Personalized function page generation method and system based on large model

Abstract

The invention provides a personalized function page generation method and system based on a large model, and belongs to the technical field of large models.The personalized function page generation method comprises the steps that firstly, function page requirement description including function expectation and style preference input by a user through a natural language is obtained, and then a requirement analysis module of the large model is called for semantic analysis; the method comprises the steps of obtaining function demand features and style preference features, then retrieving a preset component library matching page component set based on the function demand features, determining component display attributes according to the style preference features, then generating an initial function page layout according to the component set, the display attributes and a preset layout rule, and finally monitoring user interaction behavior data in real time. And a large model dynamic adjustment module is input to adjust the display attribute and the layout structure of the page component to obtain a personalized function page, so that the function page meeting the personalized requirements of the user can be efficiently and accurately generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large models, in particular to a personalized function page generation method and system based on a large model. BACKGROUND

[0002] In the current digital application boom, the design of function pages of various software and web pages is crucial to improving user experience and meeting user needs. However, traditional function page generation methods have many shortcomings.

[0003] On the one hand, traditional methods usually rely on professional developers to implement the construction of function pages through manual coding. Developers need to understand user needs in depth and translate them into specific code logic, which is not only time-consuming and labor-intensive, but also requires high professional skills of developers. At the same time, due to the differences in technical level and understanding ability of different developers, the generated pages are uneven in functional integrity and user experience.

[0004] On the other hand, although existing partially automated page generation tools can simplify the development process to some extent, they often lack accurate grasp of user's personalized needs. The above tools usually generate pages based on pre-set templates and rules, and it is difficult to flexibly adjust according to user's expectations of page functions and preferences of presentation styles. Users can only choose from limited templates and cannot obtain personalized function pages that meet their own needs, thereby affecting user experience and satisfaction. In addition, traditional methods lack effective monitoring and dynamic adjustment mechanism for user interaction behavior after page generation, and cannot optimize and improve the page according to user's actual use, resulting in poor practicability and adaptability of the page. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a personalized function page generation method based on a large model, which comprises: obtaining a function page requirement description input by a user through natural language, the function page requirement description containing user's expectations of page functions and preferences of page presentation styles; calling a requirement analysis module of a large model to perform semantic analysis processing on the function page requirement description, to obtain function requirement features and style preference features of the user, the function requirement features reflecting the user's expected page function types and interaction logic, and the style preference features reflecting the user's expected page color matching, font style and layout structure; retrieve a preset component library based on the functional requirement features, match to obtain a page component set corresponding to the functional requirement, and determine display attributes of each component in the page component set based on the style preference features, the display attributes including component size, position proportion, and color parameters; generate an initial functional page layout according to the page component set, the display attributes of each component, and a preset layout rule; monitor user interaction behavior data with the initial functional page in real time, input the interaction behavior data into a dynamic adjustment module of the large model, adjust the display attributes of the page components and the page layout structure based on the interaction behavior data, and obtain a personalized functional page.

[0006] In another aspect, the embodiment of the present application also provides a personalized functional page generation system based on a large model, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0007] Based on the above aspects, the embodiment of the present application can make the user express his own requirements more directly by obtaining the functional page requirement description input by the user in natural language, can accurately extract the functional requirement features and the style preference features of the user by calling the requirement analysis module of the large model to perform semantic analysis on the requirement description, can realize intelligent selection and personalized customization of the page components by retrieving a preset component library based on the functional requirement features and matching a page component set, and determining the display attributes of the components according to the style preference features, can improve the efficiency and accuracy of page generation, can generate an initial functional page layout according to the page component set, the display attributes, and a preset layout rule, can then monitor the interaction behavior data of the user with the initial functional page in real time, and input the interaction behavior data into the dynamic adjustment module of the large model, can dynamically adjust the display attributes of the page components and the page layout structure based on the interaction behavior data, can make the page adaptively optimized according to the actual use of the user, can continuously fit the requirements and use habits of the user, can efficiently and accurately generate a page meeting the personalized requirements of the user, and can significantly improve the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 is an execution flow diagram of the personalized functional page generation method based on a large model provided by the embodiment of the present application.

[0009] Figure 2 is a schematic diagram of exemplary hardware and software components of the personalized functional page generation system based on a large model provided by the embodiment of the present application. DETAILED DESCRIPTION

[0010] The application will be described in detail below with reference to the accompanying drawings, Figure 1 is a flowchart of a personalized function page generation method based on a large model provided by an embodiment of the application. The personalized function page generation method based on a large model will be described in detail below.

[0011] Step S110: Obtain a function page requirement description input by a user in natural language, the function page requirement description containing the user's expectation of page functions and preference for page presentation styles.

[0012] This embodiment takes an online learning platform as an application scenario, and a user can input a natural language description through an interactive entrance provided by the online learning platform. The natural language description needs to comprehensively cover two aspects, one is the expectation of page functions, such as the functions of course resource management, student homework correction, and learning data statistics, and the other is the preference for page presentation styles, such as color matching, font selection, and module layout of the page.

[0013] Users with different identities will have different demand emphases. Teachers may pay more attention to the implementation of teaching management related functions, students may pay more attention to learning resource acquisition and interactive communication functions, and parents may focus on the viewing function of student learning conditions.

[0014] Step S111: Receive the natural language requirement description submitted by the user through the input interface, the input interface supporting text input, voice-to-text input, and image with text description input.

[0015] The input interface of the online learning platform is composed of multiple function modules. The text input module is manifested as a text input box of the web page end and the mobile end of the platform, and the user can directly input text description by typing on the keyboard in the text input box. The input box has the functions of automatic line breaking and word count prompting, and when the input content reaches a certain length, a scroll bar can be displayed to facilitate the user to view all the content.

[0016] The voice-to-text input module includes a voice collection unit and a voice recognition unit. After the user clicks the voice input button, the voice collection unit starts the microphone permission and begins to collect the voice information of the user. During the collection process, a waveform graph is displayed in real time to feedback the current recording state. The collected voice information is transmitted to the voice recognition unit, which uses a deep learning based voice recognition model to pre-process the voice signal, including noise reduction, framing, and other operations, and then converts the processed signal into corresponding text information.

[0017] The Image Caption Input module allows users to upload an image file and enter a supplementary description of the image content in the text box below the image. The platform supports common image formats and performs format verification and size restrictions during the upload process. If the image does not meet the requirements, the user will be prompted to re-upload. After a successful upload, the image and accompanying text description will be stored in a linked format.

[0018] Step S112: Identify the identity type of the input user, where the identity types include student, teacher, and parent. Users with different identity types correspond to different emphasis on the demand description.

[0019] The platform identifies a user's identity type based on their login information. When a user logs in, their account information is verified and the corresponding identity identifier is retrieved from the user database. The identity identifier is set and stored during user registration, forming a unique association with the account information.

[0020] If the user is not logged in, the input interface will prompt the user to log in first, otherwise the function page requirement description cannot be submitted. The login status is monitored in real time. If the login status fails during the input process, the system will retain the current input content and prompt the user to log in again.

[0021] Step S113: For users with student identity, extract the functional expectation statements related to learning content in the natural language demand description input by the user; for users with teacher identity, extract the functional expectation statements related to teaching management in the natural language demand description input by the user; for users with parent identity, extract the functional expectation statements related to the display of student learning status in the natural language demand description input by the user.

[0022] For students, we use a combination of keyword matching and semantic analysis to extract relevant expressions. We first define a set of keywords related to the learning content, such as "course video," "exercise exercises," and "knowledge point explanation." We then segment the user's input text, matching the segmentation results with the keyword set. We also use semantic analysis to determine the core meaning of the sentence, filtering out expressions of functional expectations related to the learning content.

[0023] For users with teacher identities, a set of keywords related to teaching management is also set, such as "homework assignment", "grade entry", "student attendance", etc. Through a similar word segmentation and matching process, combined with semantic understanding, the functional expectation expressions related to teaching management are extracted.

[0024] For users with parental identity, set a set of keywords related to the display of students' learning status, such as "learning progress", "test scores", "classroom performance", etc., and extract the corresponding functional expectation statements according to the above method.

[0025] Step S114: Redundancy information filtering processing is performed on the extracted demand expressions of users of different identity types, and core demand descriptions are retained as functional page demand descriptions input by users through natural language.

[0026] The redundancy information filtering processing includes removing repetitive content and meaningless expressions. The system first compares the extracted demand expressions, identifies repetitive sentences or paragraphs, and retains only one of them.

[0027] For meaningless expressions such as adverbs, repeated exclamation marks, etc., the system establishes a meaningless vocabulary library, matches the vocabulary in the demand expression with the vocabulary in the library, and filters out the matched vocabulary. After filtering, the remaining content is determined as the core demand description.

[0028] Step S120: A demand analysis module of a large model is called to perform semantic analysis processing on the functional page demand description, to obtain functional demand features and style preference features of the user, the functional demand features reflecting the page function type and interaction logic expected by the user, and the style preference features reflecting the page color matching, font style and layout structure expected by the user.

