Form page construction management method and system based on AI large model dialogue generation

Abstract

The application provides a form page construction management method and system based on AI large model dialogue generation, relating to the technical field of human-computer interaction, the method comprising: step 1, performing intent analysis on user intent description through a pre-trained neural network model, identifying the user intent type, and extracting form field elements, the form field elements including field name, data type and verification rule; step 2, constructing a form layout virtual plane based on the form field elements, and establishing a basic layout framework on the form layout virtual plane; setting layout coordinate points in the basic layout framework; according to the data characteristics of the form field elements, the basic layout framework is divided into multiple field distribution areas. The application realizes form page construction and business process automatic execution based on dialogue interaction by intelligently analyzing user intent, optimizing form layout, matching components and automatically processing business processes, improving form construction efficiency and business processing standardization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of human-computer interaction, in particular to a form page construction management method and system based on AI large model dialogue generation. BACKGROUND

[0002] In the existing information system development practice, the construction of the form page usually depends on the manual code writing of the developer or the configuration using the graphical interface tool. Such method may face some challenges when dealing with the dynamic business demand expressed through the natural language dialogue. One problem worth paying attention to is how to efficiently and reasonably convert the user intent described in the natural language into a structured form page. When the user proposes the demand in the form of dialogue, the existing technical solution may still have some deficiencies in the process of automatic understanding and conversion. In particular, if only the traditional neural network model is used for intent analysis, the output may focus more on identifying the basic form field, and the ability in deeply analyzing the complex logical association between the fields, the data constraint and the overall layout planning may be limited.

[0003] For example, for the demand of "employee leave form", the field layout of the form generated by the existing technology may not fully consider the logical association between the leave type and the start and end time, thereby possibly affecting the convenience and intuitiveness of the final user in filling out the form. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a form page construction management method and system based on AI large model dialogue generation, and to improve the efficiency of the whole business process.

[0005] To solve the above technical problems, the technical solution of the present application is as follows:

[0006] In a first aspect, a form page construction management method based on AI large model dialogue generation, the method comprising:

[0007] Step 1: performing intent analysis on the user intent description through a pre-trained neural network model, identifying the user intent type, and extracting form field elements, the form field elements including field name, data type and verification rule;

[0008] Step 2: constructing a form layout virtual plane based on the form field elements, and establishing a basic layout framework on the form layout virtual plane; setting layout coordinate points in the basic layout framework; dividing the basic layout framework into multiple field distribution areas according to the data characteristics of the form field elements;

[0009] Step 3: mapping the form field elements to the field distribution areas, and analyzing the distribution density and association of the fields in each field distribution area after mapping to obtain an analysis result;

[0010] Step 4, according to the analysis result, generate layout optimization parameters, and adjust the basic layout framework based on the layout optimization parameters to obtain the optimized basic layout framework;

[0011] Step 5, transmit the user intent type, form field elements and the optimized layout framework to the intermediate code platform, and process through the preset component matching model in the intermediate code platform to obtain the form component description;

[0012] Step 6, generate a form page based on the form component description, and display the form page in the dialogue interface; receive the form data submitted by the user through the form page, and automatically process the form data through the automatic process engine to realize the execution of the business process based on the dialogue interaction.

[0013] The second aspect is a form page construction management system generated based on intelligent dialogue, comprising:

[0014] The acquisition module is used for analyzing the user intent description through the pre-trained neural network model, identifying the user intent type, and extracting the form field elements, wherein the form field elements include field name, data type and verification rule; a form layout virtual plane is constructed based on the form field elements, and a basic layout framework is established on the form layout virtual plane; layout coordinate points are set in the basic layout framework; the basic layout framework is divided into a plurality of field distribution areas according to the data characteristics of the form field elements; the form field elements are mapped to the field distribution areas, and the distribution density and correlation of the fields in each field distribution area after mapping are analyzed to obtain an analysis result;

[0015] The processing module is used for generating layout optimization parameters according to the analysis result, and adjusting the basic layout framework based on the layout optimization parameters to obtain the optimized basic layout framework; transmitting the user intent type, form field elements and the optimized layout framework to the intermediate code platform, and processing through the preset component matching model in the intermediate code platform to obtain the form component description; generating a form page based on the form component description, and displaying the form page in the dialogue interface; receiving the form data submitted by the user through the form page, and automatically processing the form data through the automatic process engine to realize the execution of the business process based on the dialogue interaction.

[0016] The third aspect is a computer readable storage medium, wherein the computer readable storage medium stores a program, and the program is executed by a processor to realize the method.

[0017] The above-mentioned scheme of the application has at least the following beneficial effects:

[0018] The pre-trained neural network model is used for in-depth analysis of user intent description, which can accurately identify the intent type and extract the field name, data type and check rule and other key elements, solve the form design defects caused by manual understanding deviation, realize efficient and accurate conversion from natural language to form elements, build a virtual layout plane, divide the field distribution area, and dynamically optimize the layout based on the field density and correlation, so that the form fields are orderly distributed according to the data characteristics and business logic, avoiding the problems of layout confusion and logic fragmentation, and improving the convenience and intuitiveness of users filling out the form; The dialogue interaction, form generation, data submission and process processing are integrated, and the user can complete the form creation, filling and business submission through the dialogue, and the process engine automatically triggers the subsequent processing, breaking the separation of communication and form operation in business handling, and improving the efficiency of the whole business process; Through the component matching model of the code platform, the standardization of the form component calling and combination is realized, the needs of government affairs, enterprises and other scenes are met, and new components and form types can be easily expanded according to the business changes. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of a form page construction management method based on AI large model dialogue generation provided by an embodiment of the present application.

[0020] Figure 2 is a schematic diagram of a form page construction management system based on AI large model dialogue generation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0021] The exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0022] As Figure 1 shown, an embodiment of the present application proposes a form page construction management method based on AI large model dialogue generation, which comprises the following steps:

[0023] Step 1: Analyzing the user intent description by a pre-trained neural network model, identifying the user intent type, and extracting the form field elements, including the field name, data type and check rule;

[0024] Step 2: Building a form layout virtual plane based on the form field elements, and establishing a basic layout framework on the form layout virtual plane; setting layout coordinate points in the basic layout framework; according to the data characteristics of the form field elements, the basic layout framework is divided into multiple field distribution areas;

[0025] Step 3, map the form field elements to the field distribution areas, and analyze the distribution density and relevance of the fields in each field distribution area after mapping to obtain an analysis result;

[0026] Step 4, generate layout optimization parameters according to the analysis result, and adjust the basic layout framework based on the layout optimization parameters to obtain an optimized basic layout framework;

[0027] Step 5, transmit the user intent type, form field elements and optimized layout framework to the middle code platform, and process them through the pre-installed component matching model in the middle code platform to obtain a form component description;

[0028] Step 6, generate a form page based on the form component description, and display the form page in the dialogue interface; receive the form data submitted by the user through the form page, and automatically process the form data through the automated process engine to realize the execution of the business process based on the dialogue interaction.

