Humanoid large model pregnant and lying-in woman accompanying digital human system and accompanying method thereof

Abstract

The invention discloses a humanoid large model pregnant and lying-in woman accompanying digital human system and an accompanying method thereof, and the humanoid large model pregnant and lying-in woman accompanying digital human system comprises a data collection module which is used for obtaining data information corresponding to pregnant and lying-in woman accompanying, carrying out the desensitization, deduplication and availability evaluation processing of the obtained data information, and obtaining the accompanying data of a pregnant and lying-in woman accompanying data; obtaining the processed text data; the data processing module is used for performing relation extraction, entity recognition and text classification processing on the processed text data to obtain the text data subjected to structured processing; the training module is used for inputting the text data subjected to the structured processing into the CHATGLM intelligent model, and performing fine tuning training by adopting an LORA method to obtain the CHATGLM intelligent model subjected to fine tuning; and the optimization module is used for carrying out optimization processing on the finely-adjusted CHATGLM intelligent model by adopting reinforcement learning and prompt word templates fed back by human beings to obtain a pregnant and lying-in woman accompanying anthropomorphic model.

Description

Technical Field

[0001] The present invention relates to the fields of health care and artificial intelligence technology, and in particular to a large-scale anthropomorphic model of a pregnant woman accompanying a digital human system and an accompanying method thereof. Background Art

[0002] Digital human technology is a virtual image technology based on artificial intelligence technology that can simulate human appearance, voice and behavior. The technology is based on powerful computer graphics rendering and motion capture technology, and uses big data technology and artificial intelligence technology to achieve anthropomorphic interaction with real humans.

[0003] On the one hand, with the help of machine learning algorithms and deep learning technology, AI can efficiently process and analyze massive amounts of data, extract the characteristics and laws of human behavior, and enable digital humans to simulate human thinking processes and make corresponding decisions and behaviors. At the same time, the integration of AI technologies such as speech recognition, natural language processing, and computer vision gives digital humans a more natural and smooth interaction capability, enabling them to have in-depth conversations with users and understand complex instructions.

[0004] On the other hand, with the rapid development of big data technology and artificial intelligence technology and the continuous improvement of data infrastructure, in the field of health care, digital humans can conduct in-depth mining and analysis of diversified data such as user health and behavior, and thus provide service recipients with a series of functions such as question answering, health service assistant, and data processing and analysis, meeting users' high requirements for the ability of digital humans to provide personalized services.

[0005] At present, with the rapid development of artificial intelligence technology and the great improvement of computing power, various health and medical question-and-answer systems have emerged one after another, and their service scope covers almost every link of the entire medical process, from pre-consultation, patient triage to drug indications. However, the existing various technical systems, whether pure question-and-answer models or existing digital humans, have certain defects in handling management consultation and emotional comfort for pregnant women.

[0006] The mainstream health care big models on the market include iFlytek's iFlytek Spark Medical Big Model, Baidu Lingyi Big Model, Tencent Hunyuan Medical Big Model, SenseTime Big Medical Big Model, etc. These big models are generally presented in the form of question-and-answer boxes or webpage systems. Their main tasks are to complete the generation of structured medical records, auxiliary diagnosis of various types of images and medical test reports, health management consulting services for healthy and sub-healthy people, and digital intelligent management services for hospitals. These big models can handle non-emotional business interaction services and simple question-and-answer consulting services well, but basically do not have the emotional care and detailed guidance functions for pregnant women in multiple environments and situations.

[0007] In terms of data people, one of the more successful ones is the data health person "Anzhener" jointly developed by the Zhejiang Provincial Health Commission and Ant Group. It has a simple voice dialogue function, and can provide procedural functions such as intelligent guidance, online number collection, queuing, report inquiry, and medical insurance payment. However, it cannot provide the prenatal care, answer questions and other functions required by pregnant women, and does not have the ability to provide personalized services to pregnant women.

[0008] In summary, there is currently no digital human that can provide advice, services, education, and management specifically for pregnant women. Although the technical basis of existing digital human-related products is the same, they lack personalized training for the condition and environment of pregnant women.

[0009] In view of the above technical problems, the present invention proposes a large-scale anthropomorphic model of a pregnant woman accompanying digital human system and accompanying method thereof. Summary of the invention

[0010] The purpose of the present invention is to provide a large-scale anthropomorphic model of a pregnant woman accompanying digital human system and accompanying method thereof in view of the defects of the prior art.

[0011] In order to achieve the above purpose, the present invention adopts the following technical solutions: A large-scale anthropomorphic model of a pregnant woman accompanying a digital human system, comprising: The data collection module is used to obtain data information corresponding to the maternal companionship, and to perform desensitization, deduplication, and usability evaluation on the obtained data information to obtain processed text data; The data processing module is used to perform relationship extraction, entity recognition, and text classification on the processed text data to obtain structured text data; The training module is used to input the structured text data into the CHATGLM intelligent model and use the LORA method to perform fine-tuning training to obtain the fine-tuned CHATGLM intelligent model; The optimization module is used to optimize the fine-tuned CHATGLM intelligent model using reinforcement learning of human feedback and prompt word templates to obtain an anthropomorphic model of pregnant women accompanying their mothers.

[0012] Furthermore, the data collection module includes: A text processing module, used for obtaining text data in data information; A desensitization processing module is used to delete sensitive data in text data to obtain desensitized text data; A usability evaluation module is used to evaluate the usability of keywords corresponding to the desensitized text data based on the bag-of-words model; The deduplication processing module is used to check the evaluated keywords for duplication, delete the duplicate keywords, and obtain the deduplicated text data.

[0013] Furthermore, the usability evaluation module evaluates the usability of the desensitized keywords by determining whether the keywords corresponding to the text blocks are normal, and removing the text blocks whose keywords are abnormal and exceed the preset threshold range, which is expressed as: ; Among them, K represents the keyword outlier value of each data; W represents the number of keywords; w represents the keyword sequence number; Un w represents the vector cosine distance of the wth keyword in the nth data; T represents the preset threshold interval; Indicates the number of keywords that are not within the preset threshold range.

[0014] Furthermore, the deduplication processing module checks the evaluated keywords for duplication and deletes duplicate keywords, which is expressed as: ; Among them, Un represents the nth data; Un w =Un-1 w If the nth data and the n-1th data are exactly the same, the nth data will be deleted.

