A visual page construction method and apparatus based on page generation intent

Abstract

This disclosure presents a method and apparatus for constructing a visual page based on page generation intent. One specific implementation of the method includes: guiding the user to collect their page generation intent; matching a visual page template with intent description information and a pre-built visual page template library; parsing the target page template; generating indicator description information corresponding to each component description information in a component description information set based on the intent description information; generating static fill data corresponding to each indicator description information in the indicator description information set; filling the target page template with page data based on the static fill data set; performing page integrity verification on the filled page; and rendering the visual page after the filled page passes the page integrity verification. This implementation effectively reduces the user's visual page construction cost, thereby shortening the development cycle of the visual page.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, particularly to the field of data visualization and low-code technology, and specifically to a method and apparatus for constructing visual pages based on page generation intent. Background Technology

[0002] Current visual page building technologies mainly revolve around two dimensions: "page template reuse" and "low-code page configuration." The mainstream visual page building technologies can be summarized as follows:

[0003] (1) Visual page construction is performed by dragging and dropping page components;

[0004] (2) Define the structure and map the data of page components in the visual page through low-code configuration;

[0005] (3) Use existing BI tools (such as Tableau, Power BI, FineBI, etc.) to build visual pages by dragging and dropping page components and connecting to the database.

[0006] While the aforementioned existing technologies can reduce the technical difficulty of building visual pages to some extent, the following technical problems still exist: users need to manually select page templates that match their needs and manually adjust components, map and associate data, and populate them. This makes it impossible to truly lower the barrier to entry for no-code visual pages, thus preventing the development cycle of visual pages from being further compressed. Summary of the Invention

[0007] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0008] Some embodiments of this disclosure propose a visual page construction method and apparatus based on page generation intent to solve the technical problems mentioned in the background section above.

[0009] In a first aspect, some embodiments of this disclosure provide a method for constructing a visual page based on page generation intent. The method includes: collecting page generation intent from a target user in a guided manner to obtain intent description information; matching the intent description information with a pre-built visual page template library to obtain a target page template; parsing the target page template to obtain a set of component description information, wherein the component description information represents the template components contained within the target page template; generating indicator description information corresponding to each component description information in the component description information set based on the intent description information to obtain an indicator description information set; generating static fill data corresponding to each indicator description information in the indicator description information set to obtain a static fill data set, wherein the static fill data is bound one-to-one with the template components corresponding to the component description information; filling the target page template with page data based on the static fill data set to obtain a filled page; performing page integrity verification on the filled page; and rendering the visual page after the filled page passes the page integrity verification.

[0010] Secondly, some embodiments of this disclosure provide an apparatus for a visual page construction method based on page generation intent. The apparatus includes: an intent acquisition unit configured to collect page generation intent from a target user in a guided manner to obtain intent description information; a visual page template matching unit configured to perform visual page template matching based on the intent description information and a pre-built visual page template library to obtain a target page template; a template parsing unit configured to parse the target page template to obtain a set of component description information, wherein the component description information represents the template components contained in the target page template; and a first generation unit configured to generate the components based on the intent description information. The first unit generates a set of indicator description information, which is the indicator description information corresponding to each component description information in the indicator description information set. The second unit is configured to generate static fill data corresponding to each indicator description information in the indicator description information set, which is the static fill data set. The static fill data is bound one-to-one with the template component corresponding to the component description information. The third unit is configured to fill the target page template with page data according to the static fill data set, which is the filled page. The fourth unit is configured to perform page integrity verification on the filled page and perform visualization page rendering after the filled page passes the page integrity verification.

[0011] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0012] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0013] The above embodiments of this disclosure have the following beneficial effects: The visualization page construction method based on page generation intent, as described in some embodiments of this disclosure, achieves automated visualization page construction based on intent understanding, effectively reducing the user's visualization page construction cost and thus compressing the development cycle of the visualization page. Specifically, existing methods such as "drag and drop page components," "low-code configuration," and "using existing BI" all require users to manually select page templates that match their needs and manually adjust components, map and associate data, and populate them. This is especially problematic for users who haven't prepared page filling data, requiring them to spend extra time collecting this data and preventing direct previewing of the visualization page. Therefore, the visualization page construction method based on page generation intent, as described in some embodiments of this disclosure, firstly collects the target user's page generation intent in a guided manner to obtain intent description information. In practice, visualization pages mainly display key data in the form of charts and text. Furthermore, the page structure and content of visualization pages vary depending on different display requirements. Therefore, this disclosure, through guided page generation intent collection, enables target users to comprehensively describe their visualization page construction needs using natural language. Secondly, based on the aforementioned intent description information and the pre-built visual page template library, visual page template matching is performed to obtain the target page template. In practice, conventional methods for page template matching mainly involve matching user-input template tags with preset template tags. However, when the two are inconsistent or there is no corresponding preset template tag, matching will fail. Furthermore, tag matching alone is insufficient to extract implicit page construction requirements. Therefore, this disclosure combines intent description information with template matching to improve the accuracy and success rate of matching. Next, the target page template is parsed to obtain a set of component description information, where the component description information represents the template components contained within the target page template. In practice, different page templates may contain different template components, and the data configurations of different template components may also differ. Therefore, this disclosure automatically extracts the component configurations of the template components contained in the target page template through template parsing. Further, based on the aforementioned intent description information, indicator description information corresponding to each component description information in the aforementioned component description information set is generated to obtain an indicator description information set. In addition, static fill data corresponding to each indicator description information in the above indicator description information set is generated to obtain a static fill data set, wherein the static fill data is bound one-to-one with the template component corresponding to the component description information.In practice, to ensure effective data display on visualization pages, conventional generation methods often require users to collect the page data to be displayed in advance and adjust the format of the page data to ensure compatibility with the corresponding template components. This significantly increases the user's visualization page construction cost. Therefore, this disclosure uses intent description information and indicator description information as constraints to automatically generate static data. Next, based on the aforementioned static data set, the target page template is populated with page data to obtain the populated page. Finally, the populated page undergoes page integrity verification, and the visualization page is rendered only after the populated page passes the integrity verification. In summary, this method effectively reduces the user's visualization page construction cost, thereby shortening the visualization page development cycle. Attached Figure Description

[0014] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0015] Figure 1 This is a flowchart of some embodiments of the visual page construction method based on page generation intent according to the present disclosure;

[0016] Figure 2 This is a diagram illustrating the process of collecting the intent for page generation.

[0017] Figure 3 This is another illustration of the process of collecting intent for page generation;

[0018] Figure 4 This is a diagram illustrating the relationship between the template vector library, the page template information library, and the visual page templates.

