Data processing methods, apparatus, storage media, electronic devices and software products
By performing layout analysis and text recognition on images, differentiating between locked and rewritten text fields, and using a large language model for text rewriting, the problem of insufficient information extraction quality is solved, thereby improving the accuracy and diversity of information dissemination.
Patent Information
- Application Number
- CN202611264686.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
In information publishing platforms, existing technologies struggle to effectively extract structured information from images, resulting in insufficient information publishing quality and content diversity, and a tendency for platform content to become homogenized.
By performing layout analysis and text recognition on the image, the locked text field and the rewritten text field are distinguished. The large language model is used to rewrite the text, ensuring that the rewritten descriptive text meets the expression norms and differentiation requirements, and generating the target structured data.
It has improved the accuracy and uniqueness of information dissemination, curbed content homogenization, and enhanced the quality and diversity of information dissemination.
Smart Images

Figure CN123115085A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image processing technology, and in particular to a data processing method, apparatus, storage medium, electronic device, and program product. Background Technology
[0002] In various information publishing platforms, users commonly use image materials containing structured information (such as posters, screenshots, flyers, etc.) as raw data sources to generate online content. When generating online content, it is necessary to extract structured information from the images; ensuring the quality of this extraction is a pressing issue. Summary of the Invention
[0003] This application provides a data processing method, apparatus, storage medium, electronic device, and program product to improve the quality and diversity of information dissemination.
[0004] This application provides the following solution: According to the first aspect, a data processing method is provided, including: Acquire the target image; Information is extracted from the target image to generate initial structured data; Text field recognition is performed on the initial structured data to obtain locked text fields and rewritten text fields. Locked text fields carry factual parameter information, while rewritten text fields carry descriptive text. Rewrite the rewritten text field in the initial structured data to obtain rewritten descriptive text that meets the expression specifications and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than the similarity threshold. By using the locked text field and the rewritten text, the target structured data is obtained.
[0005] Optionally, text field recognition is performed on the initial structured data to obtain locked text fields and rewritten text fields, including: Based on the keys of each business field in the initial structured data, the preset first mapping relationship configuration table is queried to determine the field attributes corresponding to each business field. The first mapping relationship configuration table records the correspondence between the keys of the business fields and the field attributes. The field attributes include locking field attributes or rewriting field attributes. Based on the field attributes corresponding to each business field, determine the locked text field and the rewritten text field.
[0006] Optionally, the rewritten text field in the initial structured data is rewritten to obtain rewritten descriptive text that conforms to expression standards and / or differentiation requirements, including: The second largest language model is used to rewrite the text fields based on expression norms and synonym conversion rules; The similarity between the rewritten descriptive text and the existing descriptive text is calculated. Rewrite the rewritten description texts with similarity greater than the similarity threshold until all rewritten description texts meet the differentiation requirements.
[0007] Optionally, information is extracted from the target image to generate initial structured data, including: The layout of the target image is analyzed to obtain the position information of multiple blocks in the target image; Text recognition is performed on each block to obtain the text content of each block; Using the first major language model, based on the text content of each block and the location information of the block to which the text content belongs in the target image, information is extracted from preset fields for each block to obtain structured data for each block; the structured data of each block is used to obtain initial structured data.
[0008] Optionally, the target image is parsed to obtain the position information of multiple blocks in the target image, including: The target image is parsed using a page segmentation model to obtain multiple blocks included in the target image, the position information of each block in the target image, and the type information of each block. Based on the type information of multiple blocks, identify one or more blocks that are associated with the target business; Based on the location information of one or more blocks associated with the target service in the target image, the target image is cropped to obtain the image blocks corresponding to each block associated with the target service. Text recognition is performed on each block to obtain the text content of each block, including: optical character recognition is performed on the blocks corresponding to each target business to obtain the text content of each target business-related block.
[0009] Optionally, text recognition is performed on each block separately to obtain the text content of each block, including: Text recognition is performed on each block separately to obtain multiple candidate texts for each block and font size weight labels for each candidate text; Based on font size weight labels, candidate texts that exceed the preset font size weight range are filtered out from multiple candidate texts to obtain the text content of each block.
[0010] Optionally, the method further includes: Perform compliance semantic checks on the target structured data; If the detection results indicate that there are compliance risks in the target structured data, risk marking is performed on the text fields in the target structured data that have compliance risks; The risk-marked target structured data is sent to the user's terminal.