[0015] Furthermore, the data processing module specifically includes: The relation extraction module is used to extract relations from text data using the extraction model to complete the task of extracting data relations; The entity recognition module is used to use the vector model to identify the semantic information of the text data after the data relationship extraction is completed; The text classification module is used to store the recognized text data into a relational database, a graph database, and a vector database according to the text category and data format.

[0016] Furthermore, the CHATGLM intelligent model in the training module is the ChatGLM3-7B intelligent model.

[0017] Furthermore, the optimization module optimizes the fine-tuned CHATGLM intelligent model based on human feedback reinforcement learning as follows: The decision-making process of the fine-tuned CHATGLM intelligent model is optimized through reinforcement learning RLHF with human feedback, so that the fine-tuned CHATGLM intelligent model can self-iterate and learn according to the feedback.

[0018] Furthermore, the optimization module optimizes the fine-tuned CHATGLM intelligent model based on the prompt word template by setting a reply template for the fine-tuned CHATGLM intelligent model through preset rules and constraints.

[0019] Furthermore, it also includes a generation module for generating pregnant woman portraits and maternal and infant health data visualization management based on the obtained structured text data.

[0020] Correspondingly, a method for accompanying a digital human system for accompanying a pregnant woman with a large anthropomorphic model is also provided, which is implemented by an anthropomorphic large anthropomorphic model digital human system for accompanying a pregnant woman with a large anthropomorphic model. The method comprises: S1. The pregnant woman companion anthropomorphic model obtains questions raised by users, and performs text analysis on the obtained questions to obtain keywords corresponding to the text; S2. The pregnant woman companion anthropomorphic model determines whether the obtained keywords are compliant. If not, the user is reminded and the conversation ends without recording relevant information. If yes, whether there is relevant knowledge corresponding to the keyword search keyword in the keyword search database is determined. If yes, step S3 is executed. S3. The pregnant woman's companion anthropomorphic model forms a specific answer based on the prompt template; S4. The anthropomorphic model accompanying pregnant women recommends relevant popular science content based on the answers and records the conversation text.

[0021] Compared with the prior art, after completing the construction of the base layer, the medical middle platform model, and the maternal companion anthropomorphic model, the present invention has four major capability modules: a multimodal question answering system, a maternal health service assistant, a data collection and analysis platform, and a user data cockpit. Through the four major capability modules, six major functions are realized: real-time Q&A for pregnant women, health advice during pregnancy, popular science of health knowledge during pregnancy, joint processing of maternal medical data across departments, generation of maternal health portraits, and visualization management of maternal health data. The specific effects are summarized as follows: 1. The present invention realizes comprehensive processing and analysis of maternal data by constructing a three-level large-scale joint processing system of base, middle platform and front end, and enables the system to have the ability to conduct real-time multi-round multi-modal anthropomorphic question and answer with pregnant women through deep learning methods. By interacting with pregnant women in the form of digital humans, various questions of pregnant women can be effectively solved, and their mental anxiety can be alleviated.

[0022] 2. Through reinforcement learning and knowledge graphs, the system can provide pregnant women with pregnancy health advice (including but not limited to recommended daily energy intake, recommended ingredients, important nutrient reminders, etc.) in an anthropomorphic way after obtaining the current health data of pregnant women and conducting multiple rounds of dialogue with them. It can also publicize and popularize relevant maternal and child health knowledge based on the questions raised by pregnant women and the relevant health and medical texts or images uploaded by them.

[0023] 3. The system uses a powerful medical middle-end model to break through information barriers, obtain the medical data and test data of various departments of pregnant women, and jointly process relevant health and medical data through a data relationship extraction model. It uses knowledge graph technology to describe the current status of pregnant women in text form, and uses relevant health and medical data to create prenatal portraits of pregnant women, so that pregnant women can understand their specific conditions and make choices suitable for themselves and the fetus in terms of medical treatment, nutrition arrangements, and rest time arrangements.

[0024] 4. The system collects and analyzes data to build a user data cockpit, which facilitates relevant medical staff, pregnant women and their families to have an accurate and clear understanding of the historical data of the entire pregnancy process. On the one hand, it reduces the workload of the doctors in charge of pregnant women, quickly collects and analyzes relevant data of pregnant women, and assists them in making better medical decisions. On the other hand, it helps pregnant women and their families understand the entire process of pregnancy and childbirth, which helps them avoid risks and relieve anxiety. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a structural diagram of a large anthropomorphic model of a pregnant woman accompanying a digital human system provided in the first embodiment; Figure 2 This is a structural diagram of an anthropomorphic model of a pregnant woman accompanying a lying-in woman provided in the first embodiment; Figure 3 is a schematic diagram of structural processing of text data provided in Example 1; Figure 4 This is a flow chart of a method for accompanying a pregnant woman by a large anthropomorphic model digital human system provided in the second embodiment; Figure 5 This is a schematic diagram of the generation of pregnant woman portraits and visualization management of maternal and infant health data provided in Example 6. DETAILED DESCRIPTION

[0026] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0027] The purpose of the present invention is to provide a large-scale anthropomorphic model of a pregnant woman accompanying digital human system and accompanying method thereof in view of the defects of the prior art.

[0028] Embodiment 1

[0029] This embodiment provides a large-scale anthropomorphic model of a pregnant woman accompanying a digital human system. Figure 1 As shown, including: The data collection module 11 is used to obtain data information corresponding to the accompanying of pregnant women and perform desensitization, deduplication, and usability evaluation on the obtained data information to obtain processed text data; The data processing module 12 is used to perform relationship extraction, entity recognition, and text classification on the processed text data to obtain structured text data; The training module 13 is used to input the text data after structured processing into the CHATGLM intelligent model, and use the LORA method to perform fine-tuning training to obtain the fine-tuned CHATGLM intelligent model; The optimization module 14 is used to optimize the fine-tuned CHATGLM intelligent model by using reinforcement learning of human feedback and prompt word templates to obtain an anthropomorphic model of pregnant women accompanying their mothers.

[0030] This system is mainly composed of three layers: technical base layer, middle-stage task adjustment layer, and front-end dialogue application layer. With the support of the technical base layer and the middle-stage task adjustment layer, the front-end application dialogue layer uses a combination of ChatGLM3-7B base large model, GPLinker model, word2vec model, bge-large-zh-v1.5 model, and approximate nearest neighbor HNSW (Hierarchical Navigable Small Word) graph algorithm and other technologies to achieve the construction of knowledge graphs for maternal and child-related issues, vector database construction, various entity extraction problems, and joint analysis of data from various departments of the hospital, thereby achieving accurate answers to real-time questions, precise display of user portraits, and detailed and reliable data cockpits, ensuring smooth and effective communication between pregnant women and the system and effectively alleviating anxiety.