[0019] Figure 5 These are schematic diagrams illustrating the structure of some embodiments of the visual page building apparatus based on page generation intent according to the present disclosure;

[0020] Figure 6 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0023] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0024] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0025] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0026] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] refer to Figure 1 The diagram illustrates a flow 100 of some embodiments of a visual page construction method based on page generation intent according to the present disclosure. This visual page construction method based on page generation intent includes the following steps:

[0028] Step 101: Conduct guided page generation intent collection for the target user to obtain intent description information.

[0029] In some embodiments, the executing entity (e.g., a computing device) of the visual page building method based on page generation intent can collect the page generation intent of the target user in a guided manner to obtain intent description information.

[0030] The target user can be any user who needs a visual page generated. The intent description information can characterize the page generation intent of the aforementioned target users.

[0031] In practice, firstly, the user's natural language input request for generating a visual page can be collected. During the request collection process, a pop-up window can guide the user's input. Secondly, a finely tuned intent recognition model (e.g., ERNIE 3.0) is used to identify the intent of the visual page generation request, obtaining intent description information.

[0032] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0033] In some optional implementations of certain embodiments, the aforementioned execution entity performs guided page generation intent collection on the target user to obtain intent description information, including:

[0034] Step S1: In response to the target user entering real-time intent text in the text input box, the following processing steps are performed based on the real-time intent text:

[0035] The real-time intent text mentioned above ends with a first end-of-line character. This first end-of-line character can be a predefined custom character used to indicate the end of the real-time intent text input. For example, the first end-of-line character can be represented by "[PEND]". The text input box described above only supports text-based page-generated intent input.

[0036] Step S11: Perform text preprocessing on the above real-time intent text to obtain a filtered word set.

[0037] In practice, firstly, the aforementioned execution entity cleanses the real-time intent text to obtain cleaned intent text. Specifically, this involves converting all letters in the real-time intent text to lowercase, standardizing dates and numbers, and filtering out punctuation marks, special symbols, redundant spaces, and invisible characters. Secondly, the cleaned intent text undergoes word segmentation. Specifically, the execution entity can use the HanLP tool to segment the cleaned intent text, obtaining a text word set. Next, the text word set is filtered using a general stop word list and a dedicated stop word list, resulting in a filtered word set. The general stop word list filters meaningless words, while the dedicated stop word list filters element descriptions corresponding to the visualization page. Furthermore, the HanLP tool can be used to perform part-of-speech tagging on the filtered words in the filtered word set.

[0038] As an example, the real-time intent text could be "Generate a visualization page for data analysis to display quarterly sales data for e-commerce." The set of text words could be ["generate", "for", "display", "e-commerce", "quarterly", "sales data", "of", "data analysis", "with", "of", "visualization", "page"]. Text words filtered out using a general stop word list could be ["generate", "for", "display", "of", "with", "of"]. Text words filtered out using a specific stop word list could be ["visualization", "page"]. The resulting set of filtered words could be ["e-commerce", "quarterly", "sales data", "data analysis"].

[0039] Step S12: Encode each filtered word in the above filtered word set to generate a word vector, thus obtaining a word vector set.

[0040] In practice, word vectors can be generated by encoding the filtered words using a Transformer-based encoding network.

[0041] Step S13: In response to the empty historical intent description information queue, generate the intent guidance information corresponding to the above real-time intent text based on the above word vector set and the pre-trained intent recognition model.

[0042] The historical intent description information sequence is used to store intent description information corresponding to the real-time intent text previously entered by the target user in the current question-and-answer round. The intent recognition model in this disclosure adopts the ERNIE 3.0 Small model. The real-time intent description information represents the page generation intent corresponding to the real-time intent text. The intent guidance information represents the guiding text used to guide the target user to further supplement new real-time intent text.

[0043] In practice, the training samples for the intent recognition model use manually constructed raw samples (real-time intent text), and are manually labeled with corresponding intent tags (e.g., intent topic, applicable scenario, analysis dimension, and display scale) and intent guidance text. Furthermore, the aforementioned execution entity can expand the number of raw samples through methods such as synonym replacement, sentence structure replacement, and keyword addition / deletion to ensure semantic diversity. Specifically, the number of manually constructed raw samples can be 1000, which is then expanded to 10000. The expanded 10000 raw samples undergo manual review through random sampling to ensure labeling quality. During the model training phase, the ERNIE 3.0 Small model is fine-tuned based on the PyTorch framework to achieve domain adaptation. The intent recognition model is encapsulated as an interface service via a RESTful API for invocation. To improve the accuracy of intent parsing, during the model iteration phase, real-time user intent text and generated real-time intent description information are collected in real time to generate structured logs. Log records from structured logs are periodically extracted and manually labeled to serve as new raw samples for incremental training of the intent recognition model. Specifically, during incremental training, the model parameters are partially frozen (e.g., 90% of the underlying model parameters of the ERNIE 3.0 Small model are frozen), and the retrained intent recognition model is A / B tested. If the test results indicate that the intent recognition accuracy of the retrained intent recognition model is greater than the preset accuracy, the retrained intent recognition model is deployed; otherwise, it is rolled back to the previous version of the intent recognition model.

[0044] Step S14: In response to the historical intent description information queue being non-empty, based on the historical intent features corresponding to the historical intent description information in the historical intent description information queue, the word vector set, and the pre-trained intent recognition model, generate the real-time intent description information and intent guidance information corresponding to the real-time intent text, and add the real-time intent description information corresponding to the real-time intent text as historical intent description information to the historical intent description information queue.

[0045] In practice, to avoid the feature forgetting problem in long sequences, when the historical intent description information queue is not empty, this disclosure will combine the historical intent features and word vector set corresponding to the historical intent description information in the historical intent description information queue generated in the current question-and-answer round as input to the intent recognition model to obtain real-time intent description information and intent guidance information for real-time intent text.

[0046] Step S15: Display the intent guidance information in the above text input box.

[0047] The aforementioned intent guidance information is used to guide the text input of the target user.

[0048] In practice, intent guidance information can be appended to the real-time intent text.

[0049] Step S16: In response to the target user not entering new real-time intent text in the text input box within a preset time period or the target user entering a second end character in the text input box, generate the intent description information according to the historical intent description information queue.

[0050] The second ending symbol can be a predefined custom symbol used to indicate that the page generation intent has been fully collected. For example, the second ending symbol can be represented by "[MEND]".

[0051] In practice, historical intent description information can be read from the historical intent description information queue and converted into JSON (JavaScript Object Notation) format as intent description information.