[0029] In the embodiment of the present application, the pre-trained neural network model is used to deeply analyze the user intent description, which can accurately identify the intent type and extract key elements such as field name, data type and verification rules, solve the form design defects caused by human understanding bias, and realize efficient and accurate conversion from natural language to form elements; by constructing a virtual layout plane, dividing field distribution areas, and dynamically optimizing the layout based on field density and relevance, the form fields are distributed in order according to data characteristics and business logic, avoiding layout confusion and logical fragmentation, and improving the convenience and intuitiveness of filling out the form; integrating dialogue interaction, form generation, data submission and process processing, the user can complete form creation, filling and business submission through dialogue, and the process engine automatically triggers subsequent processing, breaking the separation of communication and form operation in business handling, and improving the efficiency of the whole business process; through the docking of the component matching model of the middle code platform, the standardized calling and combination of form components are realized, meeting the needs of government affairs, enterprises and other scenarios, and facilitating the expansion of new components and form types according to business changes.

[0030] In a preferred embodiment of the present application, before step 1, the form generation request input by the user through the natural language dialogue interface is received, and the request contains user intent description and demand information of business type, and the business type includes enterprise employee leave approval or engineering project declaration in government affair item declaration, specifically including: the system deploys the natural language dialogue interface supporting text input and voice-to-text function, and monitors the form generation request initiated by the user in real time, when the user inputs the request, the interface performs structured processing on the request data, and splits out the user intent description and the business type information, wherein the business type includes enterprise employee leave approval, engineering project declaration in government affair item declaration and the like, and then the structured request data is transmitted to the backend processing unit, and the data format is standardized and checked to ensure that it meets the input specification of the subsequent pre-trained neural network model; the step 1 includes:

[0031] At step 100, the user intent description is processed by a pre-trained neural network model for word segmentation and semantic role labeling to obtain a semantic role labeling result, which specifically includes: first, collecting form generation related corpus in the government and enterprise office scenarios, the corpus sources cover two aspects, one is the form application natural language demand text disclosed by the government service platform, such as applying for an engineering project construction permit form, submitting an employee maternity leave approval form, and the other is the dialogue record of historical form application in the enterprise OA system, finally 100,000 pieces of corpus are collected (here the actual scenario common order of magnitude is taken as an example), and each piece of corpus needs to contain three parts of core information, i.e. natural language demand description such as creating an employee 3-day sick leave form, corresponding business type such as enterprise employee leave approval, government engineering project declaration, and labeled semantic role data such as create-action role, employee-subject role, 3-day-quantity role, sick leave-type role, and form-target role. The core use of this batch of corpus is to fine-tune the pre-trained neural network model, wherein the pre-trained neural network model specifically adopts an AI large language model adapted to the Chinese form generation scenario, and the selected model is a model optimized based on the BERT-Base-Chinese architecture, which contains 12 layers of encoder, 768-dimensional hidden layer, and 12 heads of attention mechanism, and is naturally adapted to Chinese semantic understanding. Through corpus fine-tuning, the model can strengthen the semantic understanding depth of leave approval, project declaration and other natural languages, and the labeling accuracy of action role, quantity role and other semantic roles; perform a standardization cleaning process on the 100,000 pieces of collected corpus, first remove special symbols and meaningless placeholders in the text by using regular expressions, second filter invalid texts with a length less than 5 characters, such as incomplete descriptions of only forms, leave forms, third perform text deduplication and corpus division; when performing text deduplication, first split the natural language description of each piece of corpus into words, convert the words into 300-dimensional numerical vectors through the Word2Vec tool, and the vectors can reflect the semantic association of the words, for example, the vectors of leave and vacation are more similar, convert them into 300-dimensional vectors, and then take the average of all word vectors as the vector of the whole corpus. Calculate the cosine similarity of the vectors of any two pieces of corpus, when the similarity is greater than or equal to 0.95, it is determined that the corpus is repeated and only one piece is retained, and finally 80,000 pieces of valid corpus are obtained; divide the 80,000 pieces of valid corpus into a training set and a validation set in a ratio of 8:2, wherein the number of the training set is 64,000 (8x0.8), which is used for model parameter iterative update, and the number of the validation set is 16,000 (8x0.2), which is used for monitoring the generalization ability of the model in the training process.

[0032] With the AI large language model (i.e. pre-trained neural network model) as the initial framework, two aspects of structural optimization are carried out to adapt to the semantic role labeling task. First, a semantic role labeling dedicated classification head is added to the model output layer. The classification head adopts the structure of 2 layers of full connection layer and SoftMax activation. The input dimension of the first layer of full connection layer is the model hidden layer dimension (768 dimensions), the output dimension is set to 256 dimensions, and the ReLU activation function is selected to enhance the non-linear expression ability. The input dimension of the second layer of full connection layer is 256 dimensions, and the output dimension is set to 5 dimensions, corresponding to the action role, subject role, quantity role, type role and target role. The classification head parameters are initialized with Xavier uniform to avoid the problem of gradient disappearance in the early stage of training. Second, the domain adaptation layer is used to strengthen the model's ability to recognize professional terms. First, a business domain dictionary is constructed, which includes 2000+ form generation related professional terms such as leave approval process, engineering project approval number, department code and budget amount. Then, the terms in the dictionary are input into the AI large language model, and the 768-dimensional term word vector consistent with the model word embedding dimension is obtained through the Word2Vec algorithm with window size 5, minimum word frequency 3 and iteration number 100. Finally, the term word vector is weighted and fused with the original word embedding layer of the AI large language model (the fusion weight coefficient is set to 0.7 to highlight the domain semantics while preserving the model's general semantic understanding ability), forming a new word embedding layer, so that the model can more accurately recognize domain terms such as engineering project budget and leave days.