[0029] After receiving the processed core demand description, the demand analysis module of the large model starts the semantic analysis process. The demand analysis module includes multiple sub-modules, each responsible for analyzing the functional demand and style preference.

[0030] In the analysis process, the demand analysis module performs deep semantic analysis on the text, identifies descriptions about page function type, interaction logic, and descriptions about color matching, font style, and layout structure, and further generates corresponding functional demand features and style preference features.

[0031] Step S121: The functional page demand description is input into the word vector conversion layer of the large model to generate a corresponding demand word vector sequence, each vector in the demand word vector sequence corresponding to a word unit in the demand description.

[0032] After receiving the functional page demand description text, the word vector conversion layer first preprocesses the text, including removing punctuation, unifying case, etc. Then, the processed text is segmented into multiple word units, each word unit corresponding to a word or word group in the text.

[0033] For each word unit, the word vector conversion layer converts it into a fixed-dimensional vector through a pre-trained word embedding model. The above vector can capture the semantic information of the word, so that semantically similar words are closer in the vector space.

[0034] The generated demand word vector sequence is arranged according to the order of the vocabulary units in the original text to form an ordered vector sequence.

[0035] Step S122: Call the context encoding module of the large model to perform semantic association modeling on the demand word vector sequence to generate an encoding vector containing context dependency.

[0036] After receiving the required word vector sequence, the context encoding module performs semantic association modeling. This context encoding module uses the Transformer architecture to capture context information through a multi-head attention mechanism and a feedforward neural network layer.

[0037] Step S1221: Input the required word vector sequence into the multi-head attention layer of the context encoding module, and calculate the attention weight value between each vocabulary unit and other vocabulary units. The attention weight value is used to represent the semantic association strength between vocabulary units.

[0038] The multi-head attention layer maps the required word vector sequence through multiple linear transformation matrices to obtain multiple different query matrices, key matrices, and value matrices. For each matrix combination, the dot product of the query matrix and the key matrix is ​​calculated, and then the original attention weight is obtained after scaling.

[0039] The original attention weights are normalized by the softmax function to obtain the attention weight values ​​between each vocabulary unit and other vocabulary units. The above weight values ​​reflect the closeness of the association between each vocabulary unit in the current context.

[0040] Step S1222: performing weighted aggregation processing on the demand word vector sequence based on the attention weight value to generate a context enhancement vector for each vocabulary unit, wherein the context enhancement vector integrates the semantic information of other vocabulary units whose association with the context unit meets the preset association standard.

[0041] According to the attention weight value, the vector corresponding to each vocabulary unit in the demand word vector sequence is weighted. The vector of each vocabulary unit is weighted and summed with the vectors of other vocabulary units according to the attention weight value to obtain a preliminary context enhancement vector.

[0042] The preset relevance criterion is a threshold value for the attention weight value. For vocabulary units with weight values ​​below this threshold, their semantic information will not be included in the context enhancement vector of the current vocabulary unit. This ensures that the context enhancement vector only incorporates the semantic information of vocabulary units with high relevance.

[0043] Step S1223: Input the context enhancement vector into the bidirectional recurrent layer for temporal association modeling to generate a temporal coding vector containing the semantic dependency of the context, wherein each dimension of the temporal coding vector corresponds to the semantic feature of a position in the demand word vector sequence.

[0044] The bidirectional recurrent layer consists of a forward recurrent unit and a backward recurrent unit. The context enhancement vector is sequentially input into the forward recurrent unit, which calculates the current hidden state based on the current input and the hidden state at the previous moment, capturing the positive temporal association between vocabulary units.

[0045] At the same time, the context enhancement vector is input into the backward recurrent unit in reverse order. The backward recurrent unit also calculates the hidden state at the current moment based on the current input and the hidden state at the previous moment, capturing the reverse temporal association between vocabulary units.

[0046] The hidden states of the forward recurrent unit and the backward recurrent unit at each position are concatenated to generate a temporal encoding vector, which contains the contextual semantic dependencies of the vocabulary unit in the entire sequence.

[0047] Step S1224: performing layer normalization processing on the temporal coding vector to obtain a standardized temporal coding vector.

[0048] Layer normalization is performed on each feature dimension of the temporal encoding vector. The mean and variance of all vector elements in each feature dimension are calculated. The mean is then subtracted from each element and divided by the square root of the variance. The element is then multiplied by the scaling parameter and the offset parameter is added to obtain a standardized temporal encoding vector.

[0049] Through layer normalization, the distribution of temporal coding vectors in different feature dimensions is made more stable, which is beneficial to the processing of subsequent network layers.

[0050] Step S1225: The standardized temporal coding vector is dimensionally converted through the output layer of the context coding module to generate a coding vector with the same length as the required word vector sequence, and each element of the coding vector contains contextual semantic information of the corresponding vocabulary unit.

[0051] The output layer processes the normalized temporal encoding vector using a linear transformation. A linear transformation matrix is ​​used to transform the dimension of the normalized temporal encoding vector to a dimension that matches the number of vocabulary units in the required word vector sequence, generating an encoding vector.

[0052] Each element in the encoding vector corresponds to a vocabulary unit in the required word vector sequence and contains the semantic information of the vocabulary unit in the entire context.

[0053] Step S123: performing feature splitting processing on the coding vector through the feature separation layer of the large model to separate the functional coding sub-vector related to functional requirements and the style coding sub-vector related to style preferences.

[0054] The feature separation layer contains two parallel convolutional neural network branches, one for extracting the function encoding sub-vector and the other for extracting the style encoding sub-vector. The encoding vector is input into both branches simultaneously.

[0055] The branch used to extract the function encoding sub-vector uses multiple convolution kernels that have been trained to identify semantic features related to functional requirements. Through convolution and pooling operations, the function encoding sub-vector is extracted from the encoding vector.

[0056] The other branch also uses multiple convolution kernels, which are used to identify semantic features related to style preferences. After convolution and pooling operations, the style encoding sub-vector is extracted.

[0057] Step S124: Perform function dimension mapping processing on the function coding sub-vector to generate a function requirement feature including a function type identifier, an interaction logic description, and a priority ranking. The function type identifier is used to distinguish different page function categories, the interaction logic description is used to illustrate the calling relationship between functions, and the priority ranking is used to indicate the importance of each function.

[0058] Functional dimension mapping is achieved through multiple fully connected layers. The function encoding sub-vector is processed in sequence by multiple fully connected layers. Each fully connected layer performs a nonlinear transformation on the input through an activation function, gradually extracting higher-level functional features.

[0059] Step S1241: calling a function classifier to perform category determination on the function coding sub-vector, and outputting a corresponding function type identifier, which covers information display class, operation interaction class, data processing class, and feedback response class.

[0060] The feature classifier consists of multiple fully connected layers and a softmax output layer. The feature encoding subvector is input into the fully connected layer and, after multiple nonlinear transformations, enters the softmax output layer.

[0061] The softmax output layer calculates the probability of each feature type and outputs the identifier corresponding to the feature type with the highest probability. For example, if a feature is determined to belong to the information display class, the corresponding information display class identifier is output.

[0062] Step S1242: Extract the semantic features related to the interaction in the function coding sub-vector and generate an interaction logic description. The interaction logic description includes the function triggering condition, the function execution process and the function result return method. The function triggering condition is used to describe the user operation required to start the function, the function execution process is used to describe the processing steps within the function, and the function result return method is used to describe the presentation form of the information after the function is executed.

[0063] By performing semantic analysis on the function encoding sub-vector, the interaction-related features are identified. These features include the part describing the user operation, the part describing the internal processing steps of the function, and the part presenting the results.

[0064] Based on these features, we generate function trigger conditions, such as "start the function after clicking the button"; function execution process, such as "first verify user permissions, then load data, and finally process data"; function result return method, such as "display in the middle of the page in list form" and other contents, which are combined to form an interactive logic description.

[0065] Step S1243: Calculate the importance score of each function type based on the semantic strength value in the function coding sub-vector. The semantic strength value is determined by the projection length of the function coding sub-vector in the semantic activation direction. The larger the projection length, the higher the degree to which the corresponding function is emphasized in the requirement description.

[0066] First, the semantic activation direction vector corresponding to each feature type is determined. These vectors are obtained by training on a large amount of annotated data. The feature encoding subvector is projected onto the semantic activation direction vector of each feature type, and the projection length is calculated. This projection length is the semantic strength value.

[0067] According to the semantic strength value, the importance score of each functional type is calculated in a normalized manner so that the sum of the importance scores of all functional types is 1. The larger the semantic strength value, the higher the corresponding importance score.

[0068] Step S1244: sorting each function type according to the importance score to generate a priority ranking, wherein the function type with a higher sequence position in the priority ranking indicates that it should be displayed first when the page is generated.

[0069] Arrange each function type in descending order of importance to form a priority ranking. For example, if the information display type has the highest importance score, it will be ranked first in the priority ranking, followed by the operation interaction type, and so on.