[0031] like Figure 2 As shown, the technical base layer of this embodiment is composed of a data base and an intelligent model base. Millions of obstetric and gynecological health medical data and the CHATGLM large model together form the algorithm and data foundation for training the digital human accompanying pregnant women (anthropomorphic large model). After the medical middle platform model completes the data induction work such as health medical data entity recognition, relationship extraction, and text classification for different tasks, it jointly supports the medical dialogue large model to complete multiple rounds of dialogue interaction tasks and aligns with the requirements of human communication feedback.

[0032] In the data collection module 11, data information corresponding to the accompanying of pregnant women and mothers is obtained, and the obtained data information is desensitized, deduplicated, and usability evaluated to obtain processed text data.

[0033] The system's data base has millions of high-quality training data, including: basic user information, user medical process information (appointment registration, online follow-up, consultation session, evaluation information, etc.), medical intervention data (prescription, test inspection, home care, chronic disease management, etc.), academic research clinical data, pathological analysis report data, drug description data, pregnant women's popular science literature data, obstetrics and gynecology and pediatrics Q&A data. All training sets have more than 1,000,000 data samples, covering various modal data of different age groups in obstetrics and gynecology and other related departments in the past five years, including text data such as diagnosis texts and test reports, and image data such as CT and B-ultrasound tests.

[0034] The data collection module 11 specifically includes: a text processing module, a desensitization processing module, a usability evaluation module, and a deduplication processing module.

[0035] The text processing module is used to obtain text data in data information. In text data processing, this embodiment first uses OCR, document parser and other technologies to obtain text data in different file formats (txt / csv / png, etc.) of the source data set, and marks the text type (disease data, drug data, report data, etc.). Then, the long document is segmented according to text length, punctuation, title, etc., and converted into shorter text blocks. After that, the system desensitizes, removes duplication, and evaluates usability of the text data, retaining high-quality data for model training.

[0036] The desensitization processing module is used to delete sensitive data in text data and obtain desensitized text data. In order to protect privacy, after obtaining relevant text data, the system automatically deletes key privacy fields such as name, ID number, hospital HIS system number, etc., and uses digital codes to encode relevant text data, eliminating privacy leakage risks for subsequent processing.

[0037] In this embodiment, the source data set will contain patient privacy data, and the data in each field of the data set will be stored in text mode. However, each piece of data has corresponding patient identification information, namely the HIS code or ID number. Therefore, it is necessary to replace these privacy-related numbers with simple numerical numbers for training the model.

[0038] The availability evaluation module is used to evaluate the availability of keywords corresponding to the desensitized text data based on the bag-of-words model, determine whether the keywords corresponding to each piece of data are normal, and remove the data whose keyword anomalies exceed the preset threshold interval to obtain the evaluated keywords. In the availability evaluation, since the relevant text data needs to be used in the health and medical field, the accuracy requirements of the data are very high, and the text data needs to consider a series of factors such as consistency, accuracy, and completeness. Therefore, this embodiment proposes a method for determining the quality of the element word vector data of thirty dimensions of each keyword in the data set based on the bag-of-words model to evaluate the availability of text data. The specific method is as follows: In the online follow-up data set for pregnant women, we record the text records of the conversations between pregnant women and doctors, obtain the documents through the document parser and complete the long text word segmentation, obtain multiple text data, and obtain a series of keywords for each text data. Then, we vectorize all the keywords based on the word2vec model, and represent each keyword as a 30-dimensional vector Vw = [V1, V2, ..., V 30 ], where V i (i=1,2,...,30) represents the i-th dimension of Vw, w is the keyword sequence number in the data set, that is, it represents the w-th keyword in this data set.

[0039] For each keyword, calculate the geometric mean word vector G of each keyword's 30-dimensional vector M , G M It can be expressed as: G M = (V1 ⨀ V2 ⨀ ... ⨀ V 30 ) (1 / 30) ; Here, ⨀ represents the element-wise multiplication of the vectors (Hadamard product), and 1 / 30 represents the 30th root operation of the vector (element-by-element).

[0040] The geometric mean word vector G of the completed dataset M After calculation, we need to compare the vector Vw with the geometric mean word vector G M The deviation between them is calculated. At this time, the special case vector θ of pregnant women needs to be introduced. The value of θ is determined by the special case of pregnant women. The values ​​of the special case vector θ taken for different special cases of pregnant women are different, and then different θ values ​​need to be multiplied when calculating the cosine distance.

[0041] The calculation method of θ is as follows: First, calculate the geometric mean vector G0 of the data set of 1,000 pregnant women aged 25 to 30 without physical abnormalities M , and then select the pregnant women with the largest deviation from the average value to calculate the cosine distance threshold T, which is expressed as: T=d cos(Vw, G0 M ) = 1 - (Vw · G0 M) / (|Vw| · |G0 M |); Among them, ⋅ represents the dot product of the vector, |Vw| and |G0 M ∣represent the modulus of the vector respectively.

[0042] After the calculation of the cosine threshold T is completed, the maternal special situation vector θ is further set according to the special situations of different pregnant women. According to the data of twin pregnant women, diabetic pregnant women and other data in the hospital's historical data, the geometric mean vector GS of the existing abnormal pregnant women data set is obtained. M , where S is the exception number, such as G1 M For pregnant women with twins, G2 M For pregnant women with diabetes, G3 M For elderly pregnant women, etc., according to different abnormal situations, the cosine distance value QS is calculated separately, and then the special situation vector θS of pregnant women can be calculated by the ratio of Q to T: θS=QS / T; Where θS is the vector number of different maternal special situations, for example, θ1 is the situation vector of twins, and θ2 is the situation vector of diabetic maternal situations. QS is the cosine distance value of the current special situation, and T is the cosine distance threshold.