[0033] The model training is carried out with 64,000 training set annotation corpus as input, the training environment selects 4 NVIDIA A100 GPUs (single card memory 40 GB), is realized based on PyTorch 2.0 framework, adopts mixed precision training (FP16) to improve the training speed (2 times faster than FP32 training speed), the loss function selects cross entropy loss function, for the semantic role prediction result of each word segmentation unit, the cross entropy between the real annotation role is calculated, then the cross entropy of all word segmentation units in a single batch is summed and averaged to obtain the training loss value of the batch, which is used to measure the prediction deviation of the model, the optimizer selects Adam optimizer, the learning rate is set to 1e-5 (considering the training stability and convergence speed), the weight decay coefficient is 1e-4 (to prevent model overfitting), and the batch size is set to 32, that is, 32 corpus data are processed in each batch; considering that part of the corpus length is short, the gradient accumulation strategy is adopted, when the sample in a single batch is less than 32, the gradient of 2 micro-batches is accumulated and then the model parameter is updated, to ensure that the gradient information of each parameter update is stable enough, in the training process, through the gradient descent optimization mechanism built in the AI large language model, the loss value is calculated every iteration, the encoder weight, domain adaptation layer parameter and classification head weight of the model are adjusted dynamically, the prediction deviation is gradually reduced, and the stop judgment rule of the model training is set. After completing 10 rounds of training, that is, traversing 64,000 training sets 10 times, 16,000 validation sets are loaded to evaluate the semantic role annotation accuracy of the model. The accuracy calculation process is as follows: the validation set is input into the model in batches, the model predicts the semantic role of each word segmentation unit of the corpus, and the number of word segmentation units with correct prediction and real annotation is counted. For example, the validation set contains 120,000 word segmentation units, of which 118,8,000 are correctly predicted, and the accuracy is calculated according to the formula: accuracy = number of correctly annotated semantic units ÷ total number of semantic units in the validation set × 100%. In the example, the accuracy = 118,800 ÷ 120,000 × 100% = 99%. If the accuracy of the validation set after 3 consecutive training rounds is less than 0.1%, such as 98.5% in the 10th round, 98.55% in the 11th round and 98.58% in the 12th round, it is determined that the model has converged, and the training is stopped to avoid overfitting caused by overtraining. After the training is stopped, the parameters of the current optimal model are saved, including the BERT encoder weight, the domain adaptation layer parameter and the semantic role classification head parameter, and the saving format is the.pth file specially used by PyTorch. At the same time, a training log document is generated to record the training loss value, the validation set accuracy and the learning rate change curve of each training round. Thus, the pre-trained neural network model (AI large language model) is constructed.

[0034] The pre-trained neural network model (AI large language model) loads the built-in business field dictionary and uses the bidirectional maximum matching method for preliminary word segmentation. First, the text is scanned from left to right, taking the starting position of the text as the starting point, and the longest term or commonly used word in the dictionary is matched as the candidate word segmentation unit. For example, when scanning the 3-day sick leave form of the created employee, create, create employee, etc. are matched first, and the longest create in the dictionary is selected as the first candidate unit. Then, the text is scanned from right to left, taking the end position of the text as the starting point, and the candidate word segmentation unit is obtained in the same way. Then, the AI large language model corrects the candidate results of the two scans in combination with the context semantics, and excludes semantic contradictory word segmentation combinations, such as avoiding splitting 3 days, sick, and leave into 3 days, sick, and leave. Finally, the intent description is split into the smallest semantic units of create, employee, 3 days, sick leave, and form. The pre-trained neural network model (AI large language model) identifies the semantic attributes of each word segmentation unit. First, the word vector (generated by the AI large language model word embedding layer) of the word segmentation unit is input into the fully connected layer of the semantic role labeling classification head. The fully connected layer performs linear transformation on the input word vector. The calculation process is as follows: for each preset role (action role, subject role, etc.), multiply the dimension values of the word vector by the corresponding weight (the value range of the weight parameter is -0.5 to 0.5, which is dynamically adjusted through AI large language model training), and then sum them up. Then, add the bias term (the value range is -0.1 to 0.1) to get the original score of the role. Then, the original score is converted to a probability distribution through the SoftMax function. The sum of all role probabilities is 1. The AI large language model combines semantic understanding to perform secondary verification on the probability distribution, such as excluding the low probability of labeling employees as action roles. The role with the highest probability is selected as the labeling result of the word segmentation unit. For example, the action role probability of create is the highest, so it is labeled as an action role. The subject role probability of employee is the highest, so it is labeled as a subject role. The quantity role corresponds to 3 days, the type role corresponds to sick leave, and the target role corresponds to form. Finally, all word segmentation units and their corresponding roles are integrated to generate semantic role labeling results containing word segmentation-semantic role correspondence.

[0035] Step 101: Based on the semantic role annotation results, identify the user intent type through the classifier in the pre-trained neural network model. According to the user intent type, activate the corresponding attention mechanism module in the pre-trained neural network model to extract the field names of form field elements from the semantic role annotation results. Specifically, this includes: based on the semantic role annotation results, the pre-trained neural network model (an AI large language model adapted to form scenarios) calls the built-in SoftMax classifier, which is deeply integrated with the semantic encoding layer of the AI ​​large language model; input the key semantic roles in the annotation results, such as leave approval, project application, and preset intent type libraries, including leave approval form generation, project application form generation, etc., into the model. The SoftMax classifier first encodes the key semantic roles through the semantic encoding layer of the AI ​​large language model, converting natural language semantics into structured feature vectors, and then converting the feature vectors into probability distributions. The matching probability calculation formula for each intent type is as follows: ,in, Representing the The matching probability of each intent type ranges from 0 to 1 (output by the AI ​​large language model encoding layer, incorporating semantic information from the business scenario). Score the features for this intent type. Feature scores are given for all intent types. It is an intent type. It is an index of intent type. It is the first characteristic score of the intent type; the AI large language model makes a semantic rationality judgment on the probability distribution, excludes low-probability matching results that do not conform to the business logic, such as avoiding matching the leave approval form generation to the engineering project declaration form generation, and selecting the type with the highest probability as the final recognized user intent type, such as determining that the user intent type is the leave approval form generation if the leave approval form generation has the highest probability; according to the intent type, the pre-trained neural network model automatically activates the corresponding attention mechanism module, which is the core component of the AI large language model and has the ability of focusing on the context semantics, such as activating the leave-related semantic attention module corresponding to the leave approval intent; the module assigns higher attention weights to semantic units that are strongly related to the intent type, such as employee, sick leave, 3 days, and leave start date, through the context modeling capability of the AI large language model, the weight value ranges from 0 to 1, the weight of the strongly related unit is close to 1, such as the weight of sick leave is 0.92 and the weight of 3 days is 0.88, the weight of irrelevant units is close to 0, and at the same time, irrelevant semantic units with a weight lower than 0.2, such as create and need, are filtered out; the AI large language model combines the business logic of form generation, such as the leave approval form generation which needs to include core information such as employee identity, leave duration, and leave type, to complete the information of the semantic units with strong weights, such as completing the employee from the employee to the employee name, completing the 3 days to the leave days, completing the sick leave to the leave type, and keeping the original expression for the leave start date, and finally extracting form field names such as employee name, leave type, leave days, and leave start date.

