j , λ is the causal correlation bias term, t i-j is the emotion log summary semantic causal correlation factor corresponding to M i-j , and d(vi v j ) for v i and v j is the distance between v 1-1 , t 1-n , t n-1 and t n-n are the emotion log summary semantic causal correlation factors of each position in the emotion log summary semantic causal correlation topology matrix, respectively, and T is the emotion log summary semantic causal correlation topology matrix.

[0055] Correspondingly, in order to understand the causal mechanism behind the emotional changes at different time points, it is necessary to quantify the causal correlation between the emotion log summary semantic deep implicit embedding encoding vectors between any two time points, that is, to calculate the emotion log summary semantic causal correlation factor between any two emotion log summary semantic deep implicit embedding encoding vectors. The emotion log summary semantic causal correlation factor calculated by the causal correlation energy measurement function can accurately deduce the causal correlation between the features in the original time series distribution, helping the model to deeply explore the internal relationship between different features. The emotion log summary semantic causal correlation topology matrix constructed can present these causal relationships in a quantitative way, which is convenient for subsequent in-depth analysis of the internal mechanism of the emotional changes of the elderly.

[0056] In particular, by regarding the causal correlation of low-level emotional changes in a complex system as a molecular level relationship based on statistical correlation inference, the intervention prediction of emotional change causal correlation energy can be further performed on the basis of global fine-grained statistical association representation, so as to study the fine-grained structure of emotional change causal correlation and its dynamic regulation based on the high-dimensional and heterogeneous representation of emotional change causal relationship genomics. Among them, in the case that the aggregated distribution representation of the causal graph is greater than the preset threshold, the source data integration is biased based on the matrix graph node effect representation of the emotion log summary semantic causal correlation factor, and in the case that the aggregated distribution representation of the causal graph is less than the preset threshold, the condensed structure modeling can be directly performed through feature mode integration compression. In this way, not only the emotional change causal correlation energy in the system can be encoded and described, but also the implicit emotional change causal intervention prediction result can be condensed, so that the key emotional change causal correlation can be more efficiently revealed.

[0057] Then, the emotion log summary semantic causal correlation topology matrix is subjected to emotion log summary semantic causal triggering based on a gating function to obtain an emotion log summary semantic causal correlation topology feature matrix. The above process can be represented by the formula:

[0058]

[0059] wherein, ti-j is M i-j The corresponding emotional log summary semantic causal association factor T is the emotional log summary semantic causal association topology matrix, softmax is a nonlinear activation function, τ is a normalization threshold, f trigger (T) is a gating activation process for T, and M is an emotional log summary semantic causal association topology feature matrix.

[0060] It can be understood that the dynamic gating mechanism and the nonlinear activation function in the emotional log summary semantic causal trigger operation can perform more detailed feature extraction and modeling on the emotional change causal topology relationship. Specifically, the dynamic gating mechanism can accurately identify the key causal path in the complex and variable emotional log summary semantic dynamic context, thereby highlighting important emotional change associations, weakening noise interference, and making the emotional change causal relationship more clear and explicit; the nonlinear activation function greatly enhances the model's ability to capture high-order regularities in complex emotional change causal structures by introducing nonlinear characteristics, thereby improving the model's understanding and analysis ability of the emotional change causal relationship. That is, the generated emotional log summary semantic causal association topology feature matrix successfully filters out a large amount of interference information and highlights the key causal path. Among the many factors that affect the emotions of the elderly, it can accurately identify which are the core causal factors that truly dominate emotional changes and which are secondary or insignificant factors. This provides more accurate and key information for subsequent processing, thereby improving the accuracy and effectiveness of the model's analysis of the emotional change causal relationship.

