AI intelligent interaction method and system based on digital home-based care service

Abstract

The invention discloses an AI intelligent interaction method based on a digital home-based care service. The method comprises the following steps: acquiring a home-based care monitoring report of the current home-based elderly; extracting a health description vector of the current home elderly from the home-based care monitoring report; inputting the health description vector into a preset home-based care evaluation model, and outputting the care risk level of the current home-based elderly; based on the old-age care risk level of the current home elderly, a predefined old-age care strategy corresponding to the risk level is called to interact with the user; wherein the health description vector is a flattened vector of a text interaction feature generated by interaction item modeling; according to the method, the health description text is extracted from the home-based care monitoring report, the entity pairs are subjected to interactive modeling, and finally the text interactive features capable of comprehensively reflecting different health sign indexes are generated, so that the care risk assessment model can capture the interaction effect of the multi-dimensional sign indexes conveniently, and the risk assessment efficiency is improved. And more personalized and accurate old-age service experience is provided for the user.

Description

Technical Field

[0001] The present invention relates to the field of home-based elderly care services, and in particular to an AI intelligent interaction method based on digital home-based elderly care services. Background Art

[0002] In existing home-based elderly care service technologies, elderly care risk assessment usually relies on the processing of traditional structured data, such as numerical health indicators such as blood pressure and blood sugar. However, in actual application scenarios, home-based elderly care monitoring reports contain a large amount of unstructured text data recorded in natural language. These text data contain rich health information, such as disease descriptions, symptom records, and doctor's advice. The prior art with patent publication number CN117226866A discloses an intelligent elderly care companion robot, which uses intelligent voice interaction technology to provide a new type of home-based elderly care solution that integrates smart home services, personalized companionship services in emotions and entertainment, and life safety protection monitoring services, thereby improving the quality of life and safety of the elderly.

[0003] However, in the existing technology, there are significant deficiencies in the analysis of unstructured elderly care health text data. For example: There are often potential interactions between different entities in health description texts (such as "hypertension" and "obesity", etc.), and these relationships may significantly affect the results of pension risk assessment. For example, the combination of "hypertension" and "obesity" may mean a higher health risk, but existing technologies usually only analyze a single entity in isolation and fail to capture the interactions between entities, resulting in insufficient accuracy of prediction results. In addition, entity descriptions in health texts may have synonyms or variants (such as "hyperglycemia" and "high blood sugar"), and existing technologies lack unified semantic standardization processing, resulting in feature duplication or missing information. Summary of the invention

[0004] This embodiment provides an AI intelligent interaction method and system based on digital home-based elderly care services, exploring how to solve the problem of how to capture entity interactions in home-based elderly care monitoring texts.

[0005] In a first aspect, the present invention provides an AI intelligent interaction method based on digital home-based elderly care services, comprising: Obtain the current home-based elderly care monitoring report; Extracting the health description vector of the elderly living at home from the home-based elderly care monitoring report; The health description vector is input into a preset home-based elderly care assessment model to output the elderly care risk level of the current elderly people at home.

[0006] Based on the current elderly care risk level of the elderly living at home, the predefined elderly care strategy corresponding to the risk level is retrieved to interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

[0007] In some embodiments, the modeling method of the home-based elderly care evaluation model includes: S1. Obtain a home-based elderly care monitoring sample set; wherein the sample set includes: a number of home-based elderly care monitoring samples marked with home-based elderly care risk levels; S2. Performing text recognition on the home-based elderly care monitoring samples to extract text interaction features; S3, flattening the text interaction features to generate a health description vector of a one-dimensional array; S4, matching the health description vector with the elderly care risk level to generate home risk assessment training samples; S5. Obtain several home risk assessment training samples and aggregate them into a home risk assessment training set.

[0008] S6. Use the general model to receive the home risk assessment training set as input, and perform iterative supervised learning until the general model is iterated into the home-based elderly care assessment model.

[0009] In some of the embodiments, obtaining a home-based elderly care monitoring sample set includes: S1-1. Import the electronic health reports of elderly people living at home from the electronic health record system; S1-2, extracting a number of physical signs to be monitored from the electronic health report; S1-3, determine the pension risk level of the electronic health report based on a number of monitored physical signs and indicators and predefined physical sign abnormality standards; The determination expression of the pension risk level is: ; Among them, RS represents the pension risk level score, represents the i-th physical sign index to be detected, represents the risk weight of the i-th physical sign indicator to be detected, and n represents the number of physical sign indicators to be monitored; S1-4, filling a number of physical signs to be monitored and the corresponding elderly care risk levels into a structured home-based elderly care monitoring template to obtain the home-based elderly care monitoring sample; S1-5. Obtain a number of home-based elderly care monitoring samples, and aggregate them into the home-based elderly care monitoring sample set.