[0052] Step S2: In response to the target user entering new real-time intent text in the text input box within the preset time, the new real-time intent text is used as the real-time intent text, and the above processing steps are executed again.

[0053] As an example, see Figure 2 The diagram shown illustrates the process of collecting intent for page generation. Figure 3 The diagram illustrates another process for capturing the intent to generate a page. For complex visualization pages, multiple template components are often used for data display. To ensure the page's aesthetics, the layout between these components also needs adjustment. Furthermore, to ensure proper data display, the data to be displayed and the template components need to be formatted compatible. Based on the above analysis, it is clear that if a natural language-based visualization page generation method is adopted, the user needs to provide a complete and comprehensive description of their requirements for the intent recognition model to accurately capture the corresponding visualization page generation needs. Figure 2This demonstrates a conventional intent collection method. Specifically, first, the user submits the real-time intent text: "Generate a visualization page for data analysis to display quarterly e-commerce sales data." Next, the intent recognition model analyzes this text and generates corresponding intent description information: "'E-commerce Quarterly Sales Data Dashboard' Part 1: Displays total sales, total orders, gross profit margin, and active users. Part 2: Displays the trend between sales and orders, monthly task completion progress..." Further, the user inputs new real-time intent text: "Further, based on the above analysis, Part 2 displays the trend between sales and orders using a double-line chart, and the monthly task completion progress using a pie chart." It is evident that the conventional intent collection mode primarily uses multi-round queries, guided by the intent recognition model, to allow the user to continuously refine and adjust the corresponding page generation intent. During this process, the user needs to wait for the previous round of queries to end or for the previous round of queries to provide intent guidance before initiating a new round of queries. This is especially true when the visualization page is complex, requiring multiple rounds of queries to complete. Furthermore, the lack of effective intent constraints in the initial real-time intent text often leads to biased or even erroneous intent recognition results from the intent recognition model, requiring users to correct them within new real-time intent text. In summary, conventional intent acquisition methods incur significant waiting time during multi-round queries (primarily due to intent correction and the excessive processing time of large language model-based intent recognition models during response generation), greatly increasing the time cost of building visualization pages. Furthermore, Figure 3The disclosure illustrates the intent collection method. Specifically, firstly, the user requests, "Generate a visualization page for data analysis to display quarterly e-commerce sales data," ending with the first closing character "[PEND]". At this point, the aforementioned execution entity will capture the end of the user's real-time intent text description and automatically generate corresponding intent guidance information "[It is suggested to further supplement the name of the indicator to be displayed]" based on "Generate a visualization page for data analysis to display quarterly e-commerce sales data." The user can either refer to this intent guidance information or not use it when entering their real-time intent text. Next, the user further suggested that "the metrics that need to be displayed include: 'total sales revenue,' 'total number of orders,' 'gross profit margin,' 'number of active users,' 'trend between sales revenue and number of orders,' 'monthly task completion progress,' and other metrics," ending with the first closing character "[PEND]." At this point, the aforementioned execution entity will detect the end of the user's new real-time intent text description and automatically generate new, corresponding intent guidance information based on "the metrics that need to be displayed include: 'total sales revenue,' 'total number of orders,' 'gross profit margin,' 'number of active users,' 'trend between sales revenue and number of orders,' 'monthly task completion progress,' and other metrics," as well as "[Suggestion to further supplement the display method of the metrics]." The user can either refer to this intent guidance information or not use it when entering real-time intent text. Finally, the user further requested that "'Total sales revenue,' 'Total orders,' 'Gross profit margin,' and 'Active users' be displayed numerically; 'The trend between sales revenue and orders' be displayed using a double-line chart; and 'Monthly task completion progress' be displayed using a pie chart." This request was concluded with the second closing character "[MEND]". At this point, the aforementioned execution entity would generate a complete intent description based on the following: "Generate a visualization page for data analysis to display quarterly e-commerce sales data." Specifically, the required metrics to be displayed include: 'Total sales revenue,' 'Total orders,' 'Gross profit margin,' 'Active users,' 'The trend between sales revenue and orders', 'Monthly task completion progress,' and other metrics," and "'Total sales revenue,' 'Total orders,' 'Gross profit margin,' and 'Active users' should be displayed numerically; 'The trend between sales revenue and orders' should be displayed using a double-line chart; and 'Monthly task completion progress' should be displayed using a pie chart." (Comparison) Figure 2 and Figure 3 It can be observed that: Figure 3The intent acquisition method employs only a single round of queries, which can capture complete and accurate intent description information. Furthermore, since this disclosure is applied to the specific domain of visual pages, the intent recognition model can be optimized within this domain. A lightweight intent recognition model (ERNIE 3.0 Small model) is used to achieve rapid generation of intent guidance information, thereby reducing the waiting time for intent guidance information generation and improving the accuracy of the generated intent guidance information. This approach optimizes both waiting time and guidance accuracy, effectively improving the user experience.

[0054] Step 102: Match the visual page templates with the intent description information and the pre-built visual page template library to obtain the target page template.

[0055] In some embodiments, the aforementioned execution entity can perform visual page template matching based on intent description information and a pre-built visual page template library to obtain a target page template.

[0056] The Visual Page Template Library is a database storing pre-built visual page templates. The target page template is a visual page template that matches the intent description information. Each visual page template in the library has a corresponding template vector for subsequent similarity matching. The template vector can be a feature vector constructed by adjusting the template corresponding to the visual page template.

[0057] In practice, the aforementioned execution entity can convert intent description information into retrieval vectors and then filter target page templates from a visual page template library using similarity matching. The vector dimension of the template vector is the same as that of the retrieval vector.

[0058] Optionally, the visual page template library consists of a template vector library and a page template information library. The template vector library stores the template vectors corresponding to the visual page templates. The page template information library stores the page template information corresponding to the visual page templates. Template vectors and page template information corresponding to the same visual page template are associated through a page template identifier. This separation of the template vector library and page template information library achieves decoupling between them. Specifically, the template vector library of this disclosure can use the Milvus vector library. The page template information library uses a MySQL database. Template vectors can represent the template feature vectors of a visual page template. Page template information can represent the template meta-information of a visual page template. For example, page template information may include, but is not limited to: template theme, adaptation scenario, template component type, component location, and component style configuration. In particular, when the visual page template corresponding to the page template information contains multiple template components, it can contain multiple sets of template component types, component locations, and component style configurations.

[0059] In practice, when the page template information database stores the page template information corresponding to a new visual page template, the aforementioned execution entity will automatically call a large language model (e.g., GPT-4) to extract the page tag group included in the page template information. After manual review, it will trigger the generation of template vectors in the template vector database and write them into the template vector database.

