Construction method of language model assisted child growth accompanying dialogue system

Abstract

The invention belongs to the technical field of artificial intelligence. The invention provides a construction method of a language model assisted child growth accompanying dialogue system. According to the embodiment of the invention, in combination with children cognitive development theory data, cognitive characteristic analysis is carried out for children, a children cognitive data set content framework is designed, instruction fine tuning data fused with a children cognitive development theory is constructed, and data processing and standardization are carried out on the instruction fine tuning data, so that a children growth accompanying data set is constructed. Selecting a large language model as a base model of the language model assisted child growth accompanying dialogue system, performing instruction fine tuning on the large language model based on the child growth accompanying data set by using a LoRA fine tuning method, and deploying a webpage interface for the trained large language model, the dialogue performance of a large language model in the field of child growth accompanying is improved, and meanwhile a visual dialogue interaction interface can be provided for a user.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of artificial intelligence technology, and in particular to a method for constructing a child growth companion dialogue system assisted by a language model. Background Art

[0002] In recent years, large language models have shown great potential in various fields and provided powerful language processing capabilities for a variety of application scenarios. However, most existing children's growth companion dialogue systems use predefined dialogue templates and other methods. In terms of dialogue interaction, there are problems such as limited dialogue information processing capabilities, limited accuracy of interactive content richness, and lack of scientific theoretical guidance, which leads to insufficient compliance with children's development needs. In this context, combining children's growth companion tasks with large language models can not only improve the interaction quality of the dialogue system, but also enrich children's learning experience and achieve the dual goals of education and companionship. In addition, how to optimize large language models for specific fields or tasks, let the model learn knowledge in specific fields at a lower cost and faster time, and improve the dialogue performance of the dialogue system in specific tasks is a problem to be solved.

[0003] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0004] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0005] The purpose of the embodiments of the present disclosure is to provide a method for constructing a child growth companion dialogue system assisted by a language model, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0006] According to a first aspect of an embodiment of the present disclosure, a method for constructing a child growth companion dialogue system assisted by a language model is provided, the method comprising: Based on the data of children's cognitive development theory, the content framework of children's cognitive data set was constructed; Based on the content framework of the children's cognitive data set, constructing instruction fine-tuning data using an artificial-model method and a collection-improvement method; wherein the instruction fine-tuning data includes first instruction data and second instruction data; Processing and uniformly organizing the instruction fine-tuning data to obtain a child growth companion data set; Based on the LoRA method, the selected base model is fine-tuned using the child growth companion dataset to obtain a large language model for child growth companion; A child growth accompaniment dialogue system is constructed based on the child growth accompaniment dataset and the child growth accompaniment large language model.

[0007] Furthermore, based on the children's cognitive development theory data, the steps of constructing the content framework of the children's cognitive data set include: According to the children's cognitive development theory data, the cognitive characteristics of children are analyzed to construct the content framework of the children's cognitive data set, which includes several large theme data sets, each of which has several small theme sub-data sets; wherein, The large subject data set includes at least an emotional morality data set, a social etiquette data set, a safety education data set, a cultural festival data set, a life common sense data set, a special scene data set, a common sense knowledge data set and a natural world data set; The emotional morality dataset at least includes a traditional morality sub-dataset and a core value sub-dataset; the social etiquette dataset at least includes a kinship sub-dataset, a getting along etiquette sub-dataset and a competition and cooperation sub-dataset; the safety education dataset at least includes a traffic sub-dataset, a privacy sub-dataset, a water use sub-dataset and an electronic use sub-dataset; the cultural festival dataset at least includes a Spring Festival sub-dataset, a Mid-Autumn Festival sub-dataset and a Dragon Boat Festival sub-dataset; the common sense of life dataset at least includes a physical health sub-dataset, a personal hygiene sub-dataset and a diet sub-dataset; the special scene dataset at least includes a campus sub-dataset, a hospital sub-dataset and a fire station sub-dataset; the common sense knowledge data map at least includes an ancient poetry sub-dataset, a Song Dynasty poetry sub-dataset and an idiom sub-dataset; the natural world dataset at least includes a sky sub-dataset, an ocean sub-dataset, a farm sub-dataset, a river sub-dataset and a forest sub-dataset.