[0061] Specifically, in the embodiments of the present application, the emotional state time sequence evolution semantic encoding feature generation subunit is configured to perform log semantic dynamic walk coding based on a graph convolutional neural network model on the time sequence distribution of the emotional log summary semantic causal association topology feature matrix and the emotional log summary semantic embedding encoding vector to obtain an emotional log summary surface layer context dynamic walk semantic encoding vector. This process can be represented by the formula:

[0062]

[0063] wherein x i is the i-th emotional log summary semantic embedding encoding vector in the time sequence distribution of the emotional log summary semantic embedding encoding vector, M is the emotional log summary semantic causal association topology feature matrix, GCN is graph convolution processing, H surface is an emotional log summary surface layer context dynamic walk semantic encoding vector.

[0064] The emotion log summary semantic causal correlation topology feature matrix and the time series distribution of the emotion log summary semantic deep implicit embedding coding vector are subjected to the log semantic dynamic walk coding based on the graph convolutional neural network model to obtain an emotion log summary hidden layer context dynamic walk semantic coding vector, and the process can be represented by a formula as follows:

[0065]

[0066] wherein v i is the i-th emotion log summary semantic deep implicit embedding coding vector in the time series distribution of the emotion log summary semantic deep implicit embedding coding vector, M is the emotion log summary semantic causal correlation topology feature matrix, GCN is graph convolution processing, H hidden is the emotion log summary hidden layer context dynamic walk semantic coding vector;

[0067] The emotion log summary hidden layer context dynamic walk semantic coding vector and the emotion log summary surface layer context dynamic walk semantic coding vector are fused to obtain the emotion state time series evolution semantic coding vector, and the process can be represented by a formula as follows:

[0068] H final = γ·H surface + (1-γ)·H hidden

[0069] wherein H surface is the emotion log summary surface layer context dynamic walk semantic coding vector, H hidden is the emotion log summary hidden layer context dynamic walk semantic coding vector, γ is a fusion weighting parameter, and H final is the emotion state time series evolution semantic coding vector.

[0070] Correspondingly, in order to integrate the explicit semantic information in the original emotion log summary, so as to provide a clear and explicit basis for the model to intuitively understand the emotional state of the elderly, the emotion log summary semantic causal correlation topology feature matrix and the time series distribution of the emotion log summary semantic embedding coding vector are subjected to the log semantic dynamic walk coding based on the graph convolutional neural network model. Specifically, by simulating the propagation process of the emotion log summary semantic features in the topology structure through the dynamic walk mechanism, the explicit semantics between nodes are recursively aggregated from local to global, which can comprehensively and systematically integrate the explicit semantic information in the emotion log summary, and make up for the shortcomings of the traditional single-layer analysis method.

[0071] It should be understood that the emotional log summary implicit layer semantic feature involves a potential embedding of the feature, which contains more complex and hidden information. When performing semantic encoding, attention should be paid to the multi-hop propagation of high-order information and the distributed decoupling of deep features to avoid the over-smoothing phenomenon caused by deep topology feature propagation. The emotional log summary semantic deep implicit embedding coding vector is processed by the log semantic dynamic walk coding processing based on the graph convolutional neural network model, which can deeply mine the complex semantic patterns hidden in the emotional log summary, capture deep temporal dependence, enhance the generalization ability of the emotional log summary implicit layer semantic feature expression, and thus more comprehensively and deeply understand the emotional state of the elderly. That is, the emotional log summary implicit layer context dynamic walk semantic encoding vector can mine the potential influence of the long-term psychological state of the elderly on the current emotion, or the subtle emotional change relationship between different rehabilitation stages, which can provide a deep perspective and basis for the model to deeply understand the emotional changes of the elderly.

[0072] Correspondingly, the emotional log summary surface layer context dynamic walk semantic encoding vector captures the explicit semantic information of the emotional log summary, while the emotional log summary implicit layer context dynamic walk semantic encoding vector captures the implicit semantic information of the emotional log summary, and the two have strong complementarity. By fusing vectors containing different levels of semantic information, their respective advantages can be fully utilized to form an emotional state temporal evolution semantic encoding vector that is more comprehensive, accurate, has stronger discrimination ability and complete representation. That is, the fused emotional state temporal evolution semantic encoding vector comprehensively integrates the explicit and implicit semantic information in the emotional log summary and the causal relationship and multi-level context association information therein. It can sensitively capture the potential trend of the emotional changes of the elderly, thereby improving the model's understanding ability of the emotional state of the elderly.