[0036] At step 102, according to the semantic context of the field name, the corresponding data type and verification rule are determined, and the extraction of the form field element is completed, which specifically includes: the pre-trained neural network model (AI large language model adapted to the form scene) calls the built-in business semantic reasoning module, and for each extracted form field name, the semantic context of the field name is analyzed in combination with the semantic context of the field name, such as the field belonging to the business scene, the logical association with other fields, and the specific process is as follows: first, it is clear that the preset data type library includes character type, numerical type, date type and other types, and each type corresponds to a special semantic keyword (character type corresponds to name, department, address and other character / text identification keywords, numerical type corresponds to number of days, amount, quantity and other quantity measurement keywords, date type corresponds to date, time, start and end time keywords), the data type library is continuously updated and optimized by the AI large language model through learning of a large amount of government and enterprise form data, and the business semantic reasoning module analyzes the semantic composition of the form field name, splits the core semantic words from the field name, such as splitting employee name into employee and name, splitting leave days into leave and days, and splitting leave start date into leave, start and date, and the AI large language model mines the business association of the core semantic words, such as employee and name pointing to character identification information, and days and leave pointing to measurement information); calculate the semantic similarity of the field name and each data type, the calculation logic is that the business semantic reasoning module counts the number of words in the core semantic words of the field name that coincide with the special semantic keywords of a certain data type, that is, the number of semantic overlapping words, divides the number by the total number of core semantic words in the field name, and then multiplies by 100% to get the basic semantic similarity; the AI large language model corrects the basic semantic similarity in combination with the business scene, such as the basic semantic similarity of employee name being 50% (name coincides with character type keyword), and the model corrects the similarity to 100% in combination with the rule that employee name is generally character type in enterprise personnel form scene; the basic semantic similarity of leave days is 50% (days coincide with numerical type keyword), and the model corrects the similarity to 100% in combination with the characteristics of days as measurement in leave business; the basic semantic similarity of leave start date is 33.3% (date coincides with date type keyword), and the model corrects the similarity to 100% in combination with the expression habit of time type field, and finally the type with the highest corrected similarity is selected as the data type corresponding to the field, that is, employee name is character type, leave days is numerical type, and leave start date is date type.

[0037] Combined with the business attribute of the field name and the determined data type, the AI large language model calls a business rule knowledge base, which records government affairs approval specifications, enterprise management systems and the like. The corresponding verification rules are generated by continuous learning and updating of the model. For the character type of the employee name, starting from the business attribute of enterprise personnel management, the lower limit of the character length is set to 1 (greater than 0, to avoid empty value submission), and the upper limit is set to 50 (less than or equal to 50, to meet the length range of the conventional name). The regular expression generated by the AI large language model matches Chinese characters, letters and a small number of legal separators, and excludes special symbols such as @, # and $ for format verification. For the numerical type of the number of leave days, according to the enterprise leave system, the AI large language model calls from the business rule knowledge base, sets the lower limit of the numerical value to 1 (greater than 0, to avoid invalid leave application), and sets the upper limit to 90 (less than or equal to 90, to meet the upper limit requirement of the single leave time length of most enterprises). The integer verification logic of the AI large language model excludes decimals, negative numbers and zero values to ensure that the input is a positive integer without decimals. For the date type of the leave start date, the AI large language model obtains the current date through a system interface, and sets the lower limit of the date to the current date according to the conventional rule that the start date in the leave business should not be earlier than the application date. The upper limit is set to the current date plus 365 days, and the calculation method is to add 1 to the current date year, and the month and date remain unchanged, which meets the annual leave planning requirement. At the same time, the date format verification logic of the model (matching standard formats such as YYYY-MM-DD) ensures the input specification, and finally completes the extraction of the form field elements, including the field name, data type and verification rule.

[0038] In this embodiment, the structured separation of the intent description and the business type avoids the subsequent processing deviation caused by the mixed request information in the existing scheme. Through the field adaptation of the word segmentation and the semantic role labeling, the problems of insufficient business terminology recognition and rough semantic disassembly in the existing scheme are solved, so that the model can clearly capture the key semantic relationship in the intent. Through the intent type matching and the targeted attention mechanism, irrelevant information is further filtered to ensure that the field name extraction is not missed or redundant, and the problems of incorrect and missed field name extraction in manual configuration are avoided. Based on the semantic context and the business attribute, the data type and the verification rule are determined to solve the problem that the manual setting of the data type and the verification rule is easy to be out of touch with the business scenario, and to ensure that the extracted form field elements meet the data specification and are suitable for specific business requirements, such as the upper limit of the enterprise leave days and the date range of the government project declaration, thereby reducing the data errors and business conflicts in the subsequent form use.

[0039] In a preferred embodiment of the present application, the step 2 comprises:

[0040] In the embodiment of the present application, in step 200, the size parameters of the form layout virtual plane are determined according to the number and type of form field elements, and the form layout virtual plane with length and width attributes is constructed based on the size parameters, specifically including: determining the size parameters of the virtual plane according to the number and type of form field elements, setting the basic unit length as L, such as 50 pixels, the length parameter is calculated according to the total number of fields, the formula is virtual plane length = total number of fields × L × 0.8; the width parameter is calculated according to the type number, if including character type, numerical type and date type, then virtual plane width = 3 × L × 1.2; based on the calculation result, a rectangular form layout virtual plane with length and width attributes is constructed, for example, when the total number of fields is 8, virtual plane length = 8 × 50 × 0.8 = 320 pixels, width = 3 × 50 × 1.2 = 180 pixels, forming a virtual plane of 320 × 180 pixels.

[0041] In step 201, a basic layout framework is established on the form layout virtual plane according to the total number of form field elements, the basic layout framework is a rectangular area; five layout coordinate points are set at the four vertexes and the center point of the basic layout framework, the basic layout framework is divided into a plurality of field distribution areas according to the data type characteristics of the form field elements, the plurality of field distribution areas include character type field area, numerical type field area and date type field area, specifically including: a rectangular basic layout framework is established on the virtual plane with its boundary as the reference, the framework size is consistent with the virtual plane, five layout coordinate points are set at the four vertexes (left upper corner (0, 0), right upper corner (length, 0), left lower corner (0, width), right lower corner (length, width)) and the center point (length / 2, width / 2) of the framework; taking the center point as the reference, the area width is divided along the length direction according to the proportion of the number of character type, numerical type and date type fields, the formula is that the width of a certain type area = virtual plane width × (the number of fields of this type ÷ total number of fields); the complete height of the framework is maintained along the width direction, forming three continuously distributed field areas (character type, numerical type and date type), and the boundaries of each area are clear through the coordinate point connecting line.