[0008] Furthermore, based on the content framework of the children's cognitive data set, the step of constructing instruction fine-tuning data using the artificial-model method and the collection-improvement method includes: For other data sets other than the ancient poetry sub-dataset, the Song Dynasty poetry sub-dataset and the idiom sub-dataset in the content framework of the children's cognitive data set, construct the first instruction data using the artificial-model method; For the ancient poetry sub-dataset, the Song Dynasty poetry sub-dataset and the idiom sub-dataset in the content framework of the children's cognitive data set, construct the second instruction data using the collection-improvement method; The instruction fine-tuning data is composed according to the first instruction data and the second instruction data.

[0009] Furthermore, the instruction fine-tuning data is composed of instruction, input, output, system and history; wherein the input, the system and the history may be empty.

[0010] Furthermore, the step of processing and uniformly organizing the instruction fine-tuning data to obtain a child growth companion data set includes: Performing data cleaning on the first instruction data in the instruction fine-tuning data; wherein the data cleaning includes correcting wrong answers, separating instructions, and processing empty outputs; Integrate and uniformly process the second instruction data in the instruction fine-tuning data; wherein the integration and uniform processing includes content screening, data selection, format normalization and data conversion; The child growth companion data set is obtained according to the processed first instruction data and the processed second instruction data.

[0011] Furthermore, based on the LoRA method, the selected base model is fine-tuned using the child growth companion data set to obtain a large language model for child growth companionship, including: Add a rank decomposition matrix to each transformer block in the base model, and add a dimension reduction matrix A and a dimension expansion matrix B next to the linear layer of the base model; When the base model is trained using the child growth companion dataset, the original PLM parameters are frozen, and the parameters of the dimension reduction matrix A and the dimension expansion matrix B related to the task are updated; After the training is completed, the parameters of the dimension reduction matrix A and the dimension expansion matrix B are fused with the PLM original parameters, and the PLM original parameters of the base model are updated to obtain the large-scale language model for children's growth companionship.

[0012] Furthermore, based on the LoRA method, the step of fine-tuning the selected base model using the child growth companion dataset to obtain a large language model for child growth companionship also includes: When training the base model, the RoPE interpolation method is introduced to reparameterize the position encoding.

[0013] Furthermore, the step of constructing a child growth companion dialogue system according to the child growth companion dataset and the child growth companion large language model includes: Using the child growth companionship dataset as a dataset of the child growth companionship dialogue system; Using the large-scale language model for children's growth companionship as a back-end calling language model for the children's growth companionship dialogue system; A user interface is used as a visual interactive interface of the child growth companion dialogue system.

[0014] According to a second aspect of an embodiment of the present disclosure, a language model-assisted children's growth companion dialogue system is provided, the system comprising: The large-scale language model for children's growth companionship is used to generate dialogue information content related to children's growth companionship directly according to the input call instructions based on the knowledge and generation strategies learned during the fine-tuning process; The user interface is used to provide a visual interactive interface and customize some dialogue parameters.

[0015] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiments of the present disclosure, the method for constructing a child growth companion dialogue system assisted by the above-mentioned language model, on the one hand, combines the children's cognitive development theory data, analyzes the cognitive characteristics of children, designs the content framework of the children's cognitive data set, and constructs the instruction fine-tuning data integrating the children's cognitive development theory, and performs data processing and standardization on the instruction fine-tuning data to construct a child growth companion data set. On the other hand, a large language model is selected as the base model of the child growth companion dialogue system assisted by the language model, and the LoRA fine-tuning method is used to fine-tune the instructions of the large language model based on the child growth companion data set, and a web page interface is deployed for the trained large language model to improve the dialogue performance of the large language model in the field of child growth companionship, and at the same time, it can provide users with a visual dialogue interaction interface. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0017] Figure 1 A step diagram showing a method for constructing a language model-assisted children's growth companion dialogue system in an exemplary embodiment of the present disclosure; FIG2 is a schematic diagram showing a content framework of a children's cognitive data set in an exemplary embodiment of the present disclosure; Figure 3 A schematic diagram showing the data content format of instruction fine-tuning data in an exemplary embodiment of the present disclosure; Figure 4A schematic diagram showing the use of an artificial-model method to construct first instruction data in an exemplary embodiment of the present disclosure; Figure 5 A schematic diagram showing a LoRA method in an exemplary embodiment of the present disclosure is shown; Figure 6 A flowchart showing the construction of a child growth companion dialogue system assisted by a language model in an exemplary embodiment of the present disclosure; Figure 7 A schematic diagram showing a user interface in an exemplary embodiment of the present disclosure; Figure 8 A schematic diagram of a child growth companion dialogue system assisted by a language model in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0018] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0019] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0020] This example implementation first provides a method for constructing a child growth companion dialogue system assisted by a language model. Figure 1 As shown in , the method for constructing the language model-assisted child growth companion dialogue system may include: steps S101 to S105.