[0010] In some of the embodiments, text recognition is performed on the home-based elderly care monitoring sample to extract its text interaction features, including: S2-1, extracting the health description entity set of the home-based elderly care monitoring sample; S2-2. Modeling the interaction item of any two health description entities in the health description entity set to generate the text interaction feature.

[0011] In some embodiments, extracting a health description entity set of the home-based elderly care monitoring sample includes: S2-1-1. Select health description text from the home-based elderly care monitoring sample; S2-1-2, importing the health description text into the SC toolkit for entity recognition, and filtering out the original triple entity set of the health description text; S2-1-3. Semantically standardize the original triple entities in the original triple entity set to generate standard triple entities, and define them as health description entities of the home-based elderly care monitoring sample.

[0012] In some of the embodiments, selecting a health description text from the home-based elderly care monitoring sample includes: S2-1-2-1, mapping the health description text to a multidimensional vector space through word embedding to generate a word vector containing context features; S2-1-2-2, using the sequence labeling module of the SC toolkit to analyze the word vector containing the context features word by word to identify the entity name; S2-1-2-3. Based on the label classification rules preset by the SC toolkit, the entity name is assigned a category to generate an entity category; S2-1-2-4, recording the starting position and the ending position of the entity category in the health description text, and defining the enclosed area of ​​the starting position and the ending position as the entity boundary; S2-1-2-5. Filter out the entity name, entity category and entity boundary, and define their combination as the original triple entity of the health description text; wherein the triple of the original triple entity is: entity name, entity category and entity boundary.

[0013] S2-1-2-6. Filter out all original triple entities from the health description text, and define their combination as the original triple entity set.

[0014] In some embodiments, the step of defining the health description entity of the home-based elderly care monitoring sample includes: S2-1-3-1. Based on any two entity names in the original triple entity set, calculate the word vector similarity between any two original triple entities; The calculation expression of the word vector similarity between any two original triple entities is: ; Among them, CS represents the word vector similarity, Represent the word vectors of any two original triple entities, represents the dot product of the word vectors of any two original triple entities, and Represent word vectors The modulus and word vector The module length; S2-1-3-2, judging whether the word vector similarity between any two original triple entities exceeds a first similarity threshold; S2-1-3-3, if the word vector similarity exceeds the first similarity threshold, define the entity name exceeding the first similarity threshold as a synonym variant name; S2-1-3-4. Replace the synonymous variant names with standard expression names so that any two original triple entities whose word vector similarity exceeds the first similarity threshold are converted into the same standard triple entity.

[0015] In some embodiments, performing interaction modeling on any two health description entities in the health description entity set to generate the text interaction feature includes: S2-2-1, associate the word vector similarity between any two original triple entities to the health description entity set; S2-2-2, judging whether the word vector similarity between any two health description entities exceeds a second similarity threshold; S2-2-3. If the word vector similarity exceeds the second similarity threshold, select health description entity pairs with significant semantic relevance in the health description entity set; S2-2-4. Model the interaction items of the health description entity pairs to generate the text interaction features.

[0016] In some embodiments, a general model is used to receive the home risk assessment training set as input, and iterative supervised learning is performed until the general model is iterated into the home-based elderly care assessment model, including: S6-1, extract the first batch of home risk assessment training samples from the home risk assessment training set as input features; S6-2, performing forward propagation on the first batch of home risk assessment training samples, and outputting the first batch of pension risk level prediction values; S6-3, calculating the prediction error between the predicted value of the pension risk level in the first batch and the actual pension risk level; S6-4, if the prediction error does not reach the minimum error, updating the model parameters of the general model until the minimum error is reached; S6-5. If the prediction error reaches the minimum error, the model parameters of the current batch are regarded as the model parameters of the home-based elderly care evaluation model; S6-6. Export the model parameters of the home-based elderly care evaluation model, and use them to construct the home-based elderly care evaluation model.