[0070] Step S1245: combining the function type identifier, interaction logic description and priority ranking according to a preset format to generate the user's function requirement characteristics.

[0071] The default format is a structured data format, which maps the function type identifier, interaction logic description, and priority ranking to different fields of the structured data. For example, the function type identifier corresponds to the "Function Category" field, the interaction logic description corresponds to the "Interaction Rules" field, and the priority ranking corresponds to the "Display Priority" field. The combination of these forms the functional requirement characteristics.

[0072] Step S125: Perform style dimension mapping processing on the style encoding sub-vector to generate a style preference feature including color system parameters, font attribute parameters, and layout mode parameters. The color system parameters reflect the user's preference for the main color and auxiliary color of the page, the font attribute parameters reflect the user's preference for the font type and size, and the layout mode parameters reflect the user's preference for the arrangement of page elements.

[0073] The style dimension mapping process is also implemented through multiple fully connected layers. The style encoding subvector undergoes nonlinear transformations in the fully connected layers to extract style features related to color, font, and layout.

[0074] For example, step S1251: parse the color-related semantic features in the style encoding sub-vector, determine the user's preferred main color range and auxiliary color matching rules, and generate color system parameters. The main color range is represented by a coordinate interval in the color space, and the auxiliary color matching rules are used to describe the brightness contrast and hue circle distance between the auxiliary color and the main color.

[0075] The color-related part of the style encoding sub-vector is parsed to identify the main color tone information described by the user and convert it into a coordinate interval in the color space, such as the value range of the red, green, and blue channels in the RGB color space, to form the main color tone range.

[0076] At the same time, the relationship between the secondary color and the primary color is analyzed to determine the brightness contrast range and the distance range on the hue circle, forming a pairing rule for the secondary color. For example, the brightness contrast between the secondary color and the primary color should be within a certain ratio range, and the distance on the hue circle should be within a certain angle range.

[0077] Step S1252: Extracting font-related semantic features from the style encoding sub-vector, determining the user's preferred font type and font size range, and generating font attribute parameters. The font types include serif fonts, sans serif fonts, and handwriting fonts, and the font size range is represented by relative size levels.

[0078] Extract font-related semantic features from the style encoding sub-vector to identify the font type mentioned by the user, such as sans-serif font, and determine the relative size range of the font, such as medium-large to maximum, to form font attribute parameters.

[0079] Step S1253: Analyze the layout-related semantic features in the style encoding sub-vector, determine the user's preferred page element arrangement method and area division ratio, and generate layout mode parameters. The page element arrangement method includes flow layout, grid layout and hierarchical layout. The area division ratio is used to illustrate the area ratio of different functional areas in the page.

[0080] Analyze the layout-related semantic features in the style encoding sub-vector to determine the user's desired arrangement of page elements, such as a grid layout. Simultaneously, analyze the area ratios of different functional areas, such as the left functional area accounting for two-thirds and the right functional area accounting for one-third, to form layout pattern parameters.

[0081] Step S1254: The color system parameters, font attribute parameters, and layout mode parameters are associated and stored to generate the user's style preference characteristics.

[0082] The color system parameters, font attribute parameters and layout mode parameters are stored in an associated manner in the form of key-value pairs. Each parameter is used as a value in a key-value pair, and the corresponding parameter name is used as the key. The combination forms a style preference feature.

[0083] Step S130: Retrieve the preset component library based on the functional requirement characteristics, match and obtain a page component set corresponding to the functional requirement, and determine the display attributes of each component in the page component set based on the style preference characteristics, wherein the display attributes include component size, position ratio and color parameters.

[0084] The preset component library stores a large number of pre-developed page components, each with corresponding functional tags and configurable properties. The system searches the component library based on the functional type identifier in the functional requirement characteristics, matches the components that meet the requirements, and configures the display properties of the components based on the style preference characteristics.

[0085] Step S131: extracting the function type identifier and priority ranking from the function requirement characteristics, using the function type identifier as a search keyword, and searching the preset component library.

[0086] Extract various function type identifiers from the function requirement characteristics, such as information display type identifier, operation interaction type identifier, etc., use these identifiers as search keywords, and search in the index system of the preset component library.

[0087] Step S132: The preset component library stores multiple component units, each component unit includes a function matching tag, basic structure parameters and a configurable property set. The function matching tag is used to match the function type identifier, the basic structure parameters are used to describe the core function implementation method of the component, and the configurable property set is used to represent the adjustable display property range of the component.

[0088] The function matching tag of a component unit is pre-set and corresponds to the function type identifier. Infrastructure parameters include the component's code structure, data interface specifications, and a list of dependent resources. The code structure clarifies the component's implementation logic and module divisions. For example, the code structure of an information display component includes a data receiving module, a rendering module, and an update module. The data receiving module is responsible for obtaining display data from the backend, the rendering module presents the data to the page in a specific format, and the update module triggers re-rendering when the data changes.

[0089] The data interface specification defines the format and protocol for data exchange between components and the backend, including the type and format of request parameters and the structure of response data, ensuring that components can correctly retrieve and parse data. The dependency resource list lists external resources required for component operation, such as stylesheets and script files. These resources are automatically imported when the component loads.

[0090] The configurable property set covers various adjustable parameters for a component's display. For color properties, this includes the settable ranges for primary, secondary, and background colors. Font properties include optional values ​​for font type, size, and weight. Layout properties include adjustment ranges for width, height, margins, and positioning. Each configurable property has clear value constraints to ensure the component continues to display and function properly after adjustments.

[0091] Step S133: Calculate the semantic similarity between the function type identifier and the function matching label of the component unit, and select the component units whose semantic similarity meets the preset conditions as candidate components.

[0092] First, the function type identifier and component unit function matching label are converted into vector representations. The function type identifier and function matching label are processed using a word embedding model to generate corresponding vectors. These vectors are in the same vector space and can reflect their semantic information.

[0093] Then, semantic similarity is determined by calculating the cosine similarity between the two vectors. Cosine similarity is calculated based on the angle between the vectors. The smaller the angle, the larger the cosine similarity value, indicating a closer semantic connection between the two.

[0094] The calculated semantic similarity is compared with the preset condition, which is a semantic similarity threshold. If the semantic similarity between the function matching label and the function type identifier of a component unit is greater than the threshold, the component unit is included in the candidate component list.

[0095] For example, step S1331: converting the function type identifier into a corresponding function vector, converting the function matching label of the component unit into a corresponding label vector, and the function vector and the label vector are represented using the same vector space.

[0096] Both function type identifiers and function matching labels are converted using the same pre-trained word embedding model. This word embedding model is trained on a large-scale text corpus and can map text information into a high-dimensional vector space, where semantically similar texts are closer.

[0097] For example, for the functional type identifier "information display class", the word embedding model will convert it into a specific functional vector; for the functional matching label "component for displaying course information" in the component unit, it will also be converted into a label vector, and these two vectors are in the same vector space, which is convenient for subsequent similarity calculations.

[0098] Step S1332: Calculate the cosine similarity between the function vector and the label vector to obtain the semantic similarity value between the function type identifier and the function matching label.

[0099] The calculation process of cosine similarity is: first calculate the dot product of the function vector and the label vector, then calculate the modulus of the two vectors respectively, and finally divide the dot product result by the product of the two moduli. The result is the cosine similarity value.

[0100] The value ranges from -1 to 1. The closer the value is to 1, the more similar the semantics of the function type identifier and the function matching label are; the closer the value is to -1, the greater the semantic difference between the two.

[0101] Step S1333: Compare the semantic similarity value with a preset threshold value. If the semantic similarity value meets the preset condition, the corresponding component unit is included in the candidate component list.

[0102] The preset threshold is set according to the actual application scenario and requirements, and is generally between 0 and 1. When the calculated semantic similarity value is greater than or equal to the preset threshold, the function matching label and function type identifier of the component unit are considered to be semantically similar enough to meet the preset conditions, and it is added to the candidate component list.

[0103] For example, if the preset threshold is 0.7 and the semantic similarity value of a component unit is 0.8, the component unit will be included in the candidate component list; if the semantic similarity value is 0.6, it will not be included.

[0104] Step S1334: Perform functional redundancy detection on the component units in the candidate component list. If the core functions of two component units completely overlap, retain the component units with semantic similarity values ​​greater than the set similarity, remove the component units with semantic similarity values ​​not greater than the set similarity, and use the component units that have passed the redundancy detection as candidate components.

[0105] Functional redundancy detection is achieved by analyzing the core functional descriptions in the basic structural parameters of the component units. The core functional descriptions clearly define the main functional purposes of the component units, such as "displaying course list" and "submitting homework".

[0106] The core functional descriptions of all component units in the candidate component list are compared pairwise. If two component units have identical core functional descriptions, they are considered to have functional redundancy. At this point, the semantic similarity values ​​of these two component units and the functional type identifiers are compared. The component unit with the larger semantic similarity value is retained, and the other is removed.