[0011] According to the second aspect, a data processing method is provided, including: Obtain the target image specified by the user and provide the target image to the server; The target structured data obtained from the target image is obtained from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold. Display the information publishing page and populate the target structured data into the multiple fields contained in the information publishing page accordingly; In response to the detection of an action that triggers information publishing, information is published based on the information publishing page.
[0012] Optionally, acquiring the target image determined by the user and providing the target image to the server includes: displaying the target image determined by the user and a first component for triggering information recognition on the job editing page; and providing the target image to the server in response to the operation of triggering the first component. Obtain target structured data based on the target image from the server, including: obtaining target structured data based on the target image from the server and displaying the rewritten descriptive text in the editable text box on the job editing page.
[0013] Optionally, text fields in the target structured data that pose compliance risks are marked with risk tags; The target structured data is populated into multiple fields on the information publishing page, including: based on a preset second mapping relationship configuration table, the text fields in the target structured data are populated into multiple target fields on the information publishing page, and text fields with compliance risks are displayed with a style or mark that is different from other text fields to prompt users to modify the text fields with compliance risks; wherein, the second mapping relationship configuration table includes the correspondence between text fields and target fields on the information publishing page.
[0014] According to a third aspect, a data processing apparatus is provided, comprising: The image acquisition unit is configured to acquire the target image; The information extraction unit is configured to extract information from the target image and generate initial structured data; The field recognition unit is configured to perform text field recognition on the initial structured data to obtain locked text fields and rewritten text fields. The locked text fields carry factual parameter information, and the rewritten text fields carry descriptive text. The text rewriting unit is configured to rewrite the rewriting text field in the initial structured data to obtain rewritten descriptive text that conforms to the expression specifications and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than the similarity threshold. The data acquisition unit is configured to obtain the target structured data using a locked text field and the rewritten text.
[0015] According to a fourth aspect, a data processing apparatus is provided, comprising: The image providing unit is configured to acquire a target image determined by the user and provide the target image to the server. The data acquisition unit is configured to acquire target structured data based on the target image from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold. The page display unit is configured to display the information publishing page and populate the target structured data into the multiple fields contained in the information publishing page. The information publishing unit is configured to publish information based on the information publishing page in response to the detection of an operation that triggers information publishing.
[0016] According to a fifth aspect, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method of any one of the first aspects described above.
[0017] According to a sixth aspect, an electronic device is provided, comprising: One or more processors; and A memory associated with one or more processors, the memory being used to store program instructions, which, when read and executed by one or more processors, perform the steps of the method described in any of the first aspects above.
[0018] According to the specific embodiments provided in this application, the following technical effects are disclosed: By distinguishing text fields in the initial structured data into locked text fields and rewritten text fields, and rewriting the rewritten text fields to conform to expression standards and / or differentiation requirements, the content carrying descriptive text is standardized and differentiated without modifying the locked text fields carrying factual parameter information. This ensures the accuracy of the core content in the published information, while improving the standardization and uniqueness of the descriptive text. It effectively suppresses the problem of platform content homogenization caused by the simple reuse of image materials, and improves the quality and diversity of information publication.
[0019] Of course, any product implementing this application does not necessarily need to achieve all of the above advantages at the same time. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a system architecture diagram applicable to the embodiments of this application.
[0022] Figure 2 A flowchart illustrating the data processing method performed by the data processing apparatus provided in this application embodiment.
[0023] Figure 3 A flowchart illustrating the data processing method executed by the client in an embodiment of this application.
[0024] Figure 4 This is an interactive schematic diagram of the data processing method according to an embodiment of this application.
[0025] Figure 5 This is a schematic block diagram of a server-side data processing apparatus provided in an embodiment of this application.
[0026] Figure 6 A schematic block diagram of a client-side data processing apparatus provided in another embodiment of this application.
[0027] Figure 7 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.
[0029] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.
[0030] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0031] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."
[0032] For scenarios involving extracting information from image materials and generating structured content for publication, the common approach in related technologies is to use an optical character recognition (OCR) model to extract text from the input target image, obtaining plain text information. This plain text string is then input into a general language model for semantic analysis, or the plain text information is directly entered manually by the user. However, the plain text information obtained in this way is a mixture of all information and highly similar content. Furthermore, core information is easily distorted by subjective or automated modifications, affecting the quality and diversity of information publication.
[0033] In view of this, this application provides a new approach. To facilitate understanding of this application, the system architecture on which this application is based will first be described. Figure 1 This is an exemplary system architecture diagram of an embodiment of this application, such as... Figure 1 As shown, the system architecture may include: a client and a data processing device set on the server side.