[0021] Step S101: constructing a content framework of a children's cognitive data set based on children's cognitive development theory data; Step S102: Based on the content framework of the children's cognitive data set, construct instruction fine-tuning data using an artificial-model method and a collection-improvement method; wherein the instruction fine-tuning data includes first instruction data and second instruction data; Step S103: Processing and organizing the instruction fine-tuning data to obtain a child growth companion data set; Step S104: Based on the LoRA method, the selected base model is fine-tuned using the child growth companionship dataset to obtain a large language model for child growth companionship; Step S105: construct a child growth companionship dialogue system based on the child growth companionship dataset and the child growth companionship large language model.

[0022] Through the above-mentioned method for constructing a child growth companion dialogue system assisted by a language model, on the one hand, combined with the theoretical data of children's cognitive development, the cognitive characteristics of children are analyzed, the content framework of the children's cognitive data set is designed, and the instruction fine-tuning data integrating the children's cognitive development theory is constructed, and the instruction fine-tuning data is processed and standardized to construct a child growth companion data set. On the other hand, a large language model is selected as the base model of the child growth companion dialogue system assisted by a language model, and the LoRA fine-tuning method is used to fine-tune the instructions of the large language model based on the child growth companion data set, and a web interface is deployed for the trained large language model to improve the dialogue performance of the large language model in the field of child growth companionship, and at the same time provide users with a visual dialogue interaction interface.

[0023] Next, we will refer to Figures 1 to 7 The various steps of the method for constructing the above-mentioned language model-assisted child growth companion dialogue system in this example implementation are described in more detail.

[0024] In step S101, the theory of children's cognitive development provides a theoretical basis for understanding how children construct knowledge through interaction with the environment. The theory of children's cognitive development stages emphasizes the unique cognitive abilities of children at different ages, from intuitive perception in the sensorimotor stage to abstract thinking ability in the formal operational stage. For example, the design is targeted at children aged 7-11. According to the theory of cognitive development, children aged 7-11 are in the concrete operational stage, where they begin to be able to think logically, but only for concrete or familiar things. Combined with the theory of children's cognitive development, the cognitive characteristics of children in this age group are analyzed.

[0025] Based on the analysis of cognitive characteristics of children aged 7-11, and in accordance with child education and psychology theories, this application has designed a comprehensive data set content framework. This data set aims to promote the growth and development of children's thinking ability, language expression, emotional cognition, social skills, moral values, scientific exploration, etc., and covers the cognition, language, social, emotional and moral aspects of children aged 7-11. Figure 2 As shown in the figure, the content framework of the dataset specifically includes 8 major themes and 26 sub-themes.

[0026] In step S102 and step S103, instruction fine-tuning data integrating children's cognitive development theory is constructed: Instruction fine-tuning data is a key resource for fine-tuning large language models. Selecting an appropriate instruction fine-tuning dataset has a crucial impact on the fine-tuning effect of large language models. These datasets usually contain a series of text pairs that can help the model better understand and generate natural language. Currently, common fine-tuning dataset formats include: instruction following format, multi-round dialogue format, and other auxiliary formats.

[0027] This application selects the instruction following format. Figure 3 As shown, each piece of data consists of instruction, input, output, system, and history, where input, system, and history are optional. Each piece of data constructed in this application consists of three parts: instruction, input, and output. Among them, instruction is an instruction that explains or constrains the task of the model; input is used to supplement contextual information and provide additional context for specific tasks; output is the expected response that a large language model should produce. It is the result of the model's understanding and processing of instruction and input.