[0073] In the embodiments of the present application, the recognition result generation unit 123 is configured to obtain the emotion state recognition result based on the emotion state time evolution semantic encoding feature. Specifically, in the embodiments of the present application, the recognition result generation unit is configured to: input the emotion state time evolution semantic encoding vector into a classifier-based emotion state recognizer to obtain the emotion state recognition result, wherein the emotion state recognition result is used to indicate whether the emotion state of the target elderly person is positively evolved or negatively evolved. It can be understood that by inputting the emotion state time evolution semantic encoding vector into the classifier-based emotion state recognizer, the classification function of the classifier can be used to judge the processed semantic encoding vector, so as to accurately determine the evolution direction of the emotion state. Specifically, the classifier is a kind of machine learning model, and its main function is to divide the input data features into different categories. It can learn the relationship between different emotion evolution patterns and their corresponding features through a large number of labeled training data. When the training is completed, the classifier can classify the new input emotion state time evolution semantic encoding vector into the corresponding emotion state recognition result according to the learned feature mapping pattern. The generated recognition result can provide a key decision basis for psychological intervention. In one specific embodiment of the present application, the emotion state time evolution semantic encoding vector is input into the classifier-based emotion state recognizer to obtain the emotion state recognition result, wherein the emotion state recognition result is used to indicate whether the emotion state of the target elderly person is positively evolved or negatively evolved, including: using the full connection layer of the classifier to perform full connection encoding on the emotion state time evolution semantic encoding vector to obtain an emotion state time evolution semantic full connection encoding feature vector; inputting the emotion state time evolution semantic full connection encoding feature vector into the Softmax classification function of the classifier to obtain the probability value of the emotion state time evolution semantic encoding vector belonging to each classification label, wherein the classification label includes a classification label used to indicate that the emotion state of the target elderly person is positively evolved and a classification label used to indicate that the emotion state of the target elderly person is negatively evolved; and determining the classification label corresponding to the maximum probability value as the emotion state recognition result.

[0074] In the embodiments of the present application, the intervention signal generation module 130 is configured to determine whether to generate a psychological intervention prompt signal according to the emotional state recognition result. Specifically, in the embodiments of the present application, the intervention signal generation module is configured to generate the psychological intervention prompt signal in response to the emotional state of the target elderly person being negatively evolved. It can be understood that the negative evolution of the emotion of the elderly person after surgery can have many adverse effects on their rehabilitation. By generating a psychological intervention prompt signal in a timely manner, it helps medical staff to first understand that the elderly person may be experiencing emotional distress, and to intervene in a timely manner and provide personalized psychological support and intervention measures for the target elderly person to reduce the impact of negative emotions on the elderly. Specifically, the psychological intervention prompt signal can contain the basic information of the elderly person (such as name, ward number), the specific description of the emotional state change (such as "showing high anxiety for three consecutive days"), and other information, and is accurately delivered to the relevant medical staff through the internal communication system of the hospital, the work mobile phone application of the medical staff, etc. To ensure that the psychological intervention prompt signal is indeed received and actioned, the system contains a feedback mechanism, which requires the medical staff to confirm that they have read the prompt after receiving it, and need to fill in the next plan or the measures already taken. In this way, it can be ensured that when the elderly person has a negative emotional tendency, necessary psychological support can be provided quickly and effectively.