[0107] If the semantic similarity values ​​of two component units are the same, one of them is randomly retained. After the above processing, there will be no component units with completely redundant functions in the candidate component list.

[0108] Step S134: Convert the priority ranking into a numerical weight coefficient, and jointly determine the component screening order with the semantic similarity, perform weighted sorting on the candidate components, and select the K candidate components with the highest positions in the sorting sequence to form a page component set corresponding to the functional requirements.

[0109] Priority ranking reflects the importance of each feature type. When converted into numerical weight coefficients, the corresponding weight coefficients are assigned in descending order from high to low priority. For example, the highest priority feature type has a weight coefficient of 1.0, the second highest has a weight coefficient of 0.8, and so on. The specific values ​​are set according to the length and gradient of the ranking.

[0110] For each candidate component, its semantic similarity with the function type identifier is multiplied by the weight coefficient corresponding to that function type to obtain a weighted score. The candidate components are then sorted from highest to lowest weighted score, and the top K candidate components are selected to form the page component set. The value of K is determined based on the page's load capacity and the complexity of the functional requirements, ensuring that the page is not cluttered with too many components while still meeting core functional requirements.

[0111] Step S135: extracting the color system parameters, font attribute parameters and layout mode parameters from the style preference features as a basis for configuring component display attributes.

[0112] Color system parameters, font attribute parameters, and layout mode parameters are extracted from the style preference features. Color system parameters include the primary color range and auxiliary color matching rules; font attribute parameters include font type and size range; and layout mode parameters include the arrangement of page elements and the ratio of area division.

[0113] These parameters will serve as the specific basis for subsequent adjustments to component display properties to ensure that the component's display effect meets the user's style preferences.

[0114] Step S136: For each component in the page component set, adjust the color attribute of the component according to the color system parameter, adjust the text display attribute of the component according to the font attribute parameter, and adjust the size ratio and position parameter of the component according to the layout mode parameter, to generate the display attribute of each component, which includes the component size, position ratio and color parameter.

[0115] For each component, the attribute adjustment is performed according to the extracted style parameters one by one. During the adjustment process, the configurable attribute set of the component itself needs to be referred to, to ensure that the adjusted attribute is within the allowed configuration range of the component.

[0116] For example, step S1361: For each component in the page component set, parse the color configuration item in the configurable attribute set of the component, determine the main color value of the component according to the main tone range in the color system parameter, determine the auxiliary color value of the component according to the auxiliary color matching rule, and assign the main color value and the auxiliary color value to the corresponding color configuration item of the component, to complete the adjustment of the color attribute of the component.

[0117] The color configuration item of the component is parsed to determine the specific positions of the main color and the auxiliary color that need to be set, such as the color of the title, the color of the border, the color of the background, etc.

[0118] According to the main tone range in the color system parameter, a specific color value is selected as the main color value of the component within the range, such as selecting a specific blue value as the color of the title within the main tone range of blue.

[0119] According to the auxiliary color matching rule, the auxiliary color value is determined in combination with the main tone value. For example, if the auxiliary color matching rule requires that the brightness contrast of the auxiliary color and the main tone be within a certain range, then according to the brightness value of the main tone, an auxiliary color value that meets the contrast range is selected as the color of the border or the color of the background, etc., and these color values are assigned to the corresponding color configuration items.

[0120] Step S1362: Parse the font configuration item in the configurable attribute set of the component, determine the text font of the component according to the font type in the font attribute parameter, determine the text size of the component according to the font size range, and apply the text font and the text size to the font configuration item of the component, to complete the adjustment of the text display attribute of the component.

[0121] The font configuration item of the component is parsed to determine the font and font size configurable items of different text parts in the component, such as title text, content text, prompt text, etc.

[0122] According to the font type in the font attribute parameter, a corresponding font is selected for each text part of the component, such as a specific font in a sans-serif font.

[0123] Based on the font size range, combined with the overall design of the component and the importance of the text part, determine the specific font size for different text parts. For example, choose a larger font size within the font size range for the title text and a medium font size for the content text. Apply the determined font and font size to the corresponding font configuration item.

[0124] Step S1363: parse the layout configuration items in the component configurable property set, determine the size ratio of the component according to the area division ratio in the layout mode parameters, determine the position coordinates of the component on the page according to the arrangement of page elements, set the size ratio and position coordinates to the layout configuration items of the component, and complete the adjustment of the component size ratio and position parameters.

[0125] Parse the layout configuration items of the component to understand the adjustable layout parameters such as the component's width, height, positioning method, etc.

[0126] The component's size within its functional area is determined based on the area division ratio in the layout mode parameters. For example, if a functional area occupies half of the page and the component's size within that area occupies one-third, the component's overall size on the page will be one-sixth.

[0127] Based on the layout of page elements, such as a grid layout, determine the specific position coordinates of the component in the grid and specify its row and column. Set the determined size ratio and position coordinates to the component's layout configuration items to complete the layout property adjustment.

[0128] Step S1364: Integrate the adjusted color attributes, text display attributes, size ratios, and position parameters to generate display attributes including component size, position ratio, and color parameters.

[0129] The adjusted color attributes, text display attributes, size ratio, and position parameters are aggregated and organized into the component's display attributes according to a preset format. Component size is determined by size ratio, position ratio reflects the component's space usage on the page, and color parameters include information such as primary and secondary color values.

[0130] The display properties of each component are associated with the component one by one, which facilitates the generation of subsequent page layouts.

[0131] Step S137: Associating and binding the page component set with the display attributes of each component so that the display attributes of each component are adapted to the priority ranking of the component in the functional requirement characteristics.

[0132] By establishing a mapping relationship, each component in the page component set is associated and bound with its corresponding display attribute. In the binding process, the display attribute is fine-tuned by referring to the position of the component's corresponding function type in the priority order.

[0133] For example, the size proportion in the display attribute of the component corresponding to the function type with higher priority can be appropriately increased, and the position can be closer to the visual focus area of the page to highlight its importance and ensure that the display effect of the component matches the priority of the function.

[0134] Step S140: generating an initial function page layout according to the page component set, the display attribute of each component, and the preset layout rule.

[0135] The preset layout rule includes function area division principles, component placement specifications, interactive logic adaptation, and other content. The system plans and designs the page layout according to these rules, in combination with the page component set and the display attribute of each component, to determine the final position, size, and style of each component in the page, forming an initial function page layout.

[0136] Step S141: extracting the function type identifier and priority order of each component in the page component set, and dividing the page into multiple function areas based on the function area division principles in the preset layout rule, each function area corresponding to one or more function type identifiers.

[0137] The function area division principles in the preset layout rule divide the page into different function areas according to the relevance of function types and the visual layout habits of the page. For example, information display functions are concentrated in the left area of the page, operation and interaction functions are concentrated in the right area, or data processing functions are placed in the upper area of the page, and feedback response functions are placed in the lower area.

[0138] According to the function type identifier of each component, the function type corresponding to each function area is determined to ensure that components belonging to the same or related function type can be allocated to the same function area. At the same time, in combination with the priority order, a function area with a larger area or a more optimal position is allocated to the function type with higher priority.

[0139] Step S142: according to the priority order of each component, the K components at the front of the priority order sequence are allocated to the function area with a visual weight conforming to a first preset level in the page, and the M components at the back of the priority order sequence are allocated to the function area with a visual weight conforming to a second preset level in the page, the visual weight being determined by the position and area size of the function area in the page, and the visual weight of the first preset level being greater than that of the second preset level.

[0140] The first preset level of visual weight corresponds to a prominent and larger area of ​​the page, such as the upper center area of ​​the page; the second preset level corresponds to a relatively less important and smaller area, such as the edge or bottom area of ​​the page.

[0141] Based on the component priority ranking, the top K components are assigned to functional areas with a first preset visual weight. These components are typically the most frequently used or most important functional components. The bottom M components are assigned to functional areas with a second preset visual weight. The values ​​of K and M are determined based on the size of the functional area and the number of components to ensure that each functional area can reasonably accommodate the assigned components.

[0142] Step S143: Based on the size ratio and position parameters in the display properties of each component, determine the specific placement position and display size of each component in the corresponding functional area so that there is no overlap between the components.

[0143] In each functional area, the actual display size of each component is calculated based on the size ratio in the component display properties. Combined with the position parameters, the specific coordinate position of the component in the functional area is determined.

[0144] When determining the position, a collision detection mechanism is used to check whether the newly placed component overlaps with the already placed components. If there is overlap, the component's position parameters are adjusted, such as fine-tuning the coordinate values, until all components can be clearly displayed within the functional area and no components overlap.

[0145] Step S144: Analyze the functional association relationship between components, and place components with direct interaction relationships in adjacent positions according to the interactive proximity principle in the preset layout rules to shorten the user operation path.

[0146] The functional association between components is reflected in their interaction logic description, for example, the output of one component is the input of another component, or the operation of one component triggers the response of another component.

[0147] The interactive proximity principle in the preset layout rules requires that components with direct interactive relationships should be placed in adjacent positions as much as possible to reduce page jumps and movement distances during user operation.