[0028] There are four main types of construction methods for instruction fine-tuning data: Human Generated Datasets (HG), Model Constructed Datasets (MC), Collection and Improvement of Existing Datasets (CI), and Mixed Methods to Create Datasets. This application uses two construction methods: Human Generated Datasets Combined with Model Constructed Datasets (HG&MC) and Collection and Improvement of Existing Datasets (CI) to construct data.

[0029] For data structures other than ancient poems and idioms, this application adopts a comprehensive method of artificial generation combined with model construction (HG&MC). Figure 4 As shown in the figure, first, we manually annotate the data set content framework, and manually organize and refine the instruction topics. Furthermore, we use a large language model to generate instructions and answers related to these topics in combination with the manually generated instruction topics. This step aims to enrich the content of the data set. Through the efficiency of the model, we can create a variety of learning scenarios and interactive forms, such as role-playing and scenario simulation. This not only improves the coverage of the data set, but also provides children with a more vivid and interactive learning experience.

[0030] For ancient poetry and idioms, the collection and improvement (CI) method of existing datasets was selected. Public Chinese natural language processing datasets of ancient poetry and idioms were manually screened, and finally those that can stimulate children's interest and promote cognitive development were selected, and those that may cause misunderstanding or are not suitable for children to learn were eliminated.

[0031] Data processing and normalization: In the process of using large language models to assist in generating data sets, problems such as wrong answers, merged instructions, and empty outputs occurred. To address these issues, this application performed data cleaning. The following are specific treatment measures: (1) Correction of incorrect answers: We carefully reviewed and proofread the incorrect answers in the original data set. By comparing them with authoritative data and expert knowledge, we corrected the incorrect answers and ensured the accuracy of the data set.

[0032] (2) Instruction separation: The merged instructions were split and reorganized. Each instruction was extracted separately and made clear and unambiguous to facilitate subsequent processing and analysis.

[0033] (3) Empty output processing: Empty output entries in the dataset were supplemented and improved according to the context and relevant instructions. Additional data were introduced or reasonable assumptions were made when necessary to fill these gaps.

[0034] For the open source datasets collected using the existing dataset collection and improvement methods, the content is not directly applicable to this task, and the format must be integrated and unified. In order to ensure the quality and applicability of the dataset, the following steps were taken to process and improve it: (1) Content screening: First, we conducted a comprehensive screening of the collected open source datasets to identify and remove inaccurate, irrelevant, or data that did not meet the research purpose. This step ensured the pertinence of the dataset.

[0035] (2) Data selection: Based on the screening, data closely related to the task of this paper were carefully selected. This process involves an in-depth understanding and evaluation of the data to ensure that the selected data can support the research objectives.

[0036] (3) Format standardization: The selected data is formatted in a standardized manner, including unifying the structure, format and encoding of the data to facilitate further processing and analysis of the data.

[0037] (4) Data conversion: In order to make the data set compatible with the format used in this paper, necessary data conversion work was carried out. This involves converting the data from the original format to a format suitable for this paper, including data type conversion, field mapping, etc.

[0038] The processed data is integrated and standardized. Through this process, data from different sources are unified into a standardized format to ensure the consistency of the data set in terms of structure, format, and metadata. Through the above steps, this application constructs an instruction fine-tuning dataset that integrates children's cognitive development theory. The dataset contains 8 major topics and a scale of 30,000 instruction dialogues.

[0039] In step S104, fine-tune the large language model: The Atom model jointly created by the Llama Chinese community and Atomic Echo is selected as the base model. This model is an open source Chinese large language model that supports fine-tuning. The LoRA method is selected to fine-tune the Atom model.

[0040] like Figure 5 As shown in the figure, the LoRA method assumes that the parameter changes of the model during task adaptation are low-rank. Parameter changes can be simulated by low-rank decomposition, thereby indirectly training LLM with minimal parameter requirements. This method adds a trainable layer (called a rank decomposition matrix) to each transformer block, and adds a dimension reduction matrix A and a dimension expansion matrix B next to the linear layer of the model. During training, the original parameters of the pre-training model (Pre-training Language Model, PLM) are frozen, and only the parameters of matrices A and B related to the task are updated. After the model training is completed, the parameters of A + B need to be fused with the parameters of PLM. It not only retains the powerful capabilities of PLM, but also can adapt to new tasks.