[0042] In step 202, according to the business logic association of the form field elements, the field elements containing input-output relationship are allocated to the same field distribution area, wherein the character type field area is used to accommodate name, department name field elements, the numerical value type field area is used to accommodate leave days, project budget field elements, and the date type field area is used to accommodate start time, end time field elements, and specifically comprising: analyzing the business logic association of the form field elements, judging the input-output relationship by calculating the co-occurrence frequency (the number of the co-occurrence of two fields in the same business ÷ the total number of business cases) between fields, and determining that the fields with co-occurrence frequency exceeding 60% are strongly associated; allocating the strongly associated fields to the same area: the character type area accommodates text type fields such as name and department name, and the area width is evenly allocated according to the number of fields; the numerical value type area accommodates measurement type fields such as leave days and budget, and is sorted according to the numerical value precision requirement (integer first, decimal second); the date type area accommodates time type fields such as start time and end time, and is arranged according to the time sequence (start time first, end time second), so as to ensure the logical coherence of the fields in the same area.

[0043] In this embodiment, the virtual plane size and the area width are determined by quantitative calculation, avoiding the space waste or congestion problem caused by layout dependence on experience, so that the form structure is more in line with the actual demand of the number and type of fields; the area is divided according to the data type and the fields are allocated according to the business association, solving the user understanding difficulty caused by the chaotic arrangement of fields, making the character type, numerical value type and date type fields form a system, and the strongly associated fields are distributed in a centralized manner, so as to improve the filling efficiency; the area boundary is clearly defined by the coordinate point, and the fields are sorted according to the business logic, reducing the visual jump and thinking interruption of the user during filling, and reducing the operation failure rate; based on the calculation logic of the number and type of fields, the form layout can be automatically adjusted with the change of the fields, solving the problem that the fixed layout is difficult to adapt to dynamic demand.

[0044] In a preferred embodiment of the present application, the step 3 comprises:

[0045] In step 300, according to the character type field area, the numerical value type field area and the date type field area, the form field elements of the corresponding type are mapped to the corresponding field distribution area, specifically comprising: the system first reads the data type label (character type, numerical value type, date type) of each form field element, establishes the corresponding relationship between the type and the field distribution area, the character type field element corresponds to the character type field area, the numerical value type field element corresponds to the numerical value type field area, and the date type field element corresponds to the date type field area, and then all the field elements are allocated to the corresponding area one by one according to the type label of the field element, for example, the character type field elements such as employee name and department name are mapped to the character type field area, the numerical value type field elements such as leave days and project budget are mapped to the numerical value type field area, and the date type field elements such as leave start time and project end time are mapped to the date type field area, to complete the preliminary mapping.

[0046] Step 301, count the number of mapped form field elements in each field distribution area, calculate the field distribution density of the field distribution area, and analyze the semantic association and business logic dependency relationship between each form field element in the same field distribution area to obtain semantic association analysis data, including: the system performs statistical operation according to the field distribution area type (character type, numerical type, date type) in turn, first locates the character type field area, traverses all the mapped form field elements in the area, extracts the field identifier one by one and counts, records the result as N1, that is, the total number of field elements in the character type area, then traverses the numerical type field area in the same way, counts and records as N2, that is, the total number of field elements in the numerical type area, and finally traverses the date type field area, counts and records as N3, that is, the total number of field elements in the date type area; For example, the character type area contains 3 field elements of employee name, department name and post, then N1=3; The numerical type area contains 2 field elements of leave days and project budget, then N2=2; The date type area contains 2 field elements of leave start time and leave end time, then N3=2; Based on the coordinate data segmented in step 201, the length and width of each area are calculated, the length of each area = right boundary coordinate value-left boundary coordinate value, the width of each area = lower boundary coordinate value-upper boundary coordinate value, then the area of each area = length x width is calculated, for example, the left boundary coordinate of the character type area is 0, the right boundary coordinate is 200 pixels, the upper boundary coordinate is 0, and the lower boundary coordinate is 100 pixels, then the length = 200-0=200 pixels, the width = 100-0=100 pixels, the area = 200x100=20000 square pixels, recorded as S1 (character type area area); Similarly, the area of the numerical type area S2 and the area of the date type area S3 are calculated, then the field distribution density of each area = the number of field elements in each area ÷ the area of each area is calculated, for example, the character type area N1=3, S1=20000 square pixels, then the character type area density = 3÷20000=0.00015 / square pixel; The numerical type area N2=2, S2=15000 square pixels, then the numerical type area density = 2÷15000≈0.00013 / square pixel; The date type area N3=2, S3=15000 square pixels, then the date type area density = 2÷15000≈0.00013 / square pixel, to obtain the field distribution density of each area.

[0047] The system preset semantic association dictionary is called, which stores the semantic categories of fields in various business scenarios, such as the semantic categories of employee name, department name, and post being human attribute categories, the semantic categories of leave days and project budget being measurement categories, and the semantic categories of leave start time and leave end time being time categories. The semantic categories of any two field elements in the same area are compared. If the semantic categories of the two fields are completely consistent, such as the employee name and the department name being human attribute categories, it is determined that there is strong association. If there is a partial intersection between the semantic categories, such as the project budget being measurement category-amount and the material unit price being measurement category-unit price, it is determined that there is moderate association. If there is no association between the semantic categories, such as the employee name and the leave days, it is determined that there is weak association. According to the system preset business logic rule library, the rule library records the logical associations between various fields, such as the leave start time being earlier than the leave end time, and the two times having a time precedence dependency. The total project budget is obtained by summing up each sub-budget, and the two have a data subordinate dependency. The field elements in the same area are traversed to identify whether there is a logical association recorded in the rule library, and the dependency type (time precedence dependency, data subordinate dependency, etc.) is determined. If there is no logical association between the fields recorded in the rule library, it is determined that there is no dependency. Finally, the semantic association determination results (strong / moderate / weak association) of all fields in the same area and the business logic dependency type (including no dependency) are integrated to form the semantic association analysis data of the area.

[0048] In step 302, according to the field distribution density and the semantic association analysis data, the analysis results including the density index and the association degree index are obtained, which specifically include: first, the field distribution density calculated in step 301 is standardized, and 4 decimal places are uniformly retained as the density index, such as the character type area density 0.00028 being 0.0003 per square pixel after retention; then the association level (strong / moderate / weak) in the semantic association analysis data and the business logic dependency type (time precedence dependency, data subordinate dependency, etc.) are integrated into the association degree index; finally, according to the structure of the area type, the density index, and the association degree index, the two types of indexes are associated, for example, the character type area corresponds to the density 0.0003 per square pixel, the name and the department name in the area are strongly associated and have no logical dependency, the numerical type area corresponds to the density 0.0004 per square pixel, the leave days and the daily salary in the area are moderately associated and have a salary calculation logical dependency, and the date type area corresponds to the density 0.0002 per square pixel, the start time and the end time in the area are strongly associated and have a time precedence dependency, to form the complete analysis results including the field distribution area density index and the association degree index.