[0148] For example, step S1441: obtain the interaction logic description of each component in the page component set, extract the function call relationship between the components, and generate a component interaction relationship diagram, in which each node represents a component, and the connection between the nodes represents the direct interaction relationship between the components.

[0149] From the interaction logic description of each component, filter out the parts involving other components, such as "component A will be called after the operation of this component is completed" and "the output data of component B will be passed to this component", so as to determine the functional call relationship between components.

[0150] Each component is represented as a node. Lines are drawn between two nodes that have a direct interaction relationship to form a component interaction diagram. Lines can indicate the direction and type of interaction, such as one-way call or two-way data transfer.

[0151] Step S1442: Calculate the interaction frequency between components based on the component interaction relationship diagram. A larger interaction frequency value indicates a closer association between components.

[0152] Interaction frequency refers to the number of direct interactions between two components within a preset timeframe or based on historical interaction data. This is calculated by analyzing component interaction logs or simulating interaction scenarios to count the number of interactions between each component pair.

[0153] For example, if component A and component B have 10 direct interactions in one day, and component C and component D have 5 direct interactions, then the interaction frequency between component A and component B is greater than the interaction frequency between component C and component D, indicating that A and B are more closely associated.

[0154] Step S1443: According to the interactive proximity principle in the preset layout rule, component pairs whose interaction frequencies meet the preset threshold are marked as being placed adjacently.

[0155] The preset threshold is set based on the compactness of the page layout and the requirements for interaction efficiency. When the interaction frequency between two components is greater than or equal to the preset threshold, the two components are considered to be placed adjacently according to the interactive proximity principle and are marked as adjacent.

[0156] For example, if the preset threshold is 8 times and the interaction frequency between component A and component B is 10 times, they are marked as being placed adjacent to each other.

[0157] Step S1444: When allocating component positions within a functional area, component pairs with adjacent placement marks are prioritized and arranged within the same functional sub-area or adjacent functional sub-areas so that the physical distance between components is positively correlated with the interaction frequency.

[0158] Functional areas are further divided into functional sub-areas. Each functional sub-area is smaller, allowing for the centralized placement of closely related components. When assigning component locations, component pairs marked with adjacent placement are processed first and placed within the same functional sub-area. If space is insufficient, they are placed in adjacent functional sub-areas.

[0159] The higher the interaction frequency of component pairs, the closer the physical distance between them is, reflecting the positive correlation between physical distance and interaction frequency, thereby improving the convenience of user operation.

[0160] Step S1445: Calculate the length of the operation path between the adjacently placed component pairs. If the operation path length does not meet the preset standard, adjust the position parameters of the components until the operation path length meets the preset standard.

[0161] The action path length refers to the distance a user needs to move from one component to another after completing an action, typically measured in pixels on the screen. The default standard is a maximum allowable action path length.

[0162] For pairs of components placed adjacent to each other, the length of the operation path between them is calculated. If the length is greater than the preset standard, the position parameters of one or both components are fine-tuned, such as changing the coordinate values ​​to reduce the distance between them, and the operation path length is recalculated until it is less than or equal to the preset standard.

[0163] Step S1446: Record the final placement positions of the components and the corresponding interactive relationships as interactive association attributes of the initial functional page layout.

[0164] The final coordinate position of each component on the page, the functional area and functional sub-area to which it belongs, as well as the interaction relationship and adjacent placement between components are recorded to form the interaction association properties of the initial functional page layout. The above interaction association properties can reflect the spatial relationship and functional interaction relationship between components.

[0165] Step S145: globally coordinate the color parameters in the display properties of each component so that the color matching of different components conforms to the color system parameters in the style preference characteristics, forming a unified page color style.

[0166] Although the color properties of each component have been adjusted according to the color system parameters, there may be color inconsistencies between different components, so global coordination is required. First, the color parameters of all components, including the main color values ​​and auxiliary color values, are extracted and converted into a unified color space for comparison and adjustment.

[0167] Next, we verify each component's color parameters based on the color system parameters in the style preference feature, specifically the primary color range and secondary color matching rules. For components whose primary color values ​​fall outside the primary color range, we adjust their primary color values ​​to within the primary color range, ensuring that the visual difference between the adjusted colors and the original colors is within an acceptable range.

[0168] For auxiliary colors, check whether their brightness contrast and hue circle distance with the corresponding component's main color meet the auxiliary color matching rules. If not, adjust the auxiliary color value according to the rules. For example, if the brightness contrast is lower than the required value, increase the brightness difference between the auxiliary color and the main color. If the hue circle distance is not in compliance, adjust the hue of the auxiliary color so that its distance from the main color's hue circle is within the specified range.

[0169] During the adjustment process, consider the color coordination between adjacent components. For closely related functional components placed next to each other, ensure sufficient differentiation between their colors while avoiding jarring color clashes. For example, if two adjacent components both use the primary hue as their primary color, adjust their secondary colors or brightness to enhance differentiation.

[0170] After adjusting the color parameters of all components, review the overall color effect of the page to ensure that the overall color style of the page is consistent and meets the color system parameters of the style preference features. If there are any areas with inconsistent colors, make further fine-tuning until a unified and harmonious page color style is achieved.

[0171] Step S146: Integrate the placement, display size, color parameters and functional area divisions of all components to generate an initial functional page layout.

[0172] The information obtained through the above steps, such as component placement, display size, color parameters, and functional area division, is summarized. According to the hierarchical structure of the page layout, the position and range of each functional area on the page are first determined, and then the components within each functional area are arranged according to the final placement and display size.

[0173] At the same time, the color parameters of each component are applied to the corresponding component to ensure that the color display of the component conforms to the globally coordinated result. This information is integrated into a complete page layout data structure, which includes the overall size of the page, the division details of functional areas, the specific properties of each component (position coordinates, size, color parameters, etc.), and the interactive properties between components.

[0174] Through this integration process, an initial function page layout is generated, which can meet the user's functional requirements and style preferences and provide the user with a preliminary function page display.

[0175] Step S150: Monitor the interactive behavior data between the user and the initial function page in real time, input the interactive behavior data into the dynamic adjustment module of the large model, adjust the display properties and page layout structure of the page components based on the interactive behavior data, and obtain a personalized function page.

[0176] After the initial function page is displayed to the user, the system starts a real-time monitoring mechanism to capture the user's various interactive behaviors on the page. After the above interactive behavior data is collected, it is sent to the dynamic adjustment module of the large model for analysis and processing. Then, based on the analysis results, the display properties and layout structure of the page components are adjusted to make the page more in line with the user's actual usage habits and needs, and finally form a personalized function page.

[0177] Step S151: deploying a behavior monitoring component to collect real-time interaction behavior data between the user and the initial function page, wherein the interaction behavior data includes the number of component clicks, component dwell time, page scrolling track, and operation sequence.

[0178] Embed a behavior monitoring component in the code of the initial functional page. This component monitors various user action events on the page. When a user clicks a component, the monitoring component records the component's ID and the time the click occurred, accumulating the number of component clicks. When the user's mouse or touch focus remains on a component, the monitoring component records the start and end times of the stay, calculating the duration of the stay.

[0179] For page scrolling, the monitoring component listens to scroll events and records information about page position changes during the scrolling process, including the scrolling direction, speed, and page area reached, generating scrolling trajectory data. The action sequence records a series of actions performed by the user on the page in chronological order, such as clicking a component, scrolling the page, and then clicking another component. Each action includes information such as the action type, the components involved, and the time of the action.

[0180] The collected interactive behavior data is transmitted in real time to the data storage module for temporary storage and undergoes preliminary format conversion to meet the data format requirements for subsequent processing. During the data collection process, interactive behavior data involving user privacy is processed using data desensitization technology. For example, the user's operation time is obscured to remove sensitive information that may identify the user, ensuring that data collection complies with privacy protection requirements.

[0181] Step S152: pre-process the interaction behavior data, convert the number of component clicks into a click frequency parameter, convert the component dwell time into a dwell ratio parameter, convert the page scrolling track into an area attention parameter, and convert the operation sequence into an interaction path parameter.

[0182] For component clicks, count the number of clicks within a preset time interval and divide it by the time interval to get the number of clicks per unit time, which is the click frequency parameter. This click frequency parameter can reflect the user's activeness in clicking on the component.

[0183] The preprocessing of component dwell time is to divide the dwell time of each component by the total dwell time of the user on the entire page to obtain a dwell ratio parameter. The dwell ratio parameter represents the proportion of the user's dwell time on the component, which can reflect the user's attention to the component.

[0184] When processing the page scrolling trajectory, the page is divided into multiple equal areas. The user's stay time in each area and the number of times the user scrolls through the area are counted. Based on this data, the attention score of each area is calculated to form the regional attention parameter. The higher the score, the more attention the area receives from users.

[0185] For the operation sequence, the components and page locations involved in each operation are extracted in chronological order, and this information is converted into nodes and paths. The nodes represent components or page areas, and the paths represent the sequence of operations and jump relationships, thereby forming interaction path parameters. The interaction path parameters can reflect the user's operation process and path on the page.