[0043] After completing the geometric mean word vector G M After calculating the maternal special situation vector θS, the quality of each piece of data in the data set can be checked. After the system obtains the data, it calculates the geometric mean word vector of each keyword vector Vw for each piece of data, and then adds the maternal special situation vector θS to calculate the vector cosine distance according to the maternal special situation label: Un w =θS·d cos(Vw, G0 M )=θS·[1 - (Vw · G0 M ) / (|Vw| · |G0 M |)]; Finally, the keyword outliers of each data are counted and expressed as: ;

[0044] Among them, K represents the keyword outlier value of each data; W represents the number of keywords; w represents the keyword sequence number; Un w represents the vector cosine distance of the wth keyword in the nth data; T represents the preset threshold interval; Un w >T means judging whether the cosine distance falls within the threshold interval; The number of keywords that are not within the preset threshold range indicates that there may be problems with the data. When K is greater than 1 / 3, the data is considered to have problems and is removed from the training data set and does not participate in subsequent model training.

[0045] The deduplication processing module is used to check the evaluated keywords for duplicates, delete duplicate keywords, and obtain deduplicated text data. After completing the usability evaluation of the text data, check the data for duplicates. If there are two entries with completely duplicated keyword vector cosine distances, the duplicate entry with the later data sequence number needs to be deleted. That is, the formula is satisfied: ; Among them, Un represents the nth data; Un w =Un-1 w It means that the nth data and the n-1th data are exactly the same, so the nth data is deleted. After completing the text data desensitization, text data availability check and text data duplication check, the text data part processing is completed.

[0046] In this embodiment, the processing of image data such as CT and B-ultrasound examination is also included. The image processing improves the image quality by contrast adjustment, sharpening and geometric correction, so as to ensure the accuracy of subsequent model training. Among them, contrast adjustment is used to improve the overall and local contrast of the image, making the image clearer, facilitating the capture of key information and the calculation of feature vectors. Sharpening makes the edge details of the image more prominent, preventing the key edge information of various inspection images from being missed in model training. Geometric correction ensures that the images input for training are at a relatively regular angle, ensuring the accuracy of recognition in the initial training stage of the model. Since the production cost of various medical examination image data is relatively high and the possibility of error is very low, the image correction and enhancement are mainly used in the image data processing stage, and the image database is not deleted or compressed.

[0047] In the data processing module 12, the processed text data is subjected to relationship extraction, entity recognition, and text classification processing to obtain structured text data.

[0048] like Figure 3 As shown, in order to improve the training accuracy and computing efficiency of the large anthropomorphic dialogue model, this embodiment adds a medical data middle platform model between the original data layer and the model application layer to complete entity recognition, relationship extraction, text classification and other tasks for various health and medical data for different tasks.

[0049] The relation extraction module is used to extract relations from text data using the extraction model to complete the task of extracting data relations.

[0050] After desensitizing, deduplication, and usability evaluation in the data base, the extraction model GPLinker is used to extract relations from long text data. By accurately labeling health and medical text data and then fine-tuning it under supervision, the GPLinker model uses GlobalPointer-based technology to treat relation extraction as a five-tuple recognition problem, namely the first position of the subject, the tail position of the subject, the relationship between the predicate subject and the object, the first position of the object, and the tail position of the object. The five-tuple is identified and matched through five steps: preliminary recognition, entity position determination, relationship matching, multi-stage condition judgment, and global optimization to complete the task of extracting data relations.

[0051] The entity recognition module is used to use a vector model to identify semantic information from text data that has completed data relationship extraction.

[0052] After extracting the entity relationships of each phrase in the text, you need to use the vector model bge-large-zh-v1.5 to obtain the vector representation of the text block corresponding to each piece of data. By inputting the text block with completed relationship extraction into the model, the model calculates a high-dimensional vector about the characteristics of the text block through triple-based vector operations, realizing the representation of the text block in the vector space, and then completing the capture of the semantic information of the text block.

[0053] The text classification module is used to store the recognized text data into a relational database, a graph database, and a vector database according to the text category and data format.

[0054] The data related to the entity relationship extraction-quintuple construction and high-dimensional vector calculation processing steps are stored in the relational database (table data), graph database (triplet data), and vector database (vector data) according to the text category and data format. And through the approximate nearest neighbor HNSW (Hierarchical Navigable Small Word) graph algorithm, the data in the relational database, graph database, and vector database are linked with a shorter path, and then different databases are called according to different data types to provide fast retrieval support for the subsequent training of the anthropomorphic medical dialogue model and real-time question and answer for pregnant women. The relational database, graph database, and vector database jointly form a keyword retrieval database. Among them, the relational database refers to the storage of the associated information data of a certain code patient (A-45 years old-native place); the graph database stores triple data, that is, storage (A-school-student); the vector database stores the feature vector obtained after the text block is recognized and calculated.

[0055] In the training module 13, the text data after structured processing is input into the CHATGLM intelligent model, and the LORA method is used for fine-tuning training to obtain the fine-tuned CHATGLM intelligent model.

[0056] This embodiment uses the ChatGLM3-7B large model released by Zhipu AI as the model base for supervised fine-tuning. ChatGLM3-7B has powerful multimodal understanding capabilities and can easily cope with various types of information processing needs such as text and images. It also integrates AgentTuning technology to realize the functions of code generation and execution, bringing new intelligent support to the fields of programming and software development. In addition, the model is based on the latest efficient dynamic reasoning and video memory optimization technology, achieving faster reasoning speed and lower reasoning cost under the same hardware and model conditions, greatly improving application efficiency and cost performance. At the same time, the ChatGLM3-7B large model natively supports complex scenarios such as tool calls, database operations, knowledge graph search and reasoning, and operating systems. It has good multi-platform applicability and is suitable for use as an anthropomorphic large model base for accompanying pregnant women.

[0057] On the basis of the original model of ChatGLM3-7B, the CHATGLM intelligent model is fine-tuned by the LORA method. The LORA method fine-tunes the CHATGLM intelligent model in two parts: the first part is to obtain a labeled data set for fine-tuning training; the second part is to select a suitable low-rank matrix. After completing these two parts, CHATGLM is selected as the preloaded model, and the selected low-rank matrix parameters and the pregnant women conversation data set are set for training. This can achieve the competence of the large model for new tasks with fewer new parameters, effectively saving computing power costs and improving training efficiency.