[0049] The embodiment avoids the field mismatch problem that is prone to occur in manual mapping by direct correspondence of type labels and areas, ensures accurate placement of different type field elements, and solves the form confusion problem caused by mixed distribution of fields; by calculating the field distribution density, the crowdedness of fields in the area is converted into a quantifiable index, breaking the limitation of judging the rationality of layout by experience, providing data support for layout adjustment, and avoiding the impact of improper allocation of area space on user filling experience; by analyzing semantic association and business logic dependency, the implicit association between fields is accurately identified, and the layout fragmentation problem caused by ignoring field logic is solved; the density index and the association index are integrated to form the analysis result, avoiding one-sidedness caused by single-dimensional optimization.

[0050] In a preferred embodiment of the application, the step 4 comprises:

[0051] Step 400, according to the density index in the analysis result, calculating the final area proportion required by each field distribution area; according to the association index in the analysis result, determining the arrangement order and boundary connection relationship between the character type field area, the numerical type field area and the date type field area, specifically including: setting the total area of the basic layout framework, which is determined by the virtual plane size in step 200, the basic layout framework = virtual plane length x virtual plane width, and presetting a density threshold value, such as 0.0002 per square pixel; for each field distribution area, calculating the required area according to area required area = number of field elements in the area ÷ density threshold value, and then calculating the proportion according to final area proportion = area required area ÷ total of area required areas, for example, character type area N1 = 3, numerical type area N2 = 2, date type area N3 = 2, density threshold value = 0.0002 per square pixel, character type area required area = 3 ÷ 0.0002 = 15000 square pixels; numerical type area required area = 2 ÷ 0.0002 = 10000 square pixels; date type area required area = 2 ÷ 0.0002 = 10000 square pixels; total of area required areas = 35000 square pixels; final area proportion of character type area = 15000 ÷ 35000 ≈ 42.86%, numerical type and date type area each is 10000 ÷ 35000 ≈ 28.57%; according to the arrangement order in the business logic dependency relationship in the association index, if there is time-length dependency between start time, end time in the date type area and the number of days off in the numerical type area, and there is no cross-area dependency for character type area employee name as basic information, then the arrangement order is set as character type area, date type area, numerical type area, and the boundary connection relationship is determined according to the arrangement order, that is, the right boundary of the character type area is aligned with the left boundary of the date type area (seamless connection), the right boundary of the date type area is aligned with the left boundary of the numerical type area, ensuring that the adjacent area boundaries have no overlap and the spacing is 0.

[0052] Step 401, based on the final area proportion of each field distribution area and the arrangement order requirement, obtain the layout optimization parameter containing area size parameter and area position parameter, specifically including: first, calculate the area size parameter according to the final area proportion calculated in step 400 and the total area of the basic layout framework, character type area size = total area of basic layout framework x 42.86% (length = virtual plane length x 42.86%, width = virtual plane width); numerical and date type area size = total area of basic layout framework x 28.57% (length = virtual plane length x 28.57%, width = virtual plane width); for example, total area of basic layout framework = 35000 square pixels, virtual plane length = 350 pixels, width = 100 pixels, character type area length = 350 x 42.86% ≈ 150 pixels, width = 100 pixels, size is 150 x 100 square pixels; numerical and date type area length = 350 x 28.57% ≈ 100 pixels, width = 100 pixels, size is 100 x 100 square pixels, then determine the area position parameter according to the arrangement order, that is, taking the left upper corner of the basic layout framework as the origin (0, 0), character type area position = (0, 0) (left boundary x = 0, upper boundary y = 0); date type area position = (150, 0) (left boundary x = 150, upper boundary y = 0); numerical type area position = (250, 0) left boundary x = 250, upper boundary y = 0, integrate the area size parameter and the position parameter to form the layout optimization parameter.

[0053] Step 402, according to the area size parameter in the layout optimization parameter, adjust the actual size of the character type field area, numerical type field area and date type field area in the basic layout framework; according to the area position parameter in the layout optimization parameter, adjust the arrangement order and boundary connection relationship of the character type field area, numerical type field area and date type field area in the basic layout framework, to obtain the size adjustment and position adjustment result; based on the size adjustment and position adjustment result, obtain the optimized basic layout framework, specifically including: according to the size parameter in the layout optimization parameter, modify the length of each area in the basic layout framework, that is, adjust the original character type area length from the initial value, such as 120 pixels, to 150 pixels, and adjust the original numerical type and date type area length from the initial value, such as 115 pixels, to 100 pixels, and the width remains consistent with the virtual plane width (100 pixels), complete the size adjustment, fix the character type area at the (0, 0) position, move the date type area to the left boundary x = 150 (right-join the character type area), move the numerical type area to the left boundary x = 250 (right-join the date type area), ensure that the areas are arranged in the order of character type, date type and numerical type, and the adjacent area boundaries are seamlessly connected, complete the position adjustment, integrate the size adjustment and position adjustment results to form the optimized basic layout framework which meets the area size adaptation requirement, arranges in accordance with the business logic, and the boundary connection is reasonable.

[0054] In this embodiment, the area demand area and the final proportion are calculated by the density index, the problems of partial area congestion and partial area idling caused by the dependence of space allocation on experience in the layout are avoided, the space of each area is matched with the number of fields, the form space utilization rate is improved, the order of area arrangement and the boundary connection are determined according to the correlation degree index, the areas with business dependence are adjacent, the problem of user filling thought interruption caused by the dispersion of logically related areas in the layout is solved, the size and position parameters are generated by quantitative calculation, the subjective deviation of manual adjustment is avoided, and the size and position of each area can be accurately controlled; the size and position of each area in the optimized basic layout framework are clear, the fields can be directly arranged in order according to the area when the form fields are mapped subsequently, and the chaotic distribution of fields is avoided.