[0186] Each dimension of the preprocessed parameters corresponds to a different component or page area, and the value in the vector reflects the corresponding parameter value.

[0187] Step S153: Input the pre-processed click frequency parameters, dwell ratio parameters, regional attention parameters and interaction path parameters into the dynamic adjustment module of the large model to generate a user behavior feature vector.

[0188] The dynamic adjustment module of the large model receives the preprocessed parameters and inputs them into the input layer of the dynamic adjustment module. The input layer concatenates these vectors to form a comprehensive input vector that contains multiple aspects of user interaction behavior.

[0189] The integrated input vector is processed by multiple hidden layers in the dynamic adjustment module. The hidden layers utilize a fully connected neural network structure, applying nonlinear transformations to the input using activation functions to gradually extract the underlying characteristics of user behavior. The output of each hidden layer serves as the input for the next hidden layer. After multiple rounds of feature extraction and transformation, the output layer ultimately generates a user behavior feature vector.

[0190] The user behavior feature vector is a high-dimensional vector. Each dimension corresponds to a specific feature of user behavior, which can fully reflect the user's behavioral habits and preferences when using the initial function page.

[0191] Step S154: Analyze the user behavior feature vector through the demand change identification layer of the dynamic adjustment module to determine whether there is a deviation between the user's current actual demand and the initial functional demand characteristics. If there is a deviation, generate a demand adjustment coefficient.

[0192] The demand change identification layer includes a feature comparison unit and a coefficient generation unit. The feature comparison unit receives the user behavior feature vector and the initial functional requirement feature, and converts them into the same feature space for comparison and analysis.

[0193] For example, step S1541: extract the priority ranking in the initial functional requirement characteristics as the benchmark priority sequence, and convert the click frequency parameter and the stay ratio parameter in the user behavior feature vector into an actual attention sequence.

[0194] The priority ranking is extracted from the initial functional requirement characteristics. The priority ranking reflects the importance of each functional type initially expected by the user and is converted into a numerical benchmark priority sequence. Each element in the sequence corresponds to the initial importance score of a functional type.

[0195] For the user behavior feature vector, the click frequency parameter and the stay ratio parameter are extracted, and these two parameters are weightedly fused. The weights are set according to the degree of influence of the two parameters on user attention. After fusion, the actual attention score corresponding to each function type is obtained to form an actual attention sequence.

[0196] Step S1542: Calculate the deviation value between the reference priority sequence and the actual attention sequence, where the deviation value is represented by the sum of the squares of the differences between the elements at corresponding positions in the sequence.

[0197] Subtract the corresponding elements in the baseline priority sequence from the actual attention sequence, obtaining the difference at each position. Square each difference and add all the squared values ​​together to obtain the deviation value. The magnitude of this deviation value reflects the degree of difference between the user's current actual needs and the initial functional requirements.

[0198] Step S1543: Compare the deviation value with a preset deviation threshold. If the deviation value is greater than the preset deviation threshold, it is determined that there is a deviation between the user's current actual demand and the initial functional demand characteristics.

[0199] The preset deviation threshold is a critical value determined based on extensive user behavior data and page adjustment experience, used to determine whether the deviation is significant. If the calculated deviation value is greater than this threshold, it indicates that the user's actual behavior reflects a significant difference from the initial demand, and the page needs to be adjusted. If the deviation value is less than or equal to the preset deviation threshold, the user's current demand is considered to be basically consistent with the initial demand, and no adjustment is required.

[0200] Step S1544: Determine the value of the demand adjustment coefficient based on the size of the deviation value. The larger the deviation value, the larger the value of the demand adjustment coefficient. The demand adjustment coefficient is used to represent the amplitude of the adjustment.

[0201] Set a mapping relationship between the demand adjustment coefficient and the deviation value. This mapping relationship ensures that the larger the deviation value, the larger the demand adjustment coefficient value. For example, when the deviation value is within a certain range, the demand adjustment coefficient takes a value within the corresponding range and increases monotonically as the deviation value increases. This ensures that the larger the deviation value, the larger the page adjustment, better meeting the user's actual needs.

[0202] Step S1545: Determine the adjustment direction of the demand adjustment coefficient based on the difference direction between the actual attention sequence and the baseline priority sequence. If the element value in the actual attention sequence is greater than the corresponding element value in the baseline priority sequence, the adjustment direction is to enhance the properties of the corresponding component. If the element value in the actual attention sequence is less than the corresponding element value in the baseline priority sequence, the adjustment direction is to weaken the properties of the corresponding component.

[0203] Compare the element values ​​of the corresponding positions in the actual attention sequence and the baseline priority sequence one by one. When the element value in the actual attention sequence is greater than the corresponding element value in the baseline priority sequence, it means that the user's actual attention to the function type corresponding to the element is higher than the initial expectation. Therefore, the adjustment direction is to enhance the properties of the component corresponding to the function type; conversely, when the element value in the actual attention sequence is less than the corresponding element value in the baseline priority sequence, the adjustment direction is to weaken the properties of the corresponding component.

[0204] Step S1546: Integrate the value and adjustment direction of the demand adjustment coefficient to generate a demand adjustment coefficient for guiding the adjustment of component display attributes.

[0205] The value and adjustment direction of the demand adjustment coefficient are combined into a structured parameter. This parameter contains both the adjustment amplitude and adjustment direction information, which can clearly guide subsequent adjustments to the component display properties.

[0206] Step S155: Adjust the display properties of the page components based on the demand adjustment coefficient, increase the size ratio of the components whose click frequency parameters meet the first preset conditions, enhance the color contrast of the components whose dwell ratio parameters meet the second preset conditions, and optimize the component layout of the area whose regional attention parameters meet the third preset conditions.

[0207] According to the adjustment range and direction indicated by the demand adjustment coefficient, the display properties of the page components are adjusted in a targeted manner to improve the matching degree between the page and the actual needs of users.

[0208] For example, step S1551: For components whose click frequency parameters meet the first preset conditions, the size ratio of the component is increased according to the value and adjustment direction of the demand adjustment coefficient. The increase is determined by the relative value of the click frequency and the demand adjustment coefficient, and does not exceed the maximum allowable size ratio of the component.

[0209] The first pre-set condition is that the click frequency parameter is higher than a pre-set click frequency threshold, which is determined based on the average click frequency of components on the page. For components that meet this condition, the relative click frequency value is calculated, that is, the ratio of the component's click frequency to the average click frequency of all components.

[0210] Multiply the relative click frequency value by the demand adjustment coefficient to determine the increase in size share. Based on the adjustment direction (enhancement), add the increase to the component's current size share to determine the adjusted size share. Ensure that the adjusted size share does not exceed the maximum allowable size share for the component in the page layout. If it does, the maximum allowable size share will be used as the adjusted size share.

[0211] Step S1552: For components whose dwell ratio parameters meet the second preset conditions, the color contrast of the component is improved according to the value and adjustment direction of the demand adjustment coefficient. The adjustment of the color contrast is achieved by increasing the brightness difference between the main color and the background color. The degree of improvement in the color contrast is determined by the dwell ratio parameter and the demand adjustment coefficient, and remains in compliance with the color matching rules.

[0212] The second pre-set condition is that the dwell ratio parameter exceeds a preset dwell ratio threshold, which is determined based on the average dwell ratio of all components. For components that meet this condition, their dwell ratio parameter is multiplied by the value of the demand adjustment coefficient to determine the improvement in color contrast.

[0213] Improve color contrast by increasing the brightness difference between the component's primary color and the background color. This can be done by reducing the background color's brightness or increasing the primary color's brightness, depending on the amount of contrast. During this adjustment, always refer to the color matching rules in the Style Preferences feature to ensure that the adjusted color contrast meets these rules and avoids color clashes.

[0214] Step S1553: Analyze the regional attention parameters, determine the area in the page where the user attention meets the third preset condition, and adjust the layout of the components in the area to be compact, reduce the blank area between the components, and maintain the interactive relationship between the components.

[0215] The third preset condition is that the regional attention parameter is higher than a preset attention threshold, which is determined based on the average attention of each area of ​​the page. For areas that meet this condition, the current layout of the components in the area is analyzed and the size of the blank area between the components is measured.

[0216] While maintaining the interaction and non-overlapping relationships between components, fine-tune the position of components to reduce the blank space between components and make the component layout more compact. For example, appropriately reduce the spacing between adjacent components or adjust the arrangement of components to make full use of the area.

[0217] Step S1554: For components whose element values ​​in the actual attention sequence are smaller than the corresponding element values ​​in the baseline priority sequence, reduce the size ratio of the component or lower its color contrast according to the value and direction of the required adjustment coefficient to reduce its visual weight on the page.

[0218] For these components, the required size adjustment coefficient determines the reduction in size or color contrast. When reducing size, subtract the reduction from the current size to ensure that the adjusted size is no less than the minimum size required for normal component display. When reducing color contrast, reduce the brightness difference between the primary color and the background color to make the component less prominent on the page, thereby reducing its visual weight.