[0075] In summary, the artificial intelligence-based elderly postoperative rehabilitation nursing system 100 based on the embodiments of the present application is illustrated, which first performs encryption processing and data regularization on the collected emotional log data set of the target elderly person to obtain the time sequence distribution of the encrypted emotional log, then uses an artificial intelligence-based data analysis method to summarize the content and embed the semantics of the time sequence distribution of the encrypted emotional log to obtain the time sequence distribution of the emotional log summary semantic features, and obtains the emotional recognition result based on the context causal correlation coding features of the time sequence distribution of the emotional log summary semantic features, and finally determines whether to generate a psychological intervention prompt signal based on the emotional recognition result. In this way, by continuously monitoring and analyzing the emotional fluctuations and their regularities of the elderly, it helps to more accurately capture the emotional state, and thus improves the timeliness, effectiveness and intelligent degree of psychological intervention.

Claims

1. An artificial intelligence-based postoperative rehabilitation care system for the elderly, characterized by, The method comprises the following steps: a data collection preprocessing module is configured to collect emotional logs of a target elderly person to obtain a data set of the emotional logs, and to encrypt and normalize the data set of the emotional logs to obtain a time sequence distribution of encrypted emotional logs; an emotional state recognition module is configured to perform emotional analysis on the time sequence distribution of the encrypted emotional logs based on text semantics to obtain an emotional state recognition result, wherein the emotional state recognition module comprises: a log time sequence summary embedding unit configured to perform content summarization and semantic embedding on the time sequence distribution of the encrypted emotional logs to obtain a time sequence distribution of emotional log summary semantic embedding coding features; a log time sequence coding unit configured to perform emotional log context causal correlation coding on the time sequence distribution of the emotional log summary semantic embedding coding features to obtain emotional state time sequence evolution semantic coding features; and a recognition result generation unit configured to obtain the emotional state recognition result based on the emotional state time sequence evolution semantic coding features; an intervention signal generation module is configured to determine whether to generate a psychological intervention prompt signal based on the emotional state recognition result; wherein the log time sequence coding unit comprises: an emotional log summary implicit semantic feature mining subunit configured to perform emotional log summary implicit semantic feature mining on the time sequence distribution of the emotional log summary semantic embedding coding vector to obtain a time sequence distribution of an emotional log summary semantic deep implicit embedding coding vector; an emotional log summary semantic causal analysis subunit configured to perform emotional log summary semantic causal correlation factor calculation and emotional log summary semantic causal triggering on the time sequence distribution of the emotional log summary semantic deep implicit embedding coding vector to obtain an emotional log summary semantic causal correlation topology feature matrix; and an emotional state time sequence evolution semantic coding feature generation subunit configured to perform log semantic context dynamic walk coding fusion on the time sequence distribution of the emotional log summary semantic deep implicit embedding coding vector and the time sequence distribution of the emotional log summary semantic embedding coding vector based on the emotional log summary semantic causal correlation topology feature matrix to obtain an emotional state time sequence evolution semantic coding vector as the emotional state time sequence evolution semantic coding features.