[0055] In a preferred embodiment of the present application, the step 5 comprises:

[0056] Step 500, transmit the user intention type, form field element and optimized basic layout framework to the middle code platform, determine the corresponding business component library according to the user intention type through the component matching model preset in the middle code platform; determine the corresponding form container component from the business component library based on the arrangement structure of the character type field region, numerical type field region and date type field region in the optimized basic layout framework; match the corresponding input box component, drop-down selection component and date selector component from the business component library according to the field name, data type and verification rule in the form field element, specifically including: the system first transmits the user intention type such as leave approval form generation, project declaration form generation, form field element containing field name, data type, verification rule and optimized basic layout framework to the middle code platform, the middle code platform starts the preset component matching model, the first step matches the corresponding business component library according to the user intention type, if the user intention type is leave approval form generation, match the enterprise office-leave approval component library; if it is project declaration form generation, match the government affairs declaration-engineering project component library; the second step is based on the region arrangement structure of the optimized basic layout framework, such as horizontal arrangement of character type, date type and numerical type, screen the adaptive form container component from the business component library, the horizontal arrangement structure corresponds to the horizontal partition form container component, which contains a sub-container consistent with the number of regions (3 regions correspond to 3 sub-containers), the size of the sub-container matches the optimized region size parameter, such as character type region 150*100 pixels, numerical type region 100*100 pixels, the third step matches the input component according to the form field element, the field data type is character type, such as employee name, department name, match single line input box component, and write field verification rules such as character length 0-50, no special symbol into component attributes; the data type is numerical type, such as leave days, project budget, match numerical input box component, write verification rules such as 1-90 integers, non-negative; the data type is date type, such as leave start time, project deadline, match date selector component, write verification rules such as not earlier than the current date, not later than the current+365 days; if the field has fixed options, such as leave type containing sick leave and casual leave, match the drop-down selection component, and write the option list into the component attribute.

[0057] Step 501, the form container component is combined with the input box component according to the area structure of the optimized basic layout framework, and a form component description containing component type, component attribute and layout information is obtained, specifically including: the system combines the form container component with the matching input component according to the area structure of the optimized basic layout framework, combines the first sub-container of the form container in a horizontal direction, associates all single-line input box components and drop-down selection components with the character type area, combines the second sub-container with all date selector components, and combines the third sub-container with all numerical value input box components corresponding to the numerical value type area, then integrates the component type, such as the horizontal partition form container, the single-line input box, the date selector, the component attribute, such as the input box length limit, the date range, the numerical value range, and the layout information, such as the coordinates of the components in the sub-container, the first single-line input box coordinate (10, 10), the second single-line input box coordinate (10, 50), and the interval of 40 pixels, to form the form component description containing the above information.

[0058] In this embodiment, the corresponding business component library is locked through the user intention type, the form container component is matched in combination with the layout arrangement structure, the input type component is matched according to the field attribute, the precise correspondence of the component, the business scene, the layout structure and the field requirement is realized, the component mismatch or functional redundancy is avoided, the component is automatically screened relying on the preset component library and the matching model, the manual selection one by one is replaced, the component description is generated through the structured combination, the manual operation of the component selection and integration is greatly reduced, the form development cycle is shortened, the component is selected based on the standardized component library and the unified matching logic, the combination is performed in combination with the area structure of the layout framework, the component type, the attribute configuration and the layout logic of the form under different business scenes are kept standardized and unified, and the application consistency is improved.

[0059] In a preferred embodiment of the present application, the step 6 comprises:

[0060] At step 600, according to the component type, component attribute and layout information in the form component description, the corresponding form container component and input box component are instantiated, and the instantiated input box component is assembled into the corresponding form container component according to the layout information to generate a complete form page, specifically including: reading the form component description, first instantiating the form container component, that is, creating a horizontal partition form container instance according to the container size parameters in the layout information, such as total length 350 pixels and width 100 pixels, then creating three sub-container instances according to the sub-container size parameters (150x100 pixels, 100x100 pixels and 100x100 pixels), then instantiating each input component, that is, creating single-line input box, numerical input box and date selector instances according to the parameters in the component attribute, such as single-line input box length 50 characters and numerical input box range 1-90, and finally assembling each input component instance into the corresponding sub-container instance according to the coordinate position in the layout information, that is, assembling the employee name input box (coordinate (10, 10)) and the department name input box (coordinate (10, 50)) into the character type sub-container, assembling the leave start time selector (coordinate (10, 30)) into the date type sub-container, and assembling the leave number of days input box (coordinate (10, 30)) into the numerical type sub-container. After all components are assembled, a complete form page is generated.

[0061] At step 601, the form page is rendered and displayed in the dialogue interface, and the form data input and submitted by the user in the form page is received, the received form data is verified and processed by the business rules through the automatic process engine to obtain the business rule processing result, specifically including: transmitting the generated form page to the natural language dialogue interface, rendering and displaying the form page in the interface, inputting data such as employee name, leave number of days and leave start time in the form page and submitting by the user, then the interface transmits the form data to the automatic process engine, the automatic process engine calls the verification rules in the form field elements to verify the submitted data, verifies the length of character type data such as employee name length 2-10 characters, which meets the requirement of 0-50, the range of numerical type data such as leave number of days 3 days, which meets the requirement of 1-90, and the legality of date type data such as leave start time being the current date+1 day, which meets the requirement of not earlier than the current date; after the verification is passed, the engine processes the data according to the business rules such as leave number of days≤3 days being approved by the department manager and leave number of days>3 days being approved by the general manager to determine the corresponding approval level and obtain the business rule processing result.

[0062] At step 602, according to the processing result of the business rule, a corresponding approval process or data storage operation is triggered to complete the execution of the business process based on the dialogue interaction, specifically including: according to the processing result of the business rule, triggering the corresponding operation, if the processing result is the department manager approval, calling the approval process interface, automatically initiating the leave approval process of the department manager, and pushing the approval notification to the department manager account; if the processing result is data storage, such as engineering project declaration basic information, which does not need to be approved and directly archived, a data storage interface is called, and the form data is stored in a preset format, such as JSON format, to a business database, such as an enterprise office database or a government affairs declaration database, and the execution of the business process based on the dialogue interaction is completed after the approval process is completed or the data storage is successful.

[0063] In this embodiment, the components are instantiated according to the component description and assembled according to the layout, avoiding the tedious operation of manually building the form, ensuring that the structure of the form page is consistent with the optimized layout framework; the dialogue interface is directly rendered, so that the user can obtain the form without scene switching, improving the convenience of interaction; the automatic process engine first filters invalid data according to the verification rule, avoiding the flow of error data into the business link; then the data is processed according to the business rule, ensuring that the result is consistent with the actual business logic, reducing the manual verification cost and business deviation; based on the processing result, the approval or storage operation is automatically triggered, without manual process initiation, realizing seamless connection from user data submission to business process landing, shortening the business processing period, and improving the efficiency of business execution based on dialogue interaction; the user can complete form filling, submission and process tracking in the dialogue interface, reducing cross-platform operation; automatic processing reduces the risk of human error, ensures the stability and accuracy of business process execution, and improves the user's trust in business processing.