[0041] In addition to the LoRA method, the RoPE (Reparameterized Positional Encoding) interpolation method is introduced. This method allows the Transformer model to process longer contexts by reparameterizing the positional encoding, thereby improving the performance and adaptability of the model on long text tasks. This method can smoothly extend the context length of the model without large-scale modifications to the model architecture, allowing the model to understand and generate long sequence texts more effectively. By using this method, the model has improved its performance in multi-round dialogues.

[0042] In step S105, a child growth companion dialogue system is constructed: The child growth companion dialogue system implemented in this application is mainly composed of two parts: a large language model called by the back end and a web user interface deployed on the front end.

[0043] like Figure 6As shown, specifically, the large language model called by the backend is obtained by fine-tuning the Atom large language model with LoRA instructions using the data set constructed by this application, and the performance of the model in the field of children's growth companionship has been significantly improved. Figure 7 As shown in the figure, the front-end web user interface is implemented using the Gradio framework, which can quickly deploy machine learning models as web applications. The child companion dialogue model is deployed locally through Gradio, providing users with a visual interactive interface and allowing customization of some dialogue parameters.

[0044] Furthermore, this example embodiment also provides a language model-assisted child growth companion dialogue system 200. Figure 8 As shown in , the language model-assisted child growth companion dialogue system 200 may include: a child growth companion large language model 210 and a user interface 220. Among them, The large-scale language model 210 for children's growth companionship is used to directly generate dialogue information content related to children's growth companionship according to the input call instructions based on the knowledge and generation strategies learned during the fine-tuning process; The user interface 220 is used to provide a visual interactive interface and customize some dialogue parameters.

[0045] Specifically, the child growth companion dialogue system implemented in this application is mainly composed of two parts: a large language model called by the back end and a web user interface deployed on the front end.

[0046] like Figure 6 As shown, specifically, the large language model called by the backend is obtained by fine-tuning the Atom large language model with LoRA instructions using the data set constructed by this application, and the performance of the model in the field of children's growth companionship has been significantly improved. Figure 7 As shown in the figure, the front-end web user interface is implemented using the Gradio framework, which can quickly deploy machine learning models as web applications. The child companion dialogue model is deployed locally through Gradio, providing users with a visual interactive interface and allowing customization of some dialogue parameters.

[0047] Through the above-mentioned method for constructing a child growth companion dialogue system assisted by a language model, on the one hand, combined with the theoretical data of children's cognitive development, the cognitive characteristics of children are analyzed, the content framework of the children's cognitive data set is designed, and the instruction fine-tuning data integrating the children's cognitive development theory is constructed, and the instruction fine-tuning data is processed and standardized to construct a child growth companion data set. On the other hand, a large language model is selected as the base model of the child growth companion dialogue system assisted by a language model, and the LoRA fine-tuning method is used to fine-tune the instructions of the large language model based on the child growth companion data set, and a web interface is deployed for the trained large language model to improve the dialogue performance of the large language model in the field of child growth companionship, and at the same time provide users with a visual dialogue interaction interface.

[0048] It should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. in the above description indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present disclosure and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present disclosure.

[0049] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0050] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0051] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A method for constructing a language model-assisted children's growth companion dialogue system, characterized in that: The method includes: Based on the data of children's cognitive development theory, the content framework of children's cognitive data set was constructed; Based on the content framework of the children's cognitive data set, constructing instruction fine-tuning data using an artificial-model method and a collection-improvement method; wherein the instruction fine-tuning data includes first instruction data and second instruction data; Processing and uniformly organizing the instruction fine-tuning data to obtain a child growth companion data set; Based on the LoRA method, the selected base model is fine-tuned using the child growth companion dataset to obtain a large language model for child growth companion; A child growth accompaniment dialogue system is constructed based on the child growth accompaniment dataset and the child growth accompaniment large language model.

2. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 1, characterized in that: Based on the children's cognitive development theory data, the steps to construct the content framework of the children's cognitive data set include: According to the children's cognitive development theory data, the cognitive characteristics of children are analyzed to construct the content framework of the children's cognitive data set, which includes several large theme data sets, each of which has several small theme sub-data sets; wherein, The large subject data set includes at least an emotional morality data set, a social etiquette data set, a safety education data set, a cultural festival data set, a life common sense data set, a special scene data set, a common sense knowledge data set and a natural world data set; The emotional morality dataset at least includes a traditional morality sub-dataset and a core value sub-dataset; the social etiquette dataset at least includes a kinship sub-dataset, a getting along etiquette sub-dataset and a competition and cooperation sub-dataset; the safety education dataset at least includes a traffic sub-dataset, a privacy sub-dataset, a water use sub-dataset and an electronic use sub-dataset; the cultural festival dataset at least includes a Spring Festival sub-dataset, a Mid-Autumn Festival sub-dataset and a Dragon Boat Festival sub-dataset; the common sense of life dataset at least includes a physical health sub-dataset, a personal hygiene sub-dataset and a diet sub-dataset; the special scene dataset at least includes a campus sub-dataset, a hospital sub-dataset and a fire station sub-dataset; the common sense knowledge data map at least includes an ancient poetry sub-dataset, a Song Dynasty poetry sub-dataset and an idiom sub-dataset; the natural world dataset at least includes a sky sub-dataset, an ocean sub-dataset, a farm sub-dataset, a river sub-dataset and a forest sub-dataset.

3. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 2, characterized in that: Based on the content framework of the children's cognitive data set, the steps of constructing instruction fine-tuning data using the artificial-model method and the collection-improvement method include: For other data sets other than the ancient poetry sub-dataset, the Song Dynasty poetry sub-dataset and the idiom sub-dataset in the content framework of the children's cognitive data set, construct the first instruction data using the artificial-model method; For the ancient poetry sub-dataset, the Song Dynasty poetry sub-dataset and the idiom sub-dataset in the content framework of the children's cognitive data set, construct the second instruction data using the collection-improvement method; The instruction fine-tuning data is composed according to the first instruction data and the second instruction data.

4. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 3, characterized in that: The instruction fine-tuning data consists of instruction, input, output, system and history; wherein the input, the system and the history may be empty.

5. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 3, characterized in that: The step of processing and uniformly organizing the instruction fine-tuning data to obtain a child growth companion data set includes: Performing data cleaning on the first instruction data in the instruction fine-tuning data; wherein the data cleaning includes correcting wrong answers, separating instructions, and processing empty outputs; Integrate and uniformly process the second instruction data in the instruction fine-tuning data; wherein the integration and uniform processing includes content screening, data selection, format normalization and data conversion; The child growth companion data set is obtained according to the processed first instruction data and the processed second instruction data.

6. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 1, characterized in that: Based on the LoRA method, the step of fine-tuning the selected base model using the child growth companion dataset to obtain a large language model for child growth companion includes: Add a rank decomposition matrix to each transformer block in the base model, and add a dimension reduction matrix A and a dimension expansion matrix B next to the linear layer of the base model; When the base model is trained using the child growth companion dataset, the original PLM parameters are frozen, and the parameters of the dimension reduction matrix A and the dimension expansion matrix B related to the task are updated; After the training is completed, the parameters of the dimension reduction matrix A and the dimension expansion matrix B are fused with the PLM original parameters, and the PLM original parameters of the base model are updated to obtain the large-scale language model for children's growth companionship.

7. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 6, characterized in that: Based on the LoRA method, the step of fine-tuning the selected base model using the child growth companion dataset to obtain a large language model for child growth companionship also includes: When training the base model, the RoPE interpolation method is introduced to reparameterize the position encoding.

8. The method for constructing a language model-assisted children's growth companion dialogue system according to claim 1, characterized in that: The step of constructing a child growth companion dialogue system according to the child growth companion dataset and the child growth companion large language model includes: Using the child growth companionship dataset as a dataset of the child growth companionship dialogue system; Using the large-scale language model for children's growth companionship as a back-end calling language model for the children's growth companionship dialogue system; A user interface is used as a visual interactive interface of the child growth companion dialogue system.

9. A language model-assisted children's growth companion dialogue system, characterized in that: The system includes: The large-scale language model for children's growth companionship is used to generate dialogue information content related to children's growth companionship directly according to the input call instructions based on the knowledge and generation strategies learned during the fine-tuning process; The user interface is used to provide a visual interactive interface and customize some dialogue parameters.

Images