[0034] The server and client are the two main components of an application service. The server, using a server as its primary hardware infrastructure, may include one or more software service modules. In this embodiment, it is mainly responsible for handling the receiving and sending binding relationships, forming a collaborative front-end and back-end with the client.
[0035] The user terminal where the client resides can include, but is not limited to, smart mobile terminals, wearable devices, and PCs (Personal Computers). Smart mobile devices can include devices such as mobile phones, tablets, PDAs (Personal Digital Assistants), and connected car terminals. Wearable devices can include devices such as smartwatches, smart glasses, smart bracelets, VR (Virtual Reality) devices, AR (Augmented Reality) devices, and mixed reality devices (i.e., devices that support both virtual and augmented reality), etc.
[0036] The client can be an application located on the user's terminal, or it can be a plugin or software development kit (SDK) or other functional unit set in the application.
[0037] A server can be a standalone server, a server cluster, or a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a hosting product within the cloud computing service system, designed to address the shortcomings of traditional physical hosts and Virtual Private Servers (VPS) services, such as high management difficulty and weak service scalability.
[0038] In one scenario, the client can send a target image to the server and receive data processing results or instructions from the server. The server-side data processing device can perform data processing using the methods provided in the embodiments of this application, as detailed in the relevant descriptions in subsequent embodiments. It should be understood that... Figure 1 The server and client shown are merely illustrative. Depending on the implementation requirements, there can be any number of servers and clients.
[0039] Figure 2 This is a flowchart of a data processing method executed by the data processing apparatus provided in this application embodiment. The method can be performed by… Figure 1 The data processing unit in the system shown performs the operation. For example... Figure 2 As shown, the method may include the following steps: Step 201: Obtain the target image.
[0040] Step 203: Extract information from the target image to generate initial structured data.
[0041] Step 205: Perform text field recognition on the initial structured data to obtain locked text fields and rewritten text fields. Locked text fields carry factual parameter information, while rewritten text fields carry descriptive text.
[0042] Step 207: Rewrite the rewritten text field in the initial structured data to obtain rewritten descriptive text that meets the expression specifications and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than the similarity threshold.
[0043] Step 209: Use the locked text field and the rewritten text to obtain the target structured data.
[0044] As can be seen from the above process, this application distinguishes the text fields in the initial structured data into locked text fields and rewritten text fields, and rewrites the rewritten text fields in accordance with expression specifications and / or differentiation requirements. Under the premise that the locked text fields carrying factual parameter information are not modified, the content carrying descriptive text is standardized and differentiated, thereby ensuring the accuracy of core facts in the published information, while improving the standardization and uniqueness of descriptive text, effectively suppressing the problem of platform content homogenization caused by simple reuse of image materials, and improving the quality and diversity of information publication.
[0045] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments. It should be noted that the terms "first" and "second" involved in this disclosure do not have limitations in terms of size, order, or quantity, but are only used to distinguish them by name. For example, "first mapping relationship configuration table" and "second mapping relationship configuration table" are used to distinguish two mapping relationship configuration tables.
[0046] First, step 201, namely "acquiring the target image", will be described in detail with reference to the embodiments.
[0047] In this application, "target image" can broadly refer to image data carrying business information to be extracted. It can be a poster, screenshot, or flyer in the labor recruitment field, or various images from other business fields that require the extraction of business information from unstructured visual materials. For example, it could be a poster, screenshot, or flyer from the e-commerce, real estate rental and sales, or catering and lifestyle service sectors. The target image may include multiple information areas related to the target business. This application does not limit the application field for acquiring the target image.
[0048] The acquired target image can be stored in formats such as JPEG, PNG, and WEBP. Alternatively, it can be compressed, resized, or formatted by the client before being transmitted to the server. In practical implementation, the target image can contain a single information display block, or a complex layout with multiple blocks, areas arranged side-by-side, or columns. For example, a job posting might contain blocks displaying multiple job openings such as "General Worker," "Skilled Worker," and "Temporary Worker," with each block containing corresponding job information for each position.
[0049] The target image can be obtained by the client taking a photo and uploading it immediately, or by the client selecting an existing image from the local album and uploading it, or by the client reading an image from the clipboard, in-app screenshot cache, message attachment, or cloud material library and uploading it.