[0017] Compared with the prior art, the present invention provides an AI intelligent interaction method based on digital home-based elderly care services. By extracting health description text from the home-based elderly care monitoring report, and then calculating the word vector similarity between health description entities, health description entity pairs with significant semantic correlation are screened out, and these entity pairs are interactively modeled. Finally, text interaction features that can comprehensively reflect different health signs and indicators are generated, thereby facilitating the elderly care risk assessment model to capture the interaction of multi-dimensional signs and indicators, output a more accurate home-based elderly care risk level, and provide users with a more personalized and accurate elderly care service experience.

[0018] In a second aspect, the present invention provides an AI intelligent interactive system based on digital home-based elderly care services, which executes the interactive method described in the first aspect, and the system includes: The elderly care monitoring report acquisition module is used to obtain the home elderly care monitoring report of the elderly living at home; A health description vector extraction module, used to extract the health description vector of the current elderly person at home from the home-based elderly care monitoring report; The elderly care risk level output module is used to input the health description vector into a preset home-based elderly care assessment model and output the elderly care risk level of the current elderly people at home.

[0019] The pension strategy retrieval module is used to retrieve the predefined pension strategy corresponding to the risk level based on the pension risk level of the elderly living at home and interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

[0020] Compared with the prior art, the beneficial effects of the AI ​​intelligent interaction system based on digital home-based elderly care services of the present invention are the same as the beneficial effects of the AI ​​intelligent interaction method based on digital home-based elderly care services described above, so they will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of an AI intelligent interaction method based on digital home-based elderly care services according to the present invention; Figure 2 This is a schematic diagram of the modeling process of the pension assessment model of the present invention; Figure 3 It is a schematic diagram of the conversion process of the standard triple entity of the present invention; Figure 4 This is a structural block diagram of an AI intelligent interactive system based on digital home-based elderly care services in the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example 1: See Figures 1 to 3 The present invention provides an AI intelligent interaction method based on digital home-based elderly care services, comprising the following steps: Obtain the current home-based elderly care monitoring report; Extracting the health description vector of the elderly living at home from the home-based elderly care monitoring report; The health description vector is input into a preset home-based elderly care assessment model to output the elderly care risk level of the current elderly people at home.

[0024] Based on the current elderly care risk level of the elderly living at home, the predefined elderly care strategy corresponding to the risk level is retrieved to interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

[0025] This embodiment realizes the digital expression of the health information of the elderly at home by extracting the health description vector from the home-based elderly care monitoring report and combining the interactive item modeling and flattening operation of the text interaction feature. By inputting the generated health description vector into the preset home-based elderly care evaluation model, the elderly care risk level of the elderly at home can be quickly calculated and output.

[0026] Among them, predefined elderly care strategies refer to targeted intervention or management plans formulated in advance based on the elderly care risk level and the specific health status of the elderly at home. The core of these strategies is to provide appropriate health guidance, medical support and life care for the elderly at different risk levels, ensuring that the elderly can obtain appropriate elderly care services in their home environment.

[0027] In this embodiment: Low risk: For elderly people with a lower level of retirement risk, predefined retirement strategies may include healthy lifestyle recommendations (such as dietary adjustments, moderate exercise) and regular monitoring of physical condition to maintain good health.

[0028] Medium risk: For elderly people with a medium level of retirement risk, strategies may include more frequent health monitoring, more detailed health management plans (such as monitoring changes in indicators such as blood sugar and blood pressure), and appropriate community medical interventions, such as regular follow-up and medication management.

[0029] High risk: For elderly people with a higher level of risk in elderly care, predefined elderly care strategies may involve emergency intervention measures, such as arranging special examinations, personalized treatment plans, strengthening nursing support (such as regular visits by family caregivers), and even recommending transfer to a professional elderly care institution for continuous care.

[0030] In this embodiment, the modeling method of the home-based elderly care evaluation model includes: Step S1, obtaining a home-based elderly care monitoring sample set; wherein the sample set includes: a number of home-based elderly care monitoring samples marked with home-based elderly care risk levels; Step S2, performing text recognition on the home-based elderly care monitoring sample to extract its text interaction features; Step S3, flattening the text interaction features to generate a health description vector of a one-dimensional array; Specifically, text interaction features may be multi-dimensional features contained in the monitoring data of an elderly person living at home, such as blood pressure (high or low), blood sugar, weight, heart rate, sleep quality and other indicators; feature flattening refers to flattening the text interaction features carrying multi-dimensional health monitoring information into a one-dimensional array. The flattened health description vector can be directly passed into the home-based elderly care evaluation model as an input feature.