[0219] Step S1555: After the adjustment is completed, the display properties of each component are checked for consistency, so that the adjusted size ratio and color parameters conform to the overall style of the style preference characteristics to avoid visual conflicts.

[0220] Check whether the adjusted component size ratios are within the allowed range of the page layout and whether the size ratios between components are coordinated. For color parameters, check whether they comply with the color system parameters in the style preference features and whether the color combinations between components are harmonious, without any glaring or inharmonious effects.

[0221] If there are any situations that do not meet the requirements, fine-tune the display properties of the corresponding components again until the display properties of all components meet the requirements, ensuring that the overall style of the page is unified and there is no visual conflict.

[0222] Step S156: Analyze the fluency of the user operation according to the interaction path parameters. If the operation path length meets the fourth preset condition, adjust the position parameters of the relevant components, optimize the page layout structure, and shorten the interaction path.

[0223] By analyzing the interaction path parameters, the smoothness of user operations is evaluated. If the operation path is too long, the component position is adjusted to optimize the page layout and improve user operation efficiency.

[0224] Step S1561: parse the interaction path parameters, extract the sequence of steps for the user to complete a specific operation task on the page and the jump tracks between the steps, and generate an operation path diagram.

[0225] Identify the specific operation tasks completed by the user from the interaction path parameters, such as "submit the assignment and view the grade", and extract the step sequence corresponding to the task. Each step corresponds to an operation, such as "click the assignment submission component", "fill in the assignment information", "click the submit button", "click the grade query component", etc.

[0226] Based on the locations of the components involved in each step, the jump paths between the steps are drawn to form an operation path diagram. In the operation path diagram, nodes represent steps and lines represent jump paths. The length of the line reflects the size of the jump distance.

[0227] Step S1562: Calculate the jump distance between each step in the operation path diagram. The jump distance is represented by the pixel distance between the center points of the components. Convert the jump distance to a proportional distance relative to the overall size of the page. Accumulate the proportional distances between all steps to obtain the total length of the operation path.

[0228] For each jump track in the operation path diagram, determine the center point coordinates of the components involved in the steps at both ends of the track, and calculate the straight-line pixel distance between the two points, which is the jump distance.

[0229] Using the overall width and height of the page as a reference, convert the jump distance to a proportional distance: divide the jump distance by the overall page size (width or height, depending on the jump direction). Add the proportional distances between all steps to obtain the total action path length, which reflects the length of the action path required for users to complete a specific task.

[0230] Step S1563: Compare the total length of the operation path with a preset path threshold. If the total length of the operation path is greater than the preset path threshold, it is determined that the operation path is too long.

[0231] The preset path threshold is determined based on the average path length of users completing similar tasks. When the total path length exceeds this threshold, it indicates that the path to complete the task is too long and the operation is not smooth enough, and the position of the relevant components needs to be adjusted.

[0232] Step S1564: Identify the continuous steps with the longest jump distance in the operation path, and determine the relevant components involved in the continuous steps.

[0233] In the operation path diagram, identify two or more consecutive steps with the longest jump distance. These steps constitute the bottleneck part of the operation path. Determine the components involved in these steps and identify the relevant components that need to be adjusted.

[0234] Step S1565: Based on the interactive proximity principle in the preset layout rule, position parameters of related components are adjusted to place components involved in consecutive steps with the longest jump distance adjacent to each other, thereby reducing the jump distance between them.

[0235] Based on the principle of interactive proximity, adjust the positions of related components to bring them as close as possible. For example, if the jump distance between two components is the longest, you can move them to adjacent positions on the page or within the same functional sub-area to shorten the jump distance between them.

[0236] During the adjustment process, it is necessary to ensure that the movement of the component does not affect the normal layout and function of other components, while maintaining the interactive relationship between the component and other related components.

[0237] Step S1566: recalculate the adjusted total length of the operation path. If the total length of the operation path is still greater than the preset path threshold, repeatedly adjust the position parameters of the relevant components until the total length of the operation path is less than or equal to the preset path threshold.

[0238] After adjusting the positions of the relevant components, the total length of the operation path is recalculated according to the method in step S1562 and compared with the preset path threshold. If it still does not meet the requirements, the position parameters of the relevant components are adjusted again. If necessary, the adjustment range can be expanded to include more relevant components to ensure that the total length of the operation path meets the preset path threshold. During the repeated adjustment process, the interactive relationships between components must be maintained, while taking into account the rationality and aesthetics of the overall page layout to avoid page structure chaos caused by excessive adjustments.

[0239] Step S1567: Record the adjusted component position parameters and update the page layout structure to optimize the user operation path.

[0240] After multiple adjustments, the component position parameters that meet the requirements are recorded in detail, including each component's specific coordinates, size ratio, and other information. Based on these records, the page layout structure is updated, re-determining the placement and display size of each component on the page. The updated page layout structure can effectively shorten the user's operation path, improve the smoothness and efficiency of user operations, and make the personalized function page more in line with the user's actual usage habits.

[0241] After the processing of step S150, the initial function page is dynamically adjusted based on the user's interactive behavior data, and finally a function page that meets the user's personalized needs is obtained. This function page not only meets the user's initial function requirements and style preferences, but is also optimized based on the user's actual operation behavior, thereby improving the user experience.

[0242] The entire large-model-based method for generating personalized functional pages also involves protecting user privacy-sensitive data. When collecting privacy-sensitive data, such as user interaction behavior data, data desensitization technology is employed to anonymize users' personal identification information and remove information that can be directly linked to their identities. Furthermore, encrypted transmission technology is used to prevent data leakage during transmission. Access control mechanisms are implemented for data storage, ensuring that only authorized personnel can access the relevant data. Data access behavior is logged for security audits. These technical measures effectively protect user privacy and prevent the leakage of privacy-sensitive data.

[0243] Figure 2 The following diagram illustrates exemplary hardware and software components of a system 100 for generating a personalized functional page based on a large model, which can implement the concepts of the present application, according to some embodiments of the present application. For example, the processor 120 can be used in the system 100 for generating a personalized functional page based on a large model and for executing the functions of the present application.

[0244] For example, the large-scale model-based personalized function page generation system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the large-scale model-based personalized function page generation system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The large-scale model-based personalized function page generation system 100 also includes an I / O interface 150 between the computer and other input and output devices.

[0245] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned method for generating personalized function pages based on a large model is implemented.

[0246] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for generating personalized function pages based on a large model, characterized in that: The method comprises: Obtaining a function page requirement description input by a user in natural language, wherein the function page requirement description includes the user's expectations for page functions and preferences for page presentation style; Calling the requirements parsing module of the large model to perform semantic parsing on the functional page requirements description to obtain the user's functional requirements characteristics and style preference characteristics. The functional requirements characteristics reflect the user's expected page function type and interaction logic, and the style preference characteristics reflect the user's expected page color matching, font style and layout structure; Searching a preset component library based on the functional requirement characteristics to obtain a page component set corresponding to the functional requirement, and determining display attributes of each component in the page component set based on the style preference characteristics, wherein the display attributes include component size, position ratio, and color parameters; Generate an initial functional page layout based on the page component set, display properties of each component and preset layout rules; Monitor the interactive behavior data between the user and the initial function page in real time, input the interactive behavior data into the dynamic adjustment module of the large model, adjust the display properties and page layout structure of the page components based on the interactive behavior data, and obtain a personalized function page.

2. The method for generating personalized function pages based on a large model according to claim 1, characterized in that: The step of obtaining the function page requirement description input by the user in natural language includes: Receive a natural language requirement description submitted by the user through an input interface, wherein the input interface supports text input, voice-to-text input, and image-attached text description input; Identify the identity type of the input user, which includes student, teacher, and parent. Users with different identity types correspond to different requirements description emphases; For users with student identity, extract the functional expectation expressions related to learning content from the natural language demand descriptions they input; for users with teacher identity, extract the functional expectation expressions related to teaching management from the natural language demand descriptions they input; for users with parent identity, extract the functional expectation expressions related to the display of students' learning status from the natural language demand descriptions they input; Redundant information is filtered out of the extracted demand expressions of users with different identity types, and the core demand descriptions are retained as the function page demand descriptions input by users through natural language.

3. The method for generating personalized function pages based on a large model according to claim 1, characterized in that: The requirement parsing module of the large model is called to perform semantic parsing on the function page requirement description to obtain the user's function requirement characteristics and style preference characteristics, including: Input the function page requirement description into the word vector conversion layer of the large model to generate a corresponding requirement word vector sequence, where each vector in the requirement word vector sequence corresponds to a vocabulary unit in the requirement description; Calling the context encoding module of the large model to perform semantic association modeling on the demand word vector sequence to generate an encoding vector containing context dependency; Performing feature splitting processing on the coding vector through the feature separation layer of the large model to separate the functional coding sub-vector related to functional requirements and the style coding sub-vector related to style preferences; Performing function dimension mapping processing on the function coding sub-vector to generate a function requirement feature including a function type identifier, an interaction logic description, and a priority ranking. The function type identifier is used to distinguish different page function categories, the interaction logic description is used to illustrate the calling relationship between functions, and the priority ranking is used to indicate the importance of each function. The style encoding sub-vector is subjected to style dimension mapping processing to generate a style preference feature including color system parameters, font attribute parameters, and layout mode parameters. The color system parameters reflect the user's preference for the primary and secondary colors of the page, the font attribute parameters reflect the user's preference for the font type and size, and the layout mode parameters reflect the user's preference for the arrangement of page elements.