[0058] 1. Obtain an annotated data set for fine-tuning training: This problem is solved by combining the GlobalPointer technology with various clinical dialogue source data collected. After obtaining a large amount of source data on clinical conversations between pregnant women and medical staff, the GlobalPointer technology is used to clean and organize the dialogue source data. The input text is first converted into a token sequence by a tokenizer (Token sequence refers to splitting the original text into independent units (tokens), which can be words, subwords or characters), and then converted into a vector representation through an embedding layer, thereby realizing the extraction of feature vectors of relevant keywords and various emotional terms, generating corresponding emotional labels and a complete label matrix, and generating a complete processing. The pregnant women's dialogue data set with emotional labels realizes the data processing of long texts generated in the conversations between pregnant women and medical staff; see the processing methods of data collection module 11 and data processing module 12 for details.

[0059] 2. Select a suitable low-rank matrix: The selection of the low-rank matrix R is related to the training accuracy of the CHATGLM intelligent model. The larger the rank R of the matrix, the higher the training accuracy of the CHATGLM intelligent model, but the higher the computing power requirement; the smaller the rank R of the matrix, the lower the training accuracy of the CHATGLM intelligent model, but the lower the computing power requirement. In this embodiment, in order to improve the training accuracy of the fine-tuned model as much as possible, the strategy of increasing the low-rank matrix value is adopted under the condition of computing power. The selection value of the low-rank matrix starts from 8 and is 8*2 n The regular increase is that the size of the matrix value each time is twice that of the previous training. After multiple rounds of testing, this embodiment chooses to set the rank value of the low-rank matrix to 16, and improves the accuracy of the answer as much as possible when the computing power can be satisfied.

[0060] After completing the construction of the maternal and infant dialogue dataset and selecting the low-rank matrix, the relevant LORA module was loaded and trained with the dataset. Further output optimization was performed in the post-processing part of the model. Through multiple tests, the relevant output content was emotionally corrected (to ensure that the generated content was emotionally positive and warm), and natural language processing tools were used to check the grammar and spelling errors of the generated content. For some content involving medical knowledge, manual supplementation or correction was performed based on the content generated by the model to make the model more accurate. After completing the above work, the maternal-specific task fine-tuning of the CHATGLM model was completed.

[0061] For example, the beneficial effects of the fine-tuned CHATGLM intelligent model compared to the pre-fine-tuning model are as follows: Before fine-tuning: Question#What to do with morning sickness# Answer: Morning sickness is a common phenomenon in early pregnancy. About 75% of pregnant women will experience varying degrees of nausea and vomiting in the first three months of pregnancy. In most cases, this symptom will naturally subside or disappear around the fourth month of pregnancy. For morning sickness, you can try the following methods to relieve it: ① Maintain good eating habits: Eat small and frequent meals and avoid fasting. Choose light, easily digestible foods and avoid greasy, spicy and other irritating foods. Avoid eating foods with strong odors, as some odors may cause nausea. ② Proper rest: Ensure adequate sleep and reduce fatigue. Take a nap during the day, but do not stay in bed for a long time to avoid aggravating nausea. ③ Supplement vitamin B6: Vitamin B6 is believed to help relieve the symptoms of morning sickness. But before using any supplements, please consult your doctor.

[0062] If morning sickness is so severe that you can't eat, become dehydrated, or lose a lot of weight, this could be hyperemesis gravidarum, and you need to see a doctor right away. Your doctor may prescribe medication or other medical interventions. In short, when it comes to morning sickness, the most important thing is to keep communicating with your doctor to ensure that the methods you take are both safe and effective. #End of answer

[0063] After fine-tuning: Question#What to do with morning sickness# Answer #Dear expectant mother, during this special period, you may encounter some minor challenges, such as uncomfortable morning sickness. Please remember that you are not alone in your struggle, many mothers have gone through this stage, and it will pass quickly. Here are some suggestions that I hope will help you and make you feel better: Be gentle with your body: Try to eat smaller, more frequent meals, eat light, easily digestible foods, and avoid foods that may make you feel uncomfortable. Remember, every bite counts for a better tomorrow for you and your baby.

[0064] Give yourself enough rest: When your body tells you that you need to rest, please do not hesitate, lie down, close your eyes, and enjoy this moment of peace. Proper rest can not only help you restore your physical strength, but also make you feel happier.

[0065] A little vitamin B6: Sometimes a little help can make a big difference. You may consider taking a vitamin B6 supplement, which can be very helpful for relieving morning sickness. However, remember to ask your doctor first to make sure it is safe for you to do so.

[0066] Let fresh air embrace you: Open the windows and let the fresh air in! Take a few deep breaths and feel the power of nature, which may help you relieve that uncomfortable feeling.

[0067] Keep a peaceful mind: Try to relax yourself and do something you like, such as listening to music, reading or taking a walk. Sometimes, the quality of our mood directly affects our physical feelings.

[0068] Dear friends, if you feel that morning sickness is really serious, don't force yourself. It is very important to seek medical help in time. Please remember that the health of you and your baby always comes first. Although this process is a bit hard, it is also one of the most beautiful journeys in life. Believe in yourself, you will definitely be able to get through this time and welcome a healthy baby! #AnswerEnd According to the above content, through different answers, it can be seen that the fine-tuned CHATGLM model can make more suggestions and use an anthropomorphic tone when dealing with relevant issues, so as to give more encouraging and warm suggestions, indicating that the fine-tuning effect is significant.

[0069] This embodiment uses instruction fine-tuning means to improve the accuracy of the model's response. Since the ChatGLM3-7B base model uses a large amount of unlabeled text during pre-training, and the text accuracy is not high, the text content is not focused enough on the direction of health and medical treatment, and can only achieve simple tasks such as partial text completion, which cannot be applied to pregnant women's questions and answers and maternity knowledge education tasks with very high accuracy requirements, so it is necessary to use instruction fine-tuning means to enable the model to have the above capabilities. This embodiment uses the LLaMA Factory architecture in combination with the large amount of text data and image data completed in the data collection module 11 and the data processing module 12, and the maternal and child medical care data, prescription opening data and inspection data corresponding to the labeling, to achieve personalized instruction fine-tuning of the ChatGLM3-7B base model for pregnant women, that is, ChatGLM3-7B generates answers by itself, and tells ChatGLM3-7B to regenerate training after manually judging whether it is right or wrong, until ChatGLM3-7B answers are always correct. By fine-tuning the instructions, the model can better understand and follow the instructions and commands given by humans, so that the responses and push notifications will be more accurate during conversations with pregnant women and in the process of educating pregnant women and their families about pregnancy and childbirth knowledge.

[0070] The ChatGLM3-7B of this embodiment can also be connected to existing knowledge graphs or existing Internet of Things data.