[0031] Step S4: Match the health description vector and the pension risk level to generate a home risk assessment training sample; Step S5: obtain a number of home risk assessment training samples and aggregate them into a home risk assessment training set.

[0032] Step S6: Use the general model to receive the home risk assessment training set as input, and perform iterative supervised learning until the general model is iterated into the home-based elderly care assessment model.

[0033] In this embodiment, the general model can be any supervised learning model with AI intelligent learning capability, which is used to learn rules from the home risk assessment training set, and finally form a home-based elderly care assessment model that can predict the elderly care risk level, such as a random forest or a neural network.

[0034] This embodiment obtains data from a home-based elderly care monitoring sample set marked with elderly care risk levels, extracts text interaction features in the monitoring data through text recognition, and further converts these multi-dimensional health data into health description vectors in a one-dimensional array format through feature flattening operations. This ensures that the data can be easily passed as input to the home-based elderly care assessment model for learning. Next, the health description vector is matched with the elderly care risk level to generate a home-based risk assessment training sample set. Using the training samples, the general model is iteratively trained through supervised learning, and finally forms a home-based elderly care assessment model that can predict the elderly care risk level of the elderly at home.

[0035] Furthermore, the step S1 specifically includes: S1-1. Import the electronic health reports of elderly people living at home from the electronic health record system; Electronic Health Record (EHR) EHR is a general term that is widely used in various medical and health management systems. Many hospitals, community health service centers and health management agencies use EHR systems to record patients' health information. For example, they include: national health record systems or local community health management systems. Among them, electronic health reports can be physical examination reports, health follow-up records, chronic disease management reports or personal health assessment results recorded in the file by elderly people at home.

[0036] Exemplarily, these include regular or irregular physical examination results of elderly people at home, such as blood sugar, blood pressure, electrocardiogram and other indicators; records of regular follow-up of elderly people at home for chronic disease management or data tracking after health interventions; health score results generated through questionnaires and doctor evaluations, etc.

[0037] S1-2, extracting a number of physical signs to be monitored from the electronic health report; The vital signs to be monitored may include blood pressure, blood sugar, heart rate, weight, etc., which represent the current health status of the elderly at home.

[0038] S1-3, determine the pension risk level of the electronic health report based on a number of monitored physical signs and indicators and predefined physical sign abnormality standards; The predefined abnormal physical sign standard refers to setting abnormal standards for each physical sign indicator, for example: Blood pressure: If the systolic blood pressure is greater than 180 mmHg or the diastolic blood pressure is greater than 110 mmHg, it is considered high blood pressure.

[0039] Blood sugar: If the fasting blood sugar is greater than 126 mg / dL, it is considered high blood sugar.

[0040] Weight: If the weight is too light or too heavy, and exceeds a certain BMI threshold (for example, BMI < 18.5 or BMI > 30), it is considered abnormal. By converting the abnormality of each physical sign indicator into a numerical value (usually a binary classification: normal or abnormal), and then summarizing the abnormal scores of all physical signs, a comprehensive risk score is finally calculated to determine the pension risk level.

[0041] The determination expression of the pension risk level is: ; Among them, RS represents the pension risk level score, represents the i-th physical sign index to be detected, represents the risk weight of the i-th vital sign indicator to be detected, and n represents the number of vital sign indicators to be monitored.

[0042] For the pension risk level score, it can be divided into different pension risk levels: when If it is less than 3, it is judged as low risk; When RS is greater than or equal to 3 and less than 6, it is judged as medium risk; When RS is greater than or equal to 6, it is considered high risk; Of course, the actual retirement situation can be further refined based on the score, such as no risk, extremely low risk, medium-high risk, etc.

[0043] S1-4, filling a number of physical signs to be monitored and the corresponding elderly care risk levels into a structured home-based elderly care monitoring template to obtain the home-based elderly care monitoring sample; S1-5. Obtain a number of home-based elderly care monitoring samples, and aggregate them into the home-based elderly care monitoring sample set.

[0044] This embodiment provides detailed records of multiple health indicators such as blood sugar, blood pressure, weight, heart rate, etc. by importing the electronic health report of the elderly at home. Then, a number of monitored physical signs are extracted from the electronic health report to reflect the health status of the elderly at home. According to the predefined physical sign abnormality standards (such as blood pressure higher than a specific value, blood sugar exceeding the standard, etc.), the health abnormality of each sign is determined, and the elderly care risk level is calculated based on the abnormal situation. Finally, the extracted physical signs and risk levels to be monitored are filled into the structured home-based elderly care monitoring template to form a home-based elderly care monitoring sample, which is summarized as a home-based elderly care monitoring sample set.