4. The method for generating a personalized function page based on a large model according to claim 3, characterized in that: The context encoding module of the calling large model performs semantic association modeling on the demand word vector sequence to generate an encoding vector containing context dependency, including: Input the demand word vector sequence into the multi-head attention layer of the context encoding module, and calculate the attention weight value between each vocabulary unit and other vocabulary units, wherein the attention weight value is used to represent the semantic association strength between vocabulary units; Performing weighted aggregation processing on the demand word vector sequence based on the attention weight value to generate a context enhancement vector for each vocabulary unit, wherein the context enhancement vector incorporates semantic information of other vocabulary units whose association with the vocabulary unit meets a preset association standard; Input the context enhancement vector into the bidirectional recurrent layer for temporal association modeling to generate a temporal encoding vector containing the semantic dependencies of the previous and next contexts, where each dimension of the temporal encoding vector corresponds to the semantic feature of a position in the demand word vector sequence; Performing layer normalization processing on the temporal coding vector to obtain a standardized temporal coding vector; The standardized temporal coding vector is dimensionally converted through the output layer of the context coding module to generate a coding vector with the same length as the required word vector sequence, and each element of the coding vector contains the contextual semantic information of the corresponding vocabulary unit.

5. The method for generating personalized function pages based on a large model according to claim 3, characterized in that: The function dimension mapping process is performed on the function coding sub-vector to generate a function requirement feature including a function type identifier, an interaction logic description, and a priority ranking, including: Calling a function classifier to perform category determination on the function coding sub-vector and outputting a corresponding function type identifier, wherein the function type identifier includes information display class, operation interaction class, data processing class and feedback response class; Extracting interaction-related semantic features from the function encoding sub-vector to generate an interaction logic description, wherein the interaction logic description includes a function triggering condition, a function execution process, and a function result return method. The function triggering condition is used to describe the user operation required to activate the function, the function execution process is used to describe the processing steps within the function, and the function result return method is used to describe the presentation format of information after the function is executed. Calculating the importance score of each function type based on the semantic strength value in the function coding sub-vector, wherein the semantic strength value is determined by the projection length of the function coding sub-vector in the semantic activation direction, and a larger projection length indicates a higher degree of emphasis on the corresponding function in the requirement description; Sort each function type according to the importance score to generate a priority ranking, wherein the function type with a higher sequence position in the priority ranking should be displayed first when the page is generated; The function type identification, interaction logic description and priority sorting are combined according to a preset format to generate the user's function requirement characteristics.

6. The method for generating personalized function pages based on a large model according to claim 1, characterized in that: The searching of a preset component library based on the functional requirement characteristics to obtain a page component set corresponding to the functional requirement by matching, and determining the display attributes of each component in the page component set based on the style preference characteristics includes: Extract the function type identifier and priority ranking from the function requirement characteristics, use the function type identifier as a search keyword, and search the preset component library; the preset component library stores multiple component units, each component unit includes a function matching tag, basic structure parameters, and a configurable property set, the function matching tag is used to match the function type identifier, the basic structure parameters are used to describe the core function implementation method of the component, and the configurable property set is used to represent the adjustable display property range of the component; Calculate the semantic similarity between the functional type identifier and the functional matching label of the component unit, and select the component units whose semantic similarity meets the preset conditions as candidate components; Convert the priority ranking into a numerical weight coefficient, and use it together with the semantic similarity to determine the component screening order. Perform weighted sorting on the candidate components, and select the top K candidate components in the sorting sequence to form a page component set that corresponds to the functional requirements. Extracting color system parameters, font attribute parameters, and layout mode parameters from the style preference features as a basis for configuring component display attributes; For each component in the page component set, adjust the component's color properties according to the color system parameters, adjust the component's text display properties according to the font property parameters, and adjust the component's size ratio and position parameters according to the layout mode parameters to generate display properties for each component, including component size, position ratio, and color parameters. The page component set is associated and bound with the display attributes of each component so that the display attributes of each component are adapted to the priority ranking of its functional requirements.

7. The method for generating personalized function pages based on a large model according to claim 1, characterized in that: Generating the initial functional page layout according to the page component set, the display attributes of each component and the preset layout rules includes: Extract the function type identification and priority ranking of each component in the page component set, and divide the page into multiple functional areas based on the functional area division principle in the preset layout rules. Each functional area corresponds to one or more function type identifications. According to the priority ranking of each component, the K components at the top of the priority ranking sequence are assigned to the functional areas on the page whose visual weights meet the first preset level, and the M components at the bottom of the priority ranking sequence are assigned to the functional areas on the page whose visual weights meet the second preset level. The visual weights are determined by the position and size of the functional areas on the page, and the visual weights of the first preset level are greater than the visual weights of the second preset level. Based on the size ratio and position parameters in the display properties of each component, the specific placement and display size of each component is determined in the corresponding functional area to ensure that there is no overlap between components; Analyze the functional relationships between components and place directly interactive components adjacent to each other based on the interactive proximity principle in the preset layout rules, shortening the user's operation path. Globally coordinate the color parameters in the display properties of each component so that the color matching of different components conforms to the color system parameters in the style preference characteristics, forming a unified page color style; Integrate the placement, display size, color parameters and functional area division of all components to generate the initial functional page layout.

8. The method for generating personalized function pages based on a large model according to claim 1, characterized in that: The real-time monitoring of the user's interaction behavior data with the initial function page, inputting the interaction behavior data into the dynamic adjustment module of the large model, and adjusting the display attributes and page layout structure of the page components based on the interaction behavior data to obtain a personalized function page, including: Deploy a behavior monitoring component to collect real-time data on user interaction with the initial function page. The interaction behavior data includes the number of component clicks, component dwell time, page scrolling trajectory, and operation sequence. Preprocess the interaction behavior data to convert the number of component clicks into a click frequency parameter, the component dwell time into a dwell ratio parameter, the page scrolling trajectory into a regional attention parameter, and the operation sequence into an interaction path parameter; The pre-processed click frequency parameters, dwell ratio parameters, regional attention parameters, and interaction path parameters are input into the dynamic adjustment module of the large model to generate a user behavior feature vector; The demand change identification layer of the dynamic adjustment module analyzes the user behavior feature vector to determine whether there is a deviation between the user's current actual demand and the initial functional demand characteristics. If there is a deviation, a demand adjustment coefficient is generated; Adjust the display properties of page components based on the demand adjustment coefficient, increase the size ratio of components whose click frequency parameter meets the first preset condition, enhance the color contrast of components whose dwell ratio parameter meets the second preset condition, and optimize the component layout of areas whose regional attention parameter meets the third preset condition; Analyze the fluidity of user operations based on interaction path parameters. If the length of the operation path meets the fourth preset condition, adjust the position parameters of related components, optimize the page layout structure, and shorten the interaction path. Integrate the adjusted page component display properties and layout structure to generate personalized function pages.

9. The method for generating personalized function pages based on a large model according to claim 8, characterized in that: Analyzing the fluency of user operations based on interaction path parameters, and if the operation path length meets the fourth preset condition, adjusting the position parameters of relevant components, optimizing the page layout structure, and shortening the interaction path, includes: Parse the interaction path parameters, extract the sequence of steps that users take to complete specific operation tasks on the page and the jump tracks between each step, and generate an operation path diagram; Calculate the jump distance between each step in the operation path diagram. The jump distance is expressed as the pixel distance between the center points of the components. Convert the jump distance to a proportional distance relative to the overall size of the page. Accumulate the proportional distances between all steps to obtain the total length of the operation path. Comparing the total length of the operation path with a preset path threshold, and if the total length of the operation path is greater than the preset path threshold, determining that the operation path is too long; Identifying the consecutive steps with the longest jump distance in the operation path, and determining the relevant components involved in the consecutive steps; Based on the interactive proximity principle in the preset layout rules, the position parameters of related components are adjusted to place the components involved in the consecutive steps with the longest jump distance adjacent to each other to reduce the jump distance between them; Recalculating the adjusted total length of the operation path; if the total length of the operation path is still greater than the preset path threshold, repeatedly adjusting the position parameters of the relevant components until the total length of the operation path is less than or equal to the preset path threshold; Record the adjusted component position parameters and update the page layout structure to optimize the user operation path.

10. A system for generating personalized function pages based on a large model, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the large model-based personalized function page generation method described in any one of claims 1 to 9.

Images