[0060] In practice, when updating existing visual page templates, different versions of the visual page templates are assigned corresponding version numbers to support template rollback in case of anomalies in the latest visual page template. Simultaneously, redundant page template information in the page template database can be periodically cleared to eliminate redundant visual page templates. Specifically, redundant page template information is marked as not participating in template vector retrieval, thus achieving the purpose of redundancy removal.

[0061] As an example, see Figure 4 The diagram illustrates the relationship between the template vector library, the page template information library, and the visual page templates. The template vector library stores the template vectors corresponding to the visual page templates. The page template information library stores the page template information corresponding to the visual page templates. Template vectors and page template information corresponding to the same visual page template are associated through a page template identifier.

[0062] In some optional implementations of some embodiments, the above template vector library is constructed through the following steps:

[0063] Step S1: Read the above page template information database to obtain a set of page template information.

[0064] In practice, the aforementioned execution entity can read page template information from the page template information database that has not generated corresponding template vectors, and obtain the aforementioned set of page template information.

[0065] As an example, when the template vector library is built for the first time, the entire page template information library can be read to obtain a set of page template information. When the template vector library is updated, and / or when new page template information is stored in the page template information library, the page template information to be updated and / or the newly stored page template information are read from the aforementioned page template information library to form the set of page template information.

[0066] Step S2: For each page template in the above set of page template information, perform the following template vector generation steps:

[0067] Step S21: Extract tags from the above page template information to obtain page tag groups.

[0068] The aforementioned page tag group consists of a first page tag, a second page tag, and a third page tag. The first page tag represents the template theme corresponding to the visual page template, the second page tag represents the template adaptation scenario corresponding to the visual page template, and the third page tag represents the component type of the template components contained in the visual page template.

[0069] In practice, the template theme, adaptation scenarios, and template component types included in the page template information are often represented by different page tag values ​​depending on the corresponding content. Therefore, the aforementioned execution entity can extract the page tag values ​​corresponding to the template theme, adaptation scenarios, and template component types from the page template information and use them as the first page tag, the second page tag, and the third page tag to form the page tag group corresponding to the page template information.

[0070] Step S22: Clean the page tag group to obtain the cleaned page tag group.

[0071] In practice, the aforementioned execution entity can uniformly convert the letters contained in the first, second, and third page tags within the page tag group to lowercase; standardize the contained dates, numbers, etc.; and filter out the contained punctuation marks, special symbols, redundant spaces, and invisible characters, thereby obtaining a cleaned page tag group.

[0072] Step S23: Concatenate the cleaned page tags in the cleaned page tag group to obtain the tag string.

[0073] The tag string can be obtained by concatenating the cleaned page tags in the cleaned page tag group.

[0074] Step S24: Perform dense vector mapping on the above tag strings to obtain the template vector corresponding to the above page template information.

[0075] The template vector has a vector dimension of 768.

[0076] In practice, the aforementioned execution entity uses the Sentence-BERT pre-trained model to perform dense vector mapping on the tag strings, thereby automatically learning the linguistic features of the page tags and generating semantically strongly related vector representations, which serve as the template vectors corresponding to the aforementioned page template information.

[0077] Step S25: Associate the above page template information with the above template vector using the page template identifier of the visual page template corresponding to the above page template information.

[0078] Step S3: Based on the obtained template vector set, generate the vector library index corresponding to the above template vector library.

[0079] In practice, firstly, index types can be defined. Different index types correspond to different memory footprints. Specifically, index types can include: "FLAT", "IVF_FLAT", "IVF_SQ8", "IVF_PQ", "HNSW", and "SCANN". The "FLAT" type is primarily used for precise searches on small datasets and has a high memory footprint. The "IVF_FLAT" type is primarily used for general search scenarios and has a moderate memory footprint. The "IVF_SQ8" type is primarily used for memory-sensitive search scenarios and has a low memory footprint. The "IVF_PQ" type is primarily used for retrieval scenarios on very large datasets and has a very low memory footprint. The "HNSW" type is primarily used for high-performance, low-latency retrieval scenarios and has a high memory footprint. The "SCANN" type is primarily used for retrieval scenarios involving disk-based data scanning of large datasets and has a very low memory footprint. Specifically, this disclosure uses the "IVF_FLAT" type index to facilitate efficient similarity retrieval. Then, based on the index type and template vector set, a vector library index is generated. Finally, the generated vector library index is loaded into memory.

[0080] In some optional implementations of certain embodiments, the execution entity performs visual page template matching based on the aforementioned intent description information and a pre-built visual page template library to obtain a target page template, including:

[0081] Step S1: Convert the above intent description information into search text.

[0082] In practice, since the intent description information is stored in JSON format, the values ​​stored in JSON format within the intent description information can be extracted as retrieval text.

[0083] Step S2: Encode the above search text into a search vector to obtain the target search vector.

[0084] The target retrieval vector mentioned above is a 768-dimensional retrieval vector.

[0085] In practice, to ensure the consistency of the vector generation process and that the generated retrieval vectors are located in the same feature space, the Sentence-BERT pre-trained model is also used to encode the retrieval vectors of the above retrieval text to obtain the target retrieval vector.

[0086] Step S3: Perform vector matching based on the target retrieval vector and the corresponding vector library index of the template vector library to obtain a candidate template vector sequence.

[0087] Among them, the candidate template vectors in the above candidate template vector sequence are arranged in descending order of their vector matching degree with the target retrieval vector.

[0088] In practice, a cosine similarity vector matching method can be used. Based on the target retrieval vector and the vector library index, a library search is performed on the template vector library, and the top N candidate template vectors with the highest vector matching degree are selected and sorted in descending order to obtain the candidate template vector sequence. Here, N can be 5.

[0089] Step S4: Determine the candidate page template sequence based on the above candidate template vector sequence and the above page template information database.

[0090] Among them, the candidate template vector and the candidate page template are in one-to-one correspondence.

[0091] In practice, since the template vectors in the template vector library are associated with the corresponding page template information tubes in the page template information library through the page template identifier of the visual page template, the candidate page template sequence can be determined by a join query.

[0092] Step S6: In response to the fact that the vector matching degree corresponding to the first candidate template vector in the above candidate template vector sequence is greater than the preset matching degree, the first candidate page template in the above candidate page template sequence is determined as the above target page template.

[0093] The preset matching degree can be 0.85.

[0094] Step S7: In response to the fact that the vector matching degree of the first candidate template vector in the above candidate template vector sequence is less than or equal to the preset matching degree and the average vector matching degree of the candidate template vector sequence is less than or equal to the preset matching degree, the above target page template is generated according to the above candidate page template sequence.