[0045] Furthermore, step S2 specifically includes: S2-1, extracting the health description entity set of the home-based elderly care monitoring sample; S2-2. Modeling the interaction item of any two health description entities in the health description entity set to generate the text interaction feature.

[0046] This embodiment extracts a set of health description entities from home-based elderly care monitoring samples and models interaction items based on the relationships between health description entities, thereby generating text interaction features. This effectively captures the interactions between health description entities and provides more comprehensive feature information for subsequent elderly care risk assessment.

[0047] Furthermore, the step S2-1 specifically includes: S2-1-1. Select health description text from the home-based elderly care monitoring sample; S2-1-2, importing the health description text into the SC toolkit for entity recognition, and filtering out the original triple entity set of the health description text; In this embodiment, the SC toolkit refers to the SciSpaCy toolkit, which is a toolkit based on natural language processing (NLP) technology and is specifically used for tasks such as information extraction, entity recognition, and semantic standardization in medical texts. It is usually used to extract key entities from health-related texts, such as disease names, symptoms, signs, treatment methods, etc. Entity recognition using the SC toolkit means using the toolkit to analyze health description texts, identify the medical entity information contained therein, and organize these entities into a set of original triple entities.

[0048] S2-1-3. Semantically standardize the original triple entities in the original triple entity set to generate standard triple entities, and define them as health description entities of the home-based elderly care monitoring sample.

[0049] This embodiment selects health description text from home-based elderly care monitoring samples, and uses the SC toolkit to perform entity recognition on the text to extract the original triple entity set reflecting health information. The SC toolkit uses natural language processing technology to efficiently extract medical-related entity information from text and organize it in a structured manner. Then, these original triple entities are semantically standardized to ensure that entities in different forms of expression have a unified standard representation, generate a standard triple entity set, significantly improve the standardization of health description entities, and lay a high-quality data foundation for subsequent text interaction feature modeling.

[0050] In step S2-1-2, the screening step includes: S2-1-2-1, mapping the health description text to a multidimensional vector space through word embedding to generate a word vector containing context features; Among them, word embedding is a technology in natural language processing (NLP) that can capture the semantic similarity and contextual information of words by mapping them into a continuous, low-dimensional vector space. This mapping makes the vectors of semantically similar words (such as "doctor" and "nurse") in the same vector space relatively close.

[0051] S2-1-2-2, using the sequence labeling module of the SC toolkit to analyze the word vector containing the context features word by word to identify the entity name; S2-1-2-3. Based on the label classification rules preset by the SC toolkit, the entity name is assigned a category to generate an entity category; S2-1-2-4, recording the starting position and the ending position of the entity category in the health description text, and defining the enclosed area of ​​the starting position and the ending position as the entity boundary; S2-1-2-5. Filter out the entity name, entity category and entity boundary, and define their combination as the original triple entity of the health description text; wherein the triple of the original triple entity is: entity name, entity category and entity boundary.

[0052] Exemplarily, the triples are: Entity Name: Hypertension Entity Type: Disease Entity position: The position in the text where hypertension occurs (e.g., "from the 5th to the 8th character").

[0053] S2-1-2-6. Filter out all original triple entities from the health description text, and define their combination as the original triple entity set.

[0054] This embodiment maps the health description text to a multidimensional vector space to generate a word vector containing contextual features, thereby ensuring the accurate expression of the semantic information of words in the text. The sequence labeling module of the SC toolkit is used to analyze the text content word by word, identify the entity name and assign categories according to preset rules, and then locate the entity in the text, and construct an original triple entity set containing entity name, category and boundary. This step can extract structured key information from the health description text, such as the disease name and its location, and generate a comprehensive and standardized original triple entity set.

[0055] In step S2-1-3, the steps of defining the health description entity include: S2-1-3-1. Based on any two entity names in the original triple entity set, calculate the word vector similarity between any two original triple entities; The calculation expression of the word vector similarity between any two original triple entities is: ; Among them, CS represents the word vector similarity, Represent the word vectors of any two original triple entities, represents the dot product of the word vectors of any two original triple entities, and Represent word vectors The modulus and word vector The mold length.