[0071] In the optimization module 14, the fine-tuned CHATGLM intelligent model is optimized by using reinforcement learning based on human feedback and prompt word templates to obtain an anthropomorphic model of pregnant women accompanying their mothers.

[0072] This embodiment optimizes the decision-making process of the fine-tuned CHATGLM intelligent model through reinforcement learning RLHF with human feedback, so that the fine-tuned CHATGLM intelligent model can self-iterate and learn according to the feedback.

[0073] RLHF, or reinforcement learning based on human feedback, is a technical method that combines reinforcement learning with human inspection feedback. In this embodiment, the RLHF framework is combined with the intelligent base. Under the pre-trained ChatGLM3-7B model framework, the training model makes decisions based on the current state of pregnant women's questions and optional education content, and the decision made by the model is manually evaluated, and feedback is given to the model on whether the answer to the question is accurate or not, and whether the education recommendation is appropriate or not; the way to obtain the education is to analyze and process the keywords of the questioner and then directly retrieve them from the Internet or search for keywords in an existing database. Through the RLHF architecture, the model converts feedback into digital reward signals, and uses proximal policy optimization (PPO) (the model automatically generates the second round of questions and answers based on the evaluation given by humans in the first round of feedback, and summarizes the accuracy and evaluation of the feedback in the second round, and thereby completes another update of the model itself) to enable the model to continuously self-iterate learning. Through continuous iteration and optimization, the model can gradually improve the accuracy of answers to pregnant women's questions and obtain appropriate educational promotion content based on the analysis of textual interaction with pregnant women, effectively enhancing the sense of gain in interaction between pregnant women and the large anthropomorphic model of pregnant women's companions.

[0074] This embodiment uses the prompt word template for optimization, specifically: setting a reply template for the fine-tuned CHATGLM intelligent model through preset rules and constraints.

[0075] Because pregnant women need to face huge mental challenges during pregnancy and childbirth, it is necessary to set prompt word templates to achieve a gentler and more humanized response and avoid various ethical and legal risks that may arise in the response. This embodiment optimizes the model's humanization level and avoids the emergence of ethical and legal issues by setting model prompt word rules and constraints. The specific prompt word format used in this embodiment is as follows: ## Background Information You are the hospital's maternity management assistant. Now please help answer the maternity-related {{user}} questions based on the {{knowledge}} content and output them in json format.

[0076] ## Specific requirements Please provide information in the following format: export interface pregnany { answer: string / / reply content similarity : string[] / / Similar questions } Answer: Please answer questions related to pregnant mothers similarity: Please output 3 similar questions, returned in array format ## Please follow the following constraints: - Make the output as warm as possible.

[0077] - Do not restate the question.

[0078] - Be empathetic and answer in a caring tone.

[0079] - When it comes to medication, be sure to state "Please take the medication only after following your doctor's orders."

[0080] - Please abide by the national conditions, laws and regulations, such as not responding to questions about surrogacy.

[0081] - If the returned format is not JSON, please optimize it to JSON format and remove unnecessary spaces.

[0082] ## Issues that are clearly prohibited by laws and regulations: - Surrogacy is not allowed - Does not support fetal gender prediction ## Example Input: What should I pay attention to when getting vaccinated? { "answer":"Notes", "similarity":[ "What are the precautions for vaccination during pregnancy?", "What should I pay attention to when getting vaccinated during pregnancy?", "What should pregnant mothers pay attention to when getting vaccinated?" ] } The above prompt words are a set of prompt words for questions asked during pregnancy. In the actual setting of this embodiment, different prompt words are set according to the different stages of the pregnant woman, according to the pregnancy period, delivery period, and infant-rearing period to ensure that the model can accurately output answers to relevant questions in an anthropomorphic tone.

[0083] After completing the construction of the three sections of fine-tuning the above instructions, reinforcement learning based on human feedback (RLHF), and setting of prompt word templates, the personalized training of the large anthropomorphic model (that is, digital human) accompanying pregnant women was realized. The system can analyze and answer questions raised by users, store data, and provide relevant popular science education content.

[0084] This embodiment performs structured processing on the input health and medical data, and inputs it into the large model for special training in the form of vectors, tables, triples, etc. By fine-tuning the single-round and multi-round questions and answers that have been labeled with supervised instructions, obstetrics and gynecology knowledge is injected into the model, thereby realizing accurate questions and answers in the field of obstetrics and gynecology. By correcting the prompt words of the large model, the system can be competent for answering anthropomorphic questions and popularizing science in a multi-round dialogue environment, and can still effectively complete tasks in special circumstances such as elderly pregnant women, pregnant women with multiple births, and pregnant women with abnormal medical examinations. The answers are accurate and effective, and ethical risks are avoided. The system realizes pregnancy services and scientific management through the organic combination of obstetrics and gynecology medical knowledge and artificial intelligence technology, and helps eugenics.

[0085] Embodiment 2 This embodiment provides a method for accompanying a pregnant woman with a human-like large model digital human system, such as Figure 4 As shown, the accompanying method is implemented by a large anthropomorphic model of a pregnant woman accompanying a digital human system in the first embodiment, and includes: S1. The pregnant woman companion anthropomorphic model obtains questions raised by users, and performs text analysis on the obtained questions to obtain keywords corresponding to the text; The user asks questions related to pregnant women to the anthropomorphic large model digital human system for accompanying pregnant women. The anthropomorphic model for accompanying pregnant women performs text parsing on the received questions based on document parser, OCR and other technologies, and extracts keywords from the text.

[0086] S2. The pregnant woman companion anthropomorphic model determines whether the obtained keywords are compliant. If not, the user is reminded and the conversation ends without recording relevant information. If so, whether there is relevant knowledge corresponding to the keywords in the keyword search database is determined. If so, step S3 is executed. The anthropomorphic model of pregnant women accompanying their babies determines whether the extracted keywords are compliant. If not, it will send a prompt to the user that there are illegal phrases in the question and provide education on the illegal parts.

[0087] If it is compliant, the keyword will be input into the keyword retrieval database and knowledge graph, and it will be determined whether there is knowledge corresponding to the keyword in the keyword retrieval database and knowledge graph. If not, a prompt message that cannot be answered accurately will be sent to the user, and the most similar answer will be given. The conversation content will then be settled as a vector and stored in the vector database, and the vector database will be updated regularly to supplement the data.