[0095] The average vector matching degree represents the mean of the vector matching degrees corresponding to the N candidate template vectors in the candidate template vector sequence.

[0096] In practice, when the average vector matching degree is less than or equal to the preset matching degree, common template components in the candidate page template sequence can be extracted to construct a default page template, which can then be used as the target page template.

[0097] Step 103: Perform template parsing on the target page template to obtain a set of component description information.

[0098] In some embodiments, the aforementioned execution entity may perform template parsing on the target page template to obtain a set of component description information.

[0099] The component description information represents the template components contained within the target page template. The component description information may include: template component type, component location, and component style configuration.

[0100] In practice, to ensure effective data display in visualization page templates, templates often contain multiple template components with different data display types. Different visualization page templates typically have different template component compositions and varying component positions. Therefore, it is necessary to perform template parsing on the target page template to obtain a set of component description information. Specifically, this can be achieved through tree-based parsing to obtain the set of component description information.

[0101] In some optional implementations of certain embodiments, the execution entity performs template parsing on the target page template to obtain a set of component description information, including:

[0102] Step S1: Read the page template information corresponding to the target page template to obtain the template component list.

[0103] In practice, page template information is stored in JSON format, which contains information such as template component type, component location, and component style configuration. Therefore, the page template information corresponding to the target page template can be read by key-value search to obtain the template component list.

[0104] Step S2: Select template components that meet the filtering criteria from the above list of template components and use them as initial template components to obtain the initial component set.

[0105] The first filtering condition is as follows: the first attribute of the template component is true and the second attribute of the template component is not empty. The first attribute "needDataFill" indicates whether the template component needs data population, and its value is either "True" or "False". A value of "True" indicates that data population is required, and a value of "False" indicates that data population is not required. The second attribute "groupId" represents the data population identifier of the template component. Multiple template components within the same module that require data population have the same second attribute. When the second attribute is empty, it indicates a static page component, for which data population is not required. The filtering condition can be used to filter page components that are characterized as static.

[0106] Step S3: Group the initial components in the above initial component set to obtain an initial component group set.

[0107] In this group of initial components, the second property is the same for each initial component.

[0108] Step S4: For each initial component group in the initial component group set, perform the following validation steps:

[0109] Step S41: Perform integrity verification on the above initial component group.

[0110] In practice, integrity checks can be performed by determining whether the initial component group is empty or whether there are component conflicts between the initial components within the initial component group. Specifically, component conflicts can be automatically detected using predefined incompatible component types or conflict rules set according to business scenarios.

[0111] Step S42: In response to the above initial component group passing the integrity check, generate the component description information corresponding to the above initial component group in the above component description information.

[0112] In practice, when the initial component group is not empty and there are no component conflicts between the initial components within the initial component group, it indicates that the initial component group has passed the integrity check. At this time, component description information is generated based on the component parameters corresponding to the initial components within the initial component group (e.g., template component type, component position, and component style configuration).

[0113] In practice, an exception log is generated in response to the initial component group failing the integrity check, to support manual intervention and correction. Specifically, an initial component group failing the integrity check is indicated by an empty initial component group or component conflicts between initial components within the initial component group.

[0114] Step 104: Based on the intent description information, generate the indicator description information corresponding to each component description information in the component description information set to obtain the indicator description information set.

[0115] In some embodiments, the aforementioned execution entity can generate indicator description information corresponding to each component description information in the component description information set based on the intent description information, thereby obtaining the indicator description information set.

[0116] The indicator description information corresponds one-to-one with the component description information. The indicator description information can characterize the type of data to be displayed by the template component corresponding to the component description information, the data display method, and other data-related constraints (such as the range of data values).

[0117] In practice, since this disclosure guides target users to fully describe their needs for building a visualization page through intent-guided information, the intent description information can include constraints such as the data metrics to be displayed and the data display method. Simultaneously, different template components correspond to different displayable data types. Therefore, by constructing prompts based on the intent description information and component description information, a large language model (e.g., GPT-4) can be guided to generate corresponding metric description information.

[0118] In some optional implementations of certain embodiments, the execution entity generates indicator description information corresponding to each component description information in the component description information set based on the intent description information, including:

[0119] Step S1: Convert the format of the above component description information to generate the converted component description information.

[0120] In practice, component description information can be converted to Markdown format, which, compared to JSON format, retains only the information valuable for metric generation.

[0121] As an example, the formatted component description information may include: the second attribute "groupID" = M0001, which includes page components: title component, line chart component, and bar chart component.

[0122] Step S2: Construct indicator prompt information based on the above intent description information and the above format-converted component description information.

[0123] The indicator prompts are information used to guide the target inference model in content reasoning. These prompts can consist of role definition information, background information, constraints, and output format. Role definition information defines the virtual role the target inference model plays during content reasoning. Background information represents the business scenario corresponding to the generated indicator. Task instructions represent the specific indicator generation instructions. Constraints represent the constraints in the indicator generation process. Output format represents the format of the generated indicator description information.

[0124] In practice, role definition information can be pre-defined in visualization page generation scenarios. For example, role definition information could be: "You are a business metric design expert in the field of visualization page construction. You need to combine relevant information to build business metrics for multiple template components corresponding to the component description information." Background information and constraints can be extracted from intent description information. For example, background information could be: "Users need to generate a visualization page displaying quarterly e-commerce sales data." Constraints could be: "The metrics to be displayed include: 'Total sales revenue,' 'Total number of orders,' 'Gross profit margin,' 'Number of active users,' 'Trend between sales revenue and number of orders,' 'Monthly task completion progress,' and other metrics"; "'Total sales revenue,' 'Total number of orders,' 'Gross profit margin,' and 'Number of active users' should be displayed numerically"; "'Trend between sales revenue and number of orders' needs to be adapted to a double line chart"; "'Monthly task completion progress' needs to be adapted to a pie chart." Task instructions can be constructed based on intent description information and data display constraints of template components. For example, the task instruction could be: "Select an appropriate number of metrics for the template component corresponding to the component description information after this format conversion; supplement the corresponding adaptation reason for each selected metric, explaining why the metric matches the template component." The output format can be pre-set. For example, the output could be "constructed in JSON format, consisting of the template component's second attribute, the matched business metrics, and the adaptation reason." The output format could be:

[0125] {

[0126] groupID = "M0001";

[0127] BusinessMetrics = "Quarterly Sales", "Quarterly Orders";

[0128] Reason = "The template component includes a double line chart, and the constraint requirement is to use the double line chart to represent the changing trends of sales volume and order number. The line chart can intuitively describe the changing trends, and sales revenue and order number are the core indicators of quarterly sales data."