2. The artificial intelligence-based postoperative rehabilitation care system for the elderly according to claim 1, characterized in that, The data collection preprocessing module comprises: an emotional log collection unit configured to collect emotional logs of the target elderly person to obtain a data set of the emotional logs; an emotional log encryption processing unit configured to encrypt each emotional log in the data set of the emotional logs to obtain a data set of encrypted emotional logs; an emotional log time sequence distribution adjustment unit configured to adjust the data distribution pattern of the data set of the encrypted emotional logs based on the time dimension according to the time stamp to obtain the time sequence distribution of the encrypted emotional logs. 3.The AI-based postoperative rehabilitation care system for the elderly according to claim 2, characterized in that, The log time sequence summary embedding unit comprises: an emotional log summary generation subunit configured to input each encrypted emotional log in the time sequence distribution of the encrypted emotional logs into a content summarization encoder based on a large language model to obtain a time sequence distribution of emotional log summaries; The emotion log summary embedding coding subunit is configured to perform semantic embedding coding on each emotion log summary in the time sequence distribution of the emotion log summary to obtain a time sequence distribution of emotion log summary semantic embedding coding vectors as a time sequence distribution of emotion log summary semantic embedding coding features. 4.The AI-based postoperative rehabilitation care system for the elderly according to claim 3, characterized in that, The emotion log summary embedding coding subunit is configured to perform semantic embedding coding on each emotion log summary in the time sequence distribution of the emotion log summary by using a semantic encoder based on a bidirectional recurrent neural network model to obtain a time sequence distribution of emotion log summary semantic embedding coding vectors. 5.The AI-based postoperative rehabilitation care system for the elderly according to claim 1, wherein The emotion log summary semantic cause-effect analysis subunit comprises: The emotion log summary semantic cause-effect correlation factor calculation primary subunit is configured to calculate emotion log summary semantic cause-effect correlation factors between any two emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vectors to obtain an emotion log summary semantic cause-effect correlation topology matrix composed of a plurality of emotion log summary semantic cause-effect correlation factors. The emotion log summary semantic cause-effect trigger primary subunit is configured to perform emotion log summary semantic cause-effect triggering on the emotion log summary semantic cause-effect correlation topology matrix based on a gating function to obtain an emotion log summary semantic cause-effect correlation topology feature matrix. 6.The AI-based postoperative rehabilitation care system for the elderly according to claim 5, wherein The emotion log summary semantic cause-effect correlation factor calculation primary subunit is configured to: calculate a correlation matrix between any two emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vectors to obtain a time sequence distribution of emotion log summary semantic deep correlation embedding coding matrices; calculate emotion log summary semantic cause-effect correlation factors of each emotion log summary semantic deep correlation embedding coding matrix in the time sequence distribution of the emotion log summary semantic deep correlation embedding coding matrices to obtain the emotion log summary semantic cause-effect correlation topology matrix composed of a plurality of emotion log summary semantic cause-effect correlation factors, wherein the emotion log summary semantic cause-effect correlation factors are calculated from a maximum value, a mean value, a variance, and a cause-effect correlation bias term of the emotion log summary semantic deep correlation embedding coding matrix; in response to the variance of the emotion log summary semantic deep correlation embedding coding matrix being greater than or equal to a preset threshold, taking a weighted average of distances between any two emotion log summary semantic deep implicit embedding coding vectors in the time sequence distribution of the emotion log summary semantic deep implicit embedding coding vectors as the cause-effect correlation bias term, in response to the variance of the emotion log summary semantic deep correlation embedding coding matrix being less than the preset threshold, taking a weighted average of the emotion log summary semantic deep correlation embedding coding matrices as the cause-effect correlation bias term. 7.The AI-based postoperative rehabilitation care system for the elderly according to claim 6, wherein The emotion state time sequence evolution semantic coding feature generation subunit is configured to: perform log semantic dynamic walk coding on a time sequence distribution of the emotion log summary semantic cause-and-effect correlation topological feature matrix and the emotion log summary semantic deep implicit embedding coding vector based on the graph convolutional neural network model to obtain an emotion log summary hidden layer context dynamic walk semantic coding vector; perform log semantic dynamic walk coding on a time sequence distribution of the emotion log summary semantic cause-and-effect correlation topological feature matrix and the emotion log summary semantic deep implicit embedding coding vector based on the graph convolutional neural network model to obtain an emotion log summary hidden layer context dynamic walk semantic coding vector; fuse the emotion log summary hidden layer context dynamic walk semantic coding vector and the emotion log summary surface layer context dynamic walk semantic coding vector to obtain the emotion state time sequence evolution semantic coding vector. 8.The AI-based postoperative rehabilitation care system for the elderly according to claim 7, characterized in that, The recognition result generation unit is configured to: input the emotion state time sequence evolution semantic coding vector into an emotion state recognizer based on a classifier to obtain an emotion state recognition result, the emotion state recognition result being used to indicate whether the emotion state of the target elderly object is positively evolved or negatively evolved. 9.The AI-based postoperative rehabilitation care system for the elderly according to claim 8, wherein, The intervention signal generation module is configured to: in response to the emotion state of the target elderly object being negatively evolved, generate the psychological intervention prompt signal.

Images