[0088] S3. The pregnant woman's companion anthropomorphic model forms a specific answer based on the prompt template; If there is knowledge corresponding to the keyword in the keyword retrieval database and the knowledge graph, the pregnant woman companion anthropomorphic model will generate answers based on the user's questions.

[0089] S4. The anthropomorphic model accompanying pregnant women recommends relevant popular science content based on the answers and records the conversation text.

[0090] The maternal companion anthropomorphic model also recommends relevant popular science knowledge, videos, etc. based on the generated answers, and can store the text data corresponding to the current round of dialogue. Finally, it ends this round of dialogue and asks whether there are other needs. If so, it continues to return to step S1, if not, it combines the dialogue.

[0091] Compared with the existing technology, this embodiment has four major capability modules after completing the construction of the base layer, medical middle platform model, and maternal companion anthropomorphic model: multimodal question answering system, maternal health service assistant, data collection and analysis platform, and user data cockpit. The four major capability modules realize the six main functions of real-time maternal questions and answers, pregnancy health advice, pregnancy health knowledge popularization, cross-department maternal medical data joint processing, maternal health portrait generation, and maternal health data visualization management.

[0092] Embodiment 3 The accompanying method of a large anthropomorphic model pregnant woman accompanied by a digital human system provided in this embodiment is different from that in the second embodiment in that this embodiment is implemented through a specific example: The pregnant woman asks the system the question #When will morning sickness end#. After receiving the question text, the system parses the text. The keywords <morning sickness><when><end> are extracted, and this group of keywords is judged to be compliant. After judgment, it is found that this group of keywords does not contain illegal words such as <surrogacy><gender identification>, so the process can continue and the database search is performed. Through the database search, it is found that the relevant keywords <morning sickness><when><end> can be found in the database and knowledge graph. Subsequently, the pregnant woman's accompanying personified model organizes answers according to the relevant information of the prompt template. First, consider the relevant constraints of the prompt template:## Please follow the following constraints: - The output content should be as warm as possible. - Do not repeat the question. - Pay attention to empathy, and answer in a caring tone. - When it comes to drugs, be sure to state "Please take the medicine after following the doctor's advice." Then form a specific answer based on the content of the question and the constraints of the prompt template: Start of answer #Dear, I understand your discomfort very well. Morning sickness, this little naughty thing, will usually quietly put away its jokes and gradually fade out of your life around the 12th to 14th week of your pregnancy journey. However, the situation of each expectant mother is unique. Just as everyone's physique and personality are different, the duration of morning sickness also varies. Some pregnant women may say goodbye to this uncomfortable time soon, while others may need a longer time, until the middle and late stages of pregnancy, or even the entire pregnancy, before morning sickness is willing to let go. Of course, hormone fluctuations, mood swings, medication use, and daily eating habits may all be important factors affecting the time when this little naughty thing leaves. But don't worry, no matter how naughty morning sickness is, as long as we maintain good living habits, actively adjust our mentality, and seek help from a doctor when necessary, we can successfully get through this special time. Remember, you are not fighting alone, your body is preparing for the arrival of a new life in its own unique way. #End of answer After the answer is given, the system recommends popular science videos and articles about morning sickness and how to relieve it, such as "Three Tips to Relieve Morning Sickness" and "How to Minimize the Negative Effects of Morning Sickness". The system also records the text of the user's questions and the system's answers in this conversation. If the user does not ask any more questions at this point, the process ends. If the user continues to ask #I am in the 20th week of pregnancy, what should I do if I still have morning sickness#, the system will give another answer after conducting a compliance check, combining the previous conversation text, user questions, and knowledge base data: Start of answer #Dear expectant mother, if morning sickness, this "long-lasting little friend", is still with you in the 20th week of your pregnancy, don't worry too much, it may just be reluctant to leave you. Although most mothers will say goodbye to it in the early stages of pregnancy, every expectant mother's pregnancy experience is unique, just like every flower has its own time to bloom. Try to treat yourself gently and give yourself more patience and care. Keep a relaxed mood, because a good mood is the best regulator for the body. At the same time, you can also refer to the popular science videos recommended to you before, adjust your diet, and let every bite of food become a sweet source of nourishment for you and your baby. If morning sickness really makes you feel uncomfortable, you might as well find time to talk to your doctor to see if there is any good way to make this "little friend" leave early and make your pregnancy more relaxed and enjoyable. Remember, you are going through a wonderful journey, and every step deserves to be treated gently. #End of answer After the answer is completed, all operations are the same as after the first round of dialogue. If there is a follow-up dialogue, the process is the same as above. If there is no follow-up dialogue, the process ends.

[0093] Embodiment 4 The accompanying method of a large anthropomorphic model pregnant woman accompanied by a digital human system provided in this embodiment is different from that in the second embodiment in that this embodiment is implemented through a specific example: If the user asks #I want to find someone to have a surrogacy, how much does it cost#, the system will parse the text after receiving the question. The key words <surrogacy><how much does it cost> are extracted. After judgment, it is found that this group of keywords contains the illegal word <surrogacy>. The system determines that there are illegal phrases in the user's question, and the process will not continue, but remind the user: Start answering#Sorry, there are some illegal words in your question, please try again# End answering After finishing the answer, the system will not record the text information of the relevant questions and answers, and will carry out education on the illegal content and push relevant warning education films.

[0094] Embodiment 5 The accompanying method of a large anthropomorphic model pregnant woman accompanied by a digital human system provided in this embodiment is different from that in the second embodiment in that this embodiment is implemented through a specific example: If the user asks #How to give birth to a hobbit that can shoot lasers#, the system will parse the text after receiving the question text. The keywords <how><give birth><shoot laser><hobbit> are extracted, and the compliance of this group of keywords is determined. After determination, it is found that this group of keywords does not contain illegal words such as <surrogacy><gender identification>, so the process can continue and database search can be performed. Through database search, it is found that the relevant keywords cannot be searched in the knowledge base for related content, but similar content <hobbit> can be found. At this time, the system will prompt the user that it cannot give an accurate answer and reply with content related to hobbits: Start of answer #I'm sorry, I'm not able to help you solve this problem for now, but I hope the following data can help you: Hobbits are... #End of answer.

[0095] After replying, the content of the question text block is solved into a vector and stored in the vector database. In the subsequent regular updates, the database is supplemented according to actual needs.