[0129] }

[0130] Step S4: Based on the above indicator prompts and the pre-trained target inference model, generate indicator description information corresponding to the above component description information.

[0131] In practice, the target inference model can employ large language models such as Qwen3-32B. Specifically, the target inference model can be deployed locally or remotely accessed via an API (Application Programming Interface). The execution entity can input indicator prompts into the target inference model to obtain indicator descriptions.

[0132] Step 105: Generate static fill data corresponding to each indicator description information in the indicator description information set to obtain the static fill data set.

[0133] In some embodiments, the aforementioned execution entity may generate static fill data corresponding to each indicator description information in the indicator description information set, thereby obtaining a static fill data set.

[0134] There is a one-to-one correspondence between the indicator description information and the statically populated data.

[0135] In practice, based on the known business indicator types and display methods described by the indicator description information, the target inference model can be further guided to generate business indicator data.

[0136] In some optional implementations of certain embodiments, the execution entity generates static fill data corresponding to each indicator description in the indicator description information set, including:

[0137] Step S1: For each indicator description in the above indicator description information set, generate static data filling prompt information based on the above indicator description information and the component description information corresponding to the above indicator description information.

[0138] The static data filling prompts can be information used to guide the target inference model in generating static data. These prompts can consist of role definition information, template component description information, task instructions, constraints, and output format. Role definition information defines the virtual role the target inference model plays in the static data filling process. Task instructions represent specific static data filling generation instructions. Constraints represent the constraints in the static data filling process. Output format represents the format of the generated static data.

[0139] In practice, role definition information can be pre-defined in visualization page generation scenarios. For example, role definition information could be "You are a static data generation expert; you need to combine relevant information to generate logically sound and formatted demonstration data." Constraints can be extracted from indicator description information. For example, "You need to generate sales data with a quarterly time dimension," "You need to generate order numbers with a quarterly time dimension," and "The generated sales data and order numbers correspond on the time scale." Task instructions could be "Generate sales data and order numbers corresponding to the time scale, with a quarterly time unit, respectively." Output format can be pre-set; for example, the output format could be "Output sales data and order data as two-dimensional arrays."

[0140] Step S2: Based on the static data filling prompts and the target inference model, generate the static data corresponding to the indicator description information.

[0141] In practice, the target inference model can employ large language models such as Qwen3-32B. Specifically, the target inference model can be deployed locally or remotely accessed via API calls to utilize the inference services of the large language model. The execution entity can input static data fill prompts into the target inference model to obtain the static data fill.

[0142] Step 106: Based on the static data set, populate the target page template with page data to obtain the populated page.

[0143] In some embodiments, the aforementioned execution entity can fill the target page template with page data based on a static data set to obtain the filled page.

[0144] In practice, since the target page template is composed of multiple template components, and each template component corresponds to a certain static fill data, the static fill data can be associated with the corresponding template component and injected to obtain the filled page.

[0145] As an example, static data can be associated with the corresponding template component based on its second attribute to enable data filling and location. Specifically, a manual verification process can be triggered when filling fails.

[0146] Step 107: Perform page integrity verification on the filled page, and then perform visual page rendering after the filled page passes the page integrity verification.

[0147] In some embodiments, the aforementioned execution entity may perform page integrity verification on the filled page, and perform visual page rendering after the filled page passes the page integrity verification.

[0148] In practice, the aforementioned execution entity can first determine whether there are page components in the populated page that have not been populated with static data, or page components that failed to populate and were not manually verified as complete. It can also determine whether the data type of the static data matches the data type of the business metrics displayed in the template components, thereby determining whether the populated page passes the page integrity check. Once the populated page passes the page integrity check, it is sent back to the front end for visualization rendering using the DOM (Document Object Model Tree) rendering method.

[0149] In practice, this method allows users to generate visualization pages automatically under natural language guidance without any coding experience or data analysis background, especially when the user has not prepared the data to be displayed for the page. This significantly shortens the development cycle of visualization pages and improves the speed of response to user needs. In particular, when the data to be displayed for the page is missing (for example, in the context of front-end and back-end separation development, the data to be displayed for the page and the visualization page often cannot be developed synchronously with strict granular alignment), it automatically generates reasonable and realistic data, thereby intuitively generating a usable visualization page, which facilitates subsequent solution design, proof of concept, and further communication and improvement of requirements.

[0150] The above embodiments of this disclosure have the following beneficial effects: The visualization page construction method based on page generation intent, as described in some embodiments of this disclosure, achieves automated visualization page construction based on intent understanding, effectively reducing the user's visualization page construction cost and thus compressing the development cycle of the visualization page. Specifically, existing methods such as "drag and drop page components," "low-code configuration," and "using existing BI" all require users to manually select page templates that match their needs and manually adjust components, map and associate data, and populate them. This is especially problematic for users who haven't prepared page filling data, requiring them to spend extra time collecting this data and preventing direct previewing of the visualization page. Therefore, the visualization page construction method based on page generation intent, as described in some embodiments of this disclosure, firstly collects the target user's page generation intent in a guided manner to obtain intent description information. In practice, visualization pages mainly display key data in the form of charts and text. Furthermore, the page structure and content of visualization pages vary depending on different display requirements. Therefore, this disclosure, through guided page generation intent collection, enables target users to comprehensively describe their visualization page construction needs using natural language. Secondly, based on the aforementioned intent description information and the pre-built visual page template library, visual page template matching is performed to obtain the target page template. In practice, conventional methods for page template matching mainly involve matching user-input template tags with preset template tags. However, when the two are inconsistent or there is no corresponding preset template tag, matching will fail. Furthermore, tag matching alone is insufficient to extract implicit page construction requirements. Therefore, this disclosure combines intent description information with template matching to improve the accuracy and success rate of matching. Next, the target page template is parsed to obtain a set of component description information, where the component description information represents the template components contained within the target page template. In practice, different page templates may contain different template components, and the data configurations of different template components may also differ. Therefore, this disclosure automatically extracts the component configurations of the template components contained in the target page template through template parsing. Further, based on the aforementioned intent description information, indicator description information corresponding to each component description information in the aforementioned component description information set is generated to obtain an indicator description information set. In addition, static fill data corresponding to each indicator description information in the above indicator description information set is generated to obtain a static fill data set, wherein the static fill data is bound one-to-one with the template component corresponding to the component description information.In practice, to ensure effective data display on visualization pages, conventional generation methods often require users to collect the page data to be displayed in advance and adjust the format of the page data to ensure compatibility with the corresponding template components. This significantly increases the user's visualization page construction cost. Therefore, this disclosure uses intent description information and indicator description information as constraints to automatically generate static data. Next, based on the aforementioned static data set, the target page template is populated with page data to obtain the populated page. Finally, the populated page undergoes page integrity verification, and the visualization page is rendered only after the populated page passes the integrity verification. In summary, this method effectively reduces the user's visualization page construction cost, thereby shortening the visualization page development cycle.