[0056] In this implementation, the word vector similarity uses cosine similarity, and the result of cosine similarity is between -1 and 1. If the similarity is 1, it means that the two word vectors are exactly the same. If the similarity is 0, it means that the two word vectors are orthogonal and have no similarity. If the similarity is -1, it means that the two word vectors are completely opposite. In the field of natural language processing, using cosine similarity to calculate the similarity of word vectors between different entities (such as disease names, symptoms, drugs, etc.) can measure the semantic proximity between the two, so as to determine whether they belong to the same category or whether there is a certain semantic relationship.

[0057] S2-1-3-2, judging whether the word vector similarity between any two original triple entities exceeds a first similarity threshold; S2-1-3-3, if the word vector similarity exceeds the first similarity threshold, define the entity name exceeding the first similarity threshold as a synonym variant name; S2-1-3-4. Replace the synonymous variant names with standard expression names so that any two original triple entities whose word vector similarity exceeds the first similarity threshold are converted into the same standard triple entity.

[0058] Among them, the standard expression name is the standard expression name defined in the field of home-based elderly care services based on a unified standard vocabulary or industry specifications.

[0059] This embodiment effectively evaluates the semantic proximity of entity names by calculating the word vector similarity between the original triple entities. The semantic similarity between two entity word vectors is accurately measured by cosine similarity, and the entity names exceeding the threshold are determined as synonymous variant names based on the first similarity threshold. By replacing synonymous variant names with standard expression names, unification and standardization at the semantic level are achieved. This embodiment ensures that entities with different expressions can be standardized in the field of home-based elderly care services, providing high-quality semantic input for health description modeling.

[0060] Furthermore, the step S2-2 specifically includes: S2-2-1, associate the word vector similarity between any two original triple entities to the health description entity set; S2-2-2, judging whether the word vector similarity between any two health description entities exceeds a second similarity threshold; S2-2-3. If the word vector similarity exceeds the second similarity threshold, select health description entity pairs with significant semantic relevance in the health description entity set; Furthermore, the significant semantic relevance means that there is a strong semantic connection between the two health description entities, measured by the word vector similarity. If the word vector similarity of two health description entities is high to a certain extent (exceeding the preset similarity threshold), they are considered to be semantically significantly related. This correlation means that the two entities express similar concepts or have related health information in a certain context, which may affect the risk of old-age care together.

[0061] In this embodiment, by calculating the word vector similarity between health description entities, their semantically significant relationship is determined. When the similarity between two health description entities exceeds the second similarity threshold, they can be considered to be semantically significantly related, and then these entity pairs are modeled as interaction items.

[0062] S2-2-4. Model the interaction items of the health description entity pairs to generate the text interaction features.

[0063] This embodiment screens out entity pairs with significant semantic relevance by calculating the word vector similarity between health description entities. Significant semantic relevance is measured by the word vector similarity exceeding the second similarity threshold, which reflects the strong semantic association between entities, which usually means potential common influence on health information or risk factors. By modeling the interaction terms of the screened significantly related entity pairs, more multidimensional text interaction features are generated. This embodiment can not only capture the complex relationships between health description entities, but also provide richer features for pension risk assessment.

[0064] The training steps of the home-based elderly care evaluation model in this embodiment include: S6-1, extract the first batch of home risk assessment training samples from the home risk assessment training set as input features; S6-2, performing forward propagation on the first batch of home risk assessment training samples, and outputting the first batch of pension risk level prediction values; S6-3, calculating the prediction error between the predicted value of the pension risk level in the first batch and the actual pension risk level; S6-4, if the prediction error does not reach the minimum error, updating the model parameters of the general model until the minimum error is reached; S6-5. If the prediction error reaches the minimum error, the model parameters of the current batch are regarded as the model parameters of the home-based elderly care evaluation model; S6-6. Export the model parameters of the home-based elderly care evaluation model, and use them to construct the home-based elderly care evaluation model.

[0065] This embodiment gradually optimizes the parameters of the general model through an iterative supervised learning process, and finally generates a home-based elderly care assessment model. Specifically, the system extracts training samples from the home-based risk assessment training set, performs forward propagation to predict the elderly care risk level, and continuously adjusts the model parameters by calculating the error between the predicted value and the actual value. Through this iterative optimization process, the prediction accuracy of the model is gradually improved until the minimum error standard is reached, ensuring that the final generated home-based elderly care assessment model has high accuracy and reliability.