[0096] Embodiment 6 The accompanying method of the digital human system for accompanying a pregnant woman with an anthropomorphic large model provided in this embodiment is different from that in the second embodiment in that this embodiment also processes data processing modules such as pregnant woman portrait generation and visualization management of pregnant woman health data according to data information.

[0097] When user A needs to call up the health portrait of pregnant women and initiate maternal needs, the system first retrieves the relevant health and medical data of each department under user A's ID399670 (pregnant women's file data, outpatient data, various test and examination data, etc.), and then uses the pregnant women's companion anthropomorphic model to analyze the multimodal data related to user A, and obtains the relevant information of the pregnant woman <pregnancy duration 16 weeks><pregnancy age 29><jurisdiction health center: Baiyang Street Health Center><whether hospitalized: no> and special logical variable information <twin mothers: no><elderly mothers: no><diabetes Sick pregnant woman: No>, after determining the basic information, various test reports and portable data detection platform data were extracted and analyzed, and <heart rate data: 67 / min><blood pressure data: 130 / 76><weight data: 62KG><average sleep: 7.2h><blood oxygen saturation: 100%><average daily activity time: 63min><average daily stress index: 37> were obtained, and the diagnosis and treatment data of various departments were combined to obtain other disease data <respiratory system diseases: none><digestive system diseases: none><gynecological diseases: none><other unclassified diseases: none>.

[0098] After completing the data acquisition and analysis, a comprehensive portrait of the pregnant woman bound to the user A account is created, a virtual person is generated based on computer graphics design and rendering technology, and detailed factors such as body shape and expression are configured according to the relevant health conditions. The current relevant basic health data and some diet and life suggestions to ensure the subsequent health of pregnant women are displayed on the interactive page. The work of portraying the pregnant woman's current physical health status is now completed.

[0099] After viewing the digital portrait of the pregnant woman, user A visited the maternal health data cockpit. When user A visited the maternal health data cockpit, the system collected the historical medical examination data and portable device data of the pregnant woman bound to the ID, and displayed the historical status changes of the pregnant woman (pregnancy duration 0-16 weeks) and the weight of the fetus (0-110 grams) and the estimated development status (gradually showing the appearance of a baby) in a centralized image, ensuring that the visitor can see the full status data of the pregnant woman in all historical stages in one picture, achieving the function of the data cockpit. The relevant operation flow chart is as follows: Figure 5 shown.

[0100] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A large-scale anthropomorphic model of a pregnant woman accompanying a digital human system, characterized in that: include: The data collection module is used to obtain data information corresponding to the maternal companionship, and to perform desensitization, deduplication, and usability evaluation on the obtained data information to obtain processed text data; The data processing module is used to perform relationship extraction, entity recognition, and text classification on the processed text data to obtain structured text data; The training module is used to input the structured text data into the CHATGLM intelligent model and use the LORA method to perform fine-tuning training to obtain the fine-tuned CHATGLM intelligent model; The optimization module is used to optimize the fine-tuned CHATGLM intelligent model using reinforcement learning of human feedback and prompt word templates to obtain an anthropomorphic model of pregnant women accompanying their mothers.

2. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 1, characterized in that: The data collection module includes: A text processing module, used for obtaining text data in data information; A desensitization processing module is used to delete sensitive data in text data to obtain desensitized text data; A usability evaluation module is used to evaluate the usability of keywords corresponding to the desensitized text data based on the bag-of-words model; The deduplication processing module is used to check the evaluated keywords for duplication, delete the duplicate keywords, and obtain the deduplicated text data.

3. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 2, characterized in that: The usability evaluation module evaluates the usability of the desensitized keywords by determining whether the keywords corresponding to the text blocks are normal, and removing the text blocks whose keywords are abnormal and exceed the preset threshold interval, which is expressed as: ； Among them, K represents the keyword outlier value of each data; W represents the number of keywords; w represents the keyword sequence number; Un w represents the vector cosine distance of the wth keyword in the nth data; T represents the preset threshold interval; Indicates the number of keywords that are not within the preset threshold range.

4. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 3, characterized in that: The deduplication processing module checks the evaluated keywords for duplication and deletes duplicate keywords, which is expressed as: ； Among them, Un represents the nth data; Un w =Un-1 w If the nth data and the n-1th data are exactly the same, the nth data will be deleted.

5. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 1, characterized in that: The data processing module specifically includes: The relation extraction module is used to extract relations from text data using the extraction model to complete the task of extracting data relations; The entity recognition module is used to use the vector model to identify the semantic information of the text data after the data relationship extraction is completed; The text classification module is used to store the recognized text data into a relational database, a graph database, and a vector database according to the text category and data format.

6. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 1, characterized in that: The CHATGLM intelligent model in the training module is the ChatGLM3-7B intelligent model.

7. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 6, characterized in that: The optimization module optimizes the fine-tuned CHATGLM intelligent model through reinforcement learning based on human feedback as follows: The decision-making process of the fine-tuned CHATGLM intelligent model is optimized through reinforcement learning RLHF with human feedback, so that the fine-tuned CHATGLM intelligent model can self-iterate and learn according to the feedback.

8. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 6, characterized in that: The optimization module optimizes the fine-tuned CHATGLM intelligent model based on the prompt word template in the following way: setting a reply template for the fine-tuned CHATGLM intelligent model through preset rules and constraints.

9. The large-scale anthropomorphic model of a pregnant woman accompanying digital human system according to claim 1, characterized in that: It also includes a generation module for generating pregnant woman portraits and maternal and infant health data visualization management based on the structured text data obtained.

10. A method for accompanying a pregnant woman using a large anthropomorphic model digital human system, characterized in that: The method is implemented by a large-scale anthropomorphic model of a pregnant woman accompanying a digital human system as described in any one of claims 1 to 8, wherein the accompanying method comprises: S1. The pregnant woman companion anthropomorphic model obtains questions raised by users, and performs text analysis on the obtained questions to obtain keywords corresponding to the text; S2. The pregnant woman companion anthropomorphic model determines whether the obtained keywords are compliant. If not, the user is reminded and the conversation ends without recording relevant information. If yes, whether there is relevant knowledge corresponding to the keyword search keyword in the keyword search database is determined. If yes, step S3 is executed. S3. The pregnant woman's companion anthropomorphic model forms a specific answer based on the prompt template; S4. The anthropomorphic model accompanying pregnant women recommends relevant popular science content based on the answers and records the conversation text.

Images