[0064] As shown in Figure 2 The embodiment of the present application also provides a form page construction management system based on intelligent dialogue generation, which comprises:

[0065] The acquisition module is configured to perform intent analysis on the user intent description through a pre-trained neural network model, identify the user intent type, and extract form field elements, including field name, data type and verification rule; a form layout virtual plane is constructed based on the form field elements, and a basic layout framework is established on the form layout virtual plane; layout coordinate points are set in the basic layout framework; the basic layout framework is divided into a plurality of field distribution regions according to the data characteristics of the form field elements; the form field elements are mapped to the field distribution regions, and the distribution density and correlation of the fields in each field distribution region after mapping are analyzed to obtain an analysis result.

[0066] The processing module is configured to generate layout optimization parameters according to the analysis result, adjust the basic layout framework based on the layout optimization parameters to obtain an optimized basic layout framework, transmit the user intent type, the form field element and the optimized layout framework to the code platform, and process the form component description obtained through the preset component matching model in the code platform, generate a form page based on the form component description, and display the form page in the dialogue interface, receive form data submitted by the user through the form page, and automatically process the form data through the automatic process engine to implement the business process execution based on the dialogue interaction.

[0067] It should be noted that the system corresponds to the above method, and all implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0068] Embodiments of the application also provide a computer readable storage medium storing instructions which, when executed on a computer, cause the computer to perform the method described above. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0069] The above is the preferred embodiment of the application. It should be noted that, for those skilled in the art, without departing from the principles of the application, several improvements and refinements can be made, which should also be considered within the protection scope of the application.

Claims

1. A method for constructing and managing form pages based on AI large-scale model dialogue generation, characterized in that, The method includes:

1. The user intent description is parsed using a pre-trained neural network model to identify the user intent type and extract form field elements, which include field name, data type and validation rules; 2. Construct a virtual form layout plane containing length and width attributes; on the virtual form layout plane, establish a basic layout framework based on the total number of form field elements, wherein the basic layout framework is a rectangular area; set five layout coordinate points at the four vertices and the center point of the basic layout framework; divide the basic layout framework into multiple field distribution areas according to the data type characteristics of the form field elements, wherein the multiple field distribution areas include character field areas, numeric field areas, and date field areas; according to the business logic relationship of the form field elements, assign field elements containing input-output relationships to the same field distribution area; 3. Based on the character field area, numeric field area, and date field area, map the corresponding form field elements to the corresponding field distribution areas; count the number of form field elements mapped in each field distribution area, calculate the field distribution density of the field distribution area, and analyze the semantic relationships and business logic dependencies between form field elements in the same field distribution area to obtain semantic relationship analysis data; based on the field distribution density and semantic relationship analysis data, obtain analysis results including density indicators and relationship indicators; 4. Based on the density index in the analysis results, calculate the final area ratio required for each field distribution area; based on the correlation index in the analysis results, determine the arrangement order and boundary connection relationship between character field areas, numeric field areas, and date field areas; based on the final area ratio and arrangement order requirements of each field distribution area, obtain layout optimization parameters including area size parameters and area position parameters; based on the area size parameters in the layout optimization parameters, adjust the actual size of character field areas, numeric field areas, and date field areas in the basic layout framework; based on the area position parameters in the layout optimization parameters, adjust the arrangement order and boundary connection relationship of character field areas, numeric field areas, and date field areas within the basic layout framework, obtaining the size adjustment and position adjustment results; based on the size adjustment and position adjustment results, obtain the optimized basic layout framework; 5. Transmit the user intent type, form field elements, and optimized basic layout framework to the code-in-the-middle platform. Using the pre-built component matching model in the code-in-the-middle platform, determine the corresponding business component library based on the user intent type. Based on the arrangement structure of character field areas, numeric field areas, and date field areas in the optimized basic layout framework, determine the corresponding form container component from the business component library. Based on the field name, data type, and validation rules in the form field elements, match the corresponding input box component, dropdown selection component, and date picker component from the business component library. Combine the form container component and the input box component according to the area structure of the optimized basic layout framework to obtain a form component description containing component type, component attributes, and layout information.

6. Generate form pages based on form component descriptions and display the form pages in the dialog interface; receive form data submitted by users through the form pages, and automate the form data through the automated process engine to realize the execution of business processes based on dialog interaction.

2. The method for constructing and managing form pages based on AI large-scale model dialogue generation according to claim 1, characterized in that, Before step 1, a form generation request is received from the user through a natural language dialogue interface. The request includes a description of the user's intent and the requirement information of the business type, which includes employee leave approval or engineering project application in government affairs declaration.

3. The method for constructing and managing form pages based on AI large-scale model dialogue generation according to claim 2, characterized in that, Step 1 includes: The user intent description is segmented and semantic role labeled using a pre-trained neural network model to obtain the semantic role labeling results. Based on the semantic role labeling results, the user intent type is identified by the classifier in the pre-trained neural network model. According to the user intent type, the corresponding attention mechanism module in the pre-trained neural network model is activated to extract the field names of the form field elements from the semantic role labeling results. Based on the semantic context of the field names, determine the corresponding data types and validation rules to complete the extraction of form field elements.

4. The method for constructing and managing form pages based on AI large-scale model dialogue generation according to claim 3, characterized in that, Step 6 includes: Based on the component type, component attributes, and layout information in the form component description, instantiate the corresponding form container component and input box component, and according to the layout information, assemble the instantiated input box component into the corresponding form container component to generate a complete form page; The system renders and displays the form page in the dialog interface, receives the form data entered and submitted by the user on the form page, and performs validation and business rule processing on the received form data through the automated process engine to obtain the business rule processing result. Based on the processing results of business rules, the corresponding approval process or data storage operation is triggered to complete the execution of the business process based on dialogue interaction.

5. A form page construction and management system based on intelligent dialogue generation, wherein the system implements the method as described in any one of claims 1 to 4, characterized in that, include: The acquisition module is used to parse user intent descriptions using a pre-trained neural network model, identify user intent types, and extract form field elements, including field names, data types, and validation rules; it also constructs a virtual plane for form layout based on the form field elements and establishes a basic layout framework on the virtual plane for form layout. Set layout coordinate points within the basic layout framework; divide the basic layout framework into multiple field distribution areas based on the data characteristics of the form field elements; map the form field elements to the field distribution areas, and analyze the distribution density and correlation of fields in each field distribution area after mapping to obtain the analysis results; The processing module is used to generate layout optimization parameters based on the analysis results, and adjust the basic layout framework based on the layout optimization parameters to obtain the optimized basic layout framework. The user intent type, form field elements, and optimized layout framework are transmitted to the code-in-the-middle platform and processed by the pre-built component matching model in the code-in-the-middle platform to obtain the form component description. Generate a form page based on the form component description and display the form page in the dialog interface; It receives form data submitted by users through a form page and processes the form data automatically through an automated process engine to achieve business process execution based on dialogue interaction.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 4.

Images