[0066] Embodiment 2: The embodiment of the present invention also provides an AI intelligent interactive system based on digital home-based elderly care services, which is used to implement the above method embodiments, which have been described and will not be repeated here. The terms "module", "unit", "sub-unit", etc. used below can be a combination of software and / or hardware that implements predetermined functions. Although the systems described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0067] like Figure 4 As shown, Figure 4 This is a structural block diagram of an AI intelligent interactive system based on digital home-based elderly care services in the present invention, which includes: The elderly care monitoring report acquisition module is used to obtain the home elderly care monitoring report of the elderly living at home; A health description vector extraction module, used to extract the health description vector of the current elderly person at home from the home-based elderly care monitoring report; The elderly care risk level output module is used to input the health description vector into a preset home-based elderly care assessment model and output the elderly care risk level of the current elderly people at home.

[0068] The pension strategy retrieval module is used to retrieve the predefined pension strategy corresponding to the risk level based on the pension risk level of the elderly living at home and interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

[0069] In the above system, the home-based elderly care monitoring report acquisition module is used to obtain the current home-based elderly care monitoring report; the health description vector of the current home-based elderly care is extracted through the health description vector extraction module; the current home-based elderly care risk level is obtained through the elderly care risk level output module; the predefined elderly care strategy is obtained through the elderly care strategy retrieval module; and the problem of how to capture entity interactions in the home-based elderly care monitoring text is solved.

[0070] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means.

[0071] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., DVD ), or semiconductor media. The semiconductor media may be a solid state drive.

[0072] In the several embodiments provided in the present application, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the units is only a logical function division of a waterway underwater terrain change analysis system and method. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the system or unit can be electrical, mechanical or other forms.

[0073] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. An AI intelligent interaction method based on digital home-based elderly care services, characterized in that: include: Obtain the current home-based elderly care monitoring report; Extracting the health description vector of the elderly living at home from the home-based elderly care monitoring report; Input the health description vector into a preset home-based elderly care assessment model to output the elderly care risk level of the current elderly at home; Based on the current elderly care risk level of the elderly living at home, the predefined elderly care strategy corresponding to the risk level is retrieved to interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

2. According to claim 1, an AI intelligent interaction method based on digital home-based elderly care services is characterized in that: The modeling method of the home-based elderly care evaluation model includes: S1. Obtain a home-based elderly care monitoring sample set; wherein the sample set includes: a number of home-based elderly care monitoring samples marked with home-based elderly care risk levels; S2. Performing text recognition on the home-based elderly care monitoring samples to extract text interaction features; S3, flattening the text interaction features to generate a health description vector of a one-dimensional array; S4, matching the health description vector with the elderly care risk level to generate home risk assessment training samples; S5. Obtain a number of home risk assessment training samples and aggregate them into a home risk assessment training set; S6. Use the general model to receive the home risk assessment training set as input, and perform iterative supervised learning until the general model is iterated into the home-based elderly care assessment model.

3. According to claim 2, an AI intelligent interaction method based on digital home-based elderly care services is characterized in that: Obtain home-based elderly care monitoring sample set, including: S1-1. Import the electronic health reports of elderly people living at home from the electronic health record system; S1-2, extracting a number of physical signs to be monitored from the electronic health report; S1-3, determine the pension risk level of the electronic health report based on a number of monitored physical signs and indicators and predefined physical sign abnormality standards; The determination expression of the pension risk level is: ； Among them, RS represents the pension risk level score, represents the i-th physical sign index to be detected, represents the risk weight of the i-th physical sign indicator to be detected, and n represents the number of physical sign indicators to be monitored; S1-4, filling a number of physical signs to be monitored and the corresponding elderly care risk levels into a structured home-based elderly care monitoring template to obtain the home-based elderly care monitoring sample; S1-5. Obtain a number of home-based elderly care monitoring samples, and aggregate them into the home-based elderly care monitoring sample set.

4. According to claim 2, an AI intelligent interaction method based on digital home-based elderly care services is characterized in that: Perform text recognition on the home-based elderly care monitoring samples and extract their text interaction features, including: S2-1, extracting the health description entity set of the home-based elderly care monitoring sample; S2-2. Modeling the interaction item of any two health description entities in the health description entity set to generate the text interaction feature.