[0151] Further reference Figure 5 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a visual page construction method apparatus based on page generation intent. These apparatus embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this visual page building method apparatus based on page generation intent can be specifically applied to various electronic devices.

[0152] like Figure 5As shown, some embodiments of a visual page construction method apparatus 500 based on page generation intent include: an intent acquisition unit 501, a visual page template matching unit 502, a template parsing unit 503, a first generation unit 504, a second generation unit 505, a page data filling unit 506, and a page integrity verification and visual page rendering unit 507. The intent acquisition unit 501 is configured to collect the page generation intent of a target user in a guided manner to obtain intent description information; the visual page template matching unit 502 is configured to perform visual page template matching based on the intent description information and a pre-built visual page template library to obtain a target page template; the template parsing unit 503 is configured to perform template parsing on the target page template to obtain a set of component description information, wherein the component description information represents the template components contained in the target page template; the first generation unit 504 is configured to generate a pointer corresponding to each component description information in the set of component description information based on the intent description information. The system generates a set of indicator description information. A second generation unit 505 is configured to generate static fill data corresponding to each indicator description information in the set, resulting in a static fill data set. The static fill data is bound one-to-one with the template component corresponding to the component description information. A page data fill unit 506 is configured to fill the target page template with page data based on the static fill data set, resulting in a filled page. A page integrity verification and visualization page rendering unit 507 is configured to perform page integrity verification on the filled page and, after the filled page passes the page integrity verification, perform visualization page rendering.

[0153] It is understandable that the units described in the visual page building method apparatus 500 based on page generation intent are related to the reference Figure 1 The steps described in the method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the visual page construction method apparatus 500 based on page generation intent and the units contained therein, and will not be repeated here.

[0154] The following is for reference. Figure 6 It shows a schematic diagram of the structure of an electronic device (e.g., a computing device) 600 suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0155] like Figure 6As shown, the electronic device 600 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory 602 or a program loaded from a storage device 608 into a random access memory 603. The random access memory 603 also stores various programs and data required for the operation of the electronic device 600. The processing unit 601, the read-only memory 602, and the random access memory 603 are interconnected via a bus 604. An input / output interface 605 is also connected to the bus 604.

[0156] Typically, the following devices can be connected to the input / output interface 605: input devices 606 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 608 including, for example, magnetic tape, hard disk, etc.; and communication devices 609. Communication device 609 allows electronic device 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 An electronic device 600 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 6 Each box shown can represent a device or multiple devices as needed.

[0157] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a read-only memory 602. When the computer program is executed by the processing device 601, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0158] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0159] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0160] The aforementioned computer-readable medium may be included within the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: collect the target user's guided page generation intent to obtain intent description information; match the intent description information with a pre-built visual page template library to obtain a target page template; parse the target page template to obtain a set of component description information, wherein the component description information represents the template components contained within the target page template; generate indicator description information corresponding to each component description information in the component description information set based on the intent description information, obtaining an indicator description information set; generate static fill data corresponding to each indicator description information in the indicator description information set, obtaining a static fill data set, wherein the static fill data is bound one-to-one with the template components corresponding to the component description information; fill the target page template with page data based on the static fill data set to obtain a filled page; perform page integrity verification on the filled page, and render the visual page after the filled page passes the page integrity verification.

[0161] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0162] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0163] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0164] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A visual page construction method based on page generation intent, characterized in that, include: Guided page generation intent collection for target users to obtain intent description information; The target page template is obtained by matching the intent description information with a pre-built visual page template library. The target page template is parsed to obtain a set of component description information, wherein the component description information represents the template components contained in the target page template; Based on the intent description information, generate indicator description information corresponding to each component description information in the component description information set to obtain the indicator description information set. Static fill data is generated for each indicator description in the indicator description information set to obtain a static fill data set, wherein the static fill data is bound one-to-one with the template component corresponding to the component description information. Based on the static data set, the target page template is populated with page data to obtain the populated page; The filled page undergoes a page integrity check, and after the filled page passes the page integrity check, a visual page rendering is performed. The process of guiding the target user to generate page intent and collecting intent description information includes: In response to the target user entering real-time intent text in the text input box, the following processing steps are performed based on the real-time intent text, wherein the real-time intent text ends with a first terminator: The real-time intent text is preprocessed to obtain a filtered word set; Each filtered word in the filtered word set is word-encoded to generate a word vector, resulting in a word vector set; In response to an empty historical intent description information queue, based on the word vector set and the pre-trained intent recognition model, real-time intent description information and intent guidance information corresponding to the real-time intent text are generated. The historical intent description information sequence is used to store intent description information corresponding to the real-time intent text previously input by the target user in the current question-and-answer round. The real-time intent description information represents the page generation intent corresponding to the real-time intent text. The intent guidance information represents the guiding text used to guide the target user to further supplement new real-time intent text. The intent recognition model adopts the ERNIE3.0 Small model. In response to the historical intent description information queue being non-empty, based on the historical intent features corresponding to the historical intent description information in the historical intent description information queue, the word vector set, and the pre-trained intent recognition model, real-time intent description information and intent guidance information corresponding to the real-time intent text are generated, and the real-time intent description information corresponding to the real-time intent text is added to the historical intent description information queue as historical intent description information. The intent guidance information is displayed in the text input box, wherein the intent guidance information is used to guide the text input of the target user, and the intent guidance information is appended after the real-time intent text; In response to the target user not entering new real-time intent text in the text input box within a preset time period, or the target user entering a second end symbol in the text input box, the intent description information is generated according to the historical intent description information queue. The second end symbol is a predefined custom symbol used to indicate that the page generation intent has been fully collected. The step of generating static fill data corresponding to each indicator description in the indicator description information set includes: For each indicator description in the indicator description information set, static data filling prompt information is generated based on the indicator description information and the component description information corresponding to the indicator description information. The static data filling prompt information can be composed of role definition information, template component description information, task instructions, constraints, and output format. The role definition information is used to limit the virtual role played by the target inference model in the static data filling process. The task instructions represent specific static data filling instructions. The constraints represent the constraints in the static data filling process. The output format represents the format of the generated static data filling. Based on the static data filling prompt information and the target inference model, static data corresponding to the indicator description information is generated. 2.The page-based intent visualization page building method of claim 1, wherein, The visual page template library consists of a template vector library and a page template information library. The template vector library stores the template vectors corresponding to the visual page templates, and the page template information library stores the page template information corresponding to the visual page templates. The template vectors and page template information corresponding to the same visual page template are associated through a page template identifier. 3.The page-based intent visualization page building method of claim 2, wherein, The template vector library is constructed through the following steps: Read the page template information database to obtain a set of page template information; For each page template information in the page template information set, perform the following template vector generation steps: Tag extraction is performed on the page template information to obtain a page tag group, wherein the page tag group consists of a first page tag, a second page tag and a third page tag. The first page tag represents the template theme corresponding to the visual page template, the second page tag represents the template adaptation scenario corresponding to the visual page template, and the third page tag represents the component type of the template components contained in the visual page template. Perform tag cleaning on the page tag group to obtain the cleaned page tag group; The cleaned page tags in the cleaned page tag group are concatenated to obtain a tag string; Perform dense vector mapping on the tag string to obtain the template vector corresponding to the page template information; The page template information is associated with the template vector through the page template identifier of the visual page template corresponding to the page template information; Based on the obtained set of template vectors, generate the vector library index corresponding to the template vector library.