5. The AI ​​intelligent interaction method based on digital home-based elderly care service according to claim 4 is characterized in that: Extract the health description entity set of the home-based elderly care monitoring sample, including: S2-1-1. Select health description text from the home-based elderly care monitoring sample; S2-1-2, importing the health description text into the SC toolkit for entity recognition, and filtering out the original triple entity set of the health description text; S2-1-3. Semantically standardize the original triple entities in the original triple entity set to generate standard triple entities, and define them as health description entities of the home-based elderly care monitoring sample.

6. The AI ​​intelligent interaction method based on digital home-based elderly care service according to claim 5 is characterized in that: Select health description text from the home-based elderly care monitoring sample, including: S2-1-2-1, mapping the health description text to a multidimensional vector space through word embedding to generate a word vector containing context features; S2-1-2-2, using the sequence labeling module of the SC toolkit to analyze the word vector containing the context features word by word to identify the entity name; S2-1-2-3. Based on the label classification rules preset by the SC toolkit, the entity name is assigned a category to generate an entity category; S2-1-2-4, recording the starting position and the ending position of the entity category in the health description text, and defining the enclosed area of ​​the starting position and the ending position as the entity boundary; S2-1-2-5, filter out the entity name, entity category and entity boundary, and define their combination as the original triple entity of the health description text; wherein the triple of the original triple entity is: entity name, entity category and entity boundary; S2-1-2-6. Filter out all original triple entities from the health description text, and define their combination as the original triple entity set.

7. The AI ​​intelligent interaction method based on digital home-based elderly care service according to claim 3 is characterized in that: The steps for defining the health description entity of the home-based elderly care monitoring sample include: S2-1-3-1. Based on any two entity names in the original triple entity set, calculate the word vector similarity between any two original triple entities; The calculation expression of the word vector similarity between any two original triple entities is: ； Among them, CS represents the word vector similarity, Represent the word vectors of any two original triple entities, represents the dot product of the word vectors of any two original triple entities, and Represent word vectors respectively The modulus and word vector The module length; S2-1-3-2, judging whether the word vector similarity between any two original triple entities exceeds a first similarity threshold; S2-1-3-3, if the word vector similarity exceeds the first similarity threshold, define the entity name exceeding the first similarity threshold as a synonym variant name; S2-1-3-4. Replace the synonymous variant names with standard expression names so that any two original triple entities whose word vector similarity exceeds the first similarity threshold are converted into the same standard triple entity.

8. The AI ​​intelligent interaction method based on digital home-based elderly care service according to claim 5 is characterized in that: Modeling the interaction item of any two health description entities in the health description entity set to generate the text interaction feature includes: S2-2-1, associate the word vector similarity between any two original triple entities to the health description entity set; S2-2-2, judging whether the word vector similarity between any two health description entities exceeds a second similarity threshold; S2-2-3. If the word vector similarity exceeds the second similarity threshold, select health description entity pairs with significant semantic relevance in the health description entity set; S2-2-4. Model the interaction items of the health description entity pairs to generate the text interaction features.

9. The AI ​​intelligent interaction method based on digital home-based elderly care service according to claim 3 is characterized in that: Using a general model to receive the home risk assessment training set as input, and performing iterative supervised learning until the general model is iterated into the home-based elderly care assessment model, including: S6-1, extract the first batch of home risk assessment training samples from the home risk assessment training set as input features; S6-2, performing forward propagation on the first batch of home risk assessment training samples, and outputting the first batch of pension risk level prediction values; S6-3, calculating the prediction error between the predicted value of the pension risk level in the first batch and the actual pension risk level; S6-4, if the prediction error does not reach the minimum error, updating the model parameters of the general model until the minimum error is reached; S6-5. If the prediction error reaches the minimum error, the model parameters of the current batch are regarded as the model parameters of the home-based elderly care evaluation model; S6-6. Export the model parameters of the home-based elderly care evaluation model, and use them to construct the home-based elderly care evaluation model.

10. An AI intelligent interactive system based on digital home-based elderly care services, executing the interactive method according to any one of claims 1 to 9, characterized in that: The system comprises: The elderly care monitoring report acquisition module is used to obtain the home elderly care monitoring report of the elderly living at home; A health description vector extraction module, used to extract the health description vector of the current elderly person at home from the home-based elderly care monitoring report; The elderly care risk level output module is used to input the health description vector into a preset home-based elderly care assessment model and output the elderly care risk level of the current elderly at home; The pension strategy retrieval module is used to retrieve the predefined pension strategy corresponding to the risk level based on the pension risk level of the elderly living at home and interact with the user; The health description vector is a flattened vector of text interaction features generated by interaction item modeling.

Images