4. The method of claim 3, wherein, The step of matching the target page template based on the intent description information and a pre-built visual page template library includes: Convert the intent description information into search text; The search text is encoded into a search vector to obtain a target search vector, wherein the target search vector is a 768-dimensional search vector; Vector matching is performed based on the target retrieval vector and the vector library index corresponding to the template vector library to obtain a candidate template vector sequence, wherein the candidate template vectors in the candidate template vector sequence are arranged in descending order of vector matching degree with the target retrieval vector; Based on the candidate template vector sequence and the page template information database, a candidate page template sequence is determined, wherein the candidate template vector and the candidate page template correspond one-to-one; In response to the fact that the vector matching degree corresponding to the first candidate template vector in the candidate template vector sequence is greater than the preset matching degree, the first candidate page template in the candidate page template sequence is determined as the target page template; In response to the fact that the vector matching degree corresponding to the first candidate template vector in the candidate template vector sequence is less than or equal to a preset matching degree and the average vector matching degree corresponding to the candidate template vector sequence is less than or equal to a preset matching degree, the target page template is generated according to the candidate page template sequence.

5. The method of claim 4, wherein, The step of parsing the target page template to obtain a set of component description information includes: Read the page template information corresponding to the target page template to obtain a list of template components; Template components that meet the filtering criteria are selected from the list of template components to form an initial set of template components. The filtering criteria are: the first attribute of the template component is true and the second attribute of the template component is not empty. The initial components in the initial component set are grouped to obtain an initial component group set, wherein each initial component in the initial component group has the same second attribute; For each initial component group in the initial component group set, perform the following validation steps: Perform integrity verification on the initial component group; In response to the initial component group passing the integrity check, component description information corresponding to the initial component group is generated in the component description information. 6.The method of claim 5, wherein, The step of generating metric description information corresponding to each component description information in the component description information set based on the intent description information includes: The component description information is format-converted to generate format-converted component description information; Based on the intent description information and the format-converted component description information, construct indicator prompt information; Based on the indicator prompts and the pre-trained target inference model, indicator description information corresponding to the component description information is generated.

7. A visual page construction method apparatus based on page generation intent, characterized in that, include: The intent collection unit is configured to guide the target user in page generation intent collection to obtain intent description information. The visual page template matching unit is configured to perform visual page template matching based on the intent description information and a pre-built visual page template library to obtain the target page template; The template parsing unit is configured to parse the target page template to obtain a set of component description information, wherein the component description information represents the template components contained in the target page template; The first generation unit is configured to generate indicator description information corresponding to each component description information in the component description information set according to the intent description information, thereby obtaining the indicator description information set; The second generation unit is configured to generate static fill data corresponding to each indicator description information in the indicator description information set, thereby obtaining a static fill data set, wherein the static fill data is bound one-to-one with the template component corresponding to the component description information. The page data filling unit is configured to fill the target page template with page data according to the static filling data set to obtain the filled page; The page integrity verification and visualization page rendering unit is configured to perform page integrity verification on the filled page, and to perform visualization page rendering after the filled page passes the page integrity verification. The process of guiding the target user to generate page intent and collecting intent description information includes: In response to the target user entering real-time intent text in the text input box, the following processing steps are performed based on the real-time intent text, wherein the real-time intent text ends with a first terminator: The real-time intent text is preprocessed to obtain a filtered word set; Each filtered word in the filtered word set is word-encoded to generate a word vector, resulting in a word vector set; In response to an empty historical intent description information queue, based on the word vector set and the pre-trained intent recognition model, real-time intent description information and intent guidance information corresponding to the real-time intent text are generated. The historical intent description information sequence is used to store intent description information corresponding to the real-time intent text previously input by the target user in the current question-and-answer round. The real-time intent description information represents the page generation intent corresponding to the real-time intent text. The intent guidance information represents the guiding text used to guide the target user to further supplement new real-time intent text. The intent recognition model adopts the ERNIE3.0 Small model. In response to the historical intent description information queue being non-empty, based on the historical intent features corresponding to the historical intent description information in the historical intent description information queue, the word vector set, and the pre-trained intent recognition model, real-time intent description information and intent guidance information corresponding to the real-time intent text are generated, and the real-time intent description information corresponding to the real-time intent text is added to the historical intent description information queue as historical intent description information. The intent guidance information is displayed in the text input box, wherein the intent guidance information is used to guide the text input of the target user, and the intent guidance information is appended after the real-time intent text; In response to the target user not entering new real-time intent text in the text input box within a preset time period, or the target user entering a second end symbol in the text input box, the intent description information is generated according to the historical intent description information queue. The second end symbol is a predefined custom symbol used to indicate that the page generation intent has been fully collected. The step of generating static fill data corresponding to each indicator description in the indicator description information set includes: For each indicator description in the indicator description information set, static data filling prompt information is generated based on the indicator description information and the component description information corresponding to the indicator description information. The static data filling prompt information can be composed of role definition information, template component description information, task instructions, constraints, and output format. The role definition information is used to limit the virtual role played by the target inference model in the static data filling process. The task instructions represent specific static data filling instructions. The constraints represent the constraints in the static data filling process. The output format represents the format of the generated static data filling. Based on the static data filling prompt information and the target inference model, static data corresponding to the indicator description information is generated.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

Images