[0050] After acquiring the target image, before extracting information from it, preprocessing operations can be performed. Specifically, perspective transformation can be applied to tilted or distorted images captured by mobile phones to correct the direction of text lines; grayscale conversion, binarization, reflection suppression, and local contrast enhancement can be used to remove paper wrinkles, reflective spots, and background noise, while brightening light-colored small text; a preliminary judgment can be made on whether the image is relevant to the target business domain. If the judgment result indicates that the image has a low degree of relevance to the target business, a prompt message can be returned to guide the user to re-upload an image that matches the business scenario corresponding to the target business domain.
[0051] The following describes step 203, namely "extracting information from the target image and generating initial structured data", in detail with reference to the embodiments.
[0052] In this application, information extraction can be a composite processing procedure from unstructured visual images to structured business data. It can involve combining visual layout information from the target image, text recognition results, and business field extraction logic to form initial structured data corresponding to the business fields of the current business scenario.
[0053] The initial structured data can be the first structured data obtained after the information extraction described above, which maps images to fields. This initial structured data can be in the form of JSON key-value pairs, a dictionary, or a data table. The keys of the business fields are typically predefined as standardized field names matching the business scenario corresponding to the target business domain, while the values are the raw information identified from the image. For example, in the recruitment field, the keys of the business fields could be "Job Title," "Salary Range," "Job Responsibilities," "Benefits," etc.; the values corresponding to the keys of each business field could be "General Worker," "xxx-xxx Yuan / Month," "Assembly Line Product Assembly...", "Insurance and Accommodation Included," etc.
[0054] The following describes in detail step 205, namely "recognizing text fields in the initial structured data to obtain locked text fields and rewritten text fields", with reference to the embodiments.
[0055] In this application, locked text fields carry factual parameter information, while rewritten text fields carry descriptive text. Factual parameter information can be extracted from the target image, reflecting objective reality and should not be arbitrarily modified; this type of information is deterministic and verifiable. Locked text fields can be completely locked fields or restricted fields that only allow format standardization but not numerical changes. For example, in a recruitment scenario, the text fields corresponding to "job title," "salary range," "salary structure," "age requirement," and "employment period" in a recruitment poster are all factual parameter information.
[0056] Descriptive text can be extracted from a target image and is used to explain, elaborate, or introduce, but it is not immutable. This type of information is usually presented as narrative paragraphs, and its semantics can be optimized, worded consistently, restructured, or replaced with synonyms within a certain range, as long as the basic business meaning remains unchanged. For example, in a recruitment scenario, the text fields corresponding to "job responsibilities," "benefits," and "job description" in a recruitment poster are all descriptive text.
[0057] Text field recognition can be a process of classifying the attributes of various business fields in the initial structured data to determine whether each field belongs to a locked type (prohibited from modification) or a rewrite type (allowed for modification) in subsequent processing. Its recognition determines the processing permissions of business fields in the initial structured data, that is, whether the business field carries factual parameter information or descriptive text, rather than the recognition of the text characters themselves.
[0058] The following describes in detail step 207, namely, "rewriting the rewritten text field in the initial structured data to obtain rewritten descriptive text that conforms to expression standards and / or differentiation requirements," with reference to an embodiment.
[0059] In this application, the differentiation requirement includes a similarity threshold between the rewritten descriptive text and the existing descriptive text. The existing descriptive text can be existing descriptive text of the same type generated by historical publishing operations. The similarity threshold can be a fixed threshold or dynamically adjusted based on the business category of the target business domain, geographical context, or publishing time.
[0060] Expression standards can be a set of pre-defined textual language rules designed for industry expression habits or platform standardization requirements within a target business domain. Specific forms may include, but are not limited to: industry standard terminology replacement rules, sentence integrity rules, colloquial expression correction rules, risky wording avoidance rules, and text length control rules. Different target business domains can each maintain their own set of expression standard rule libraries to adapt to their unique industry terminology systems and publishing standards.
[0061] Differentiation requirements can refer to the similarity constraints that the rewritten descriptive text must meet compared to existing published descriptive text of the same type. The core of this requirement is to stipulate a preset similarity threshold between the rewritten descriptive text and existing descriptive text to ensure that each rewritten descriptive text has sufficient distinguishability from similar existing descriptive text.
[0062] Text rewriting refers to the optimization and reconstruction of rewritten text fields in initial structured data at the natural language processing level, ensuring that the original basic business meaning remains unchanged while meeting the requirements of expression standardization and differentiation. Text rewriting is not an arbitrary modification of field values in the initial structured data or unfounded generative fabrication, but rather a standardization process that focuses solely on descriptive text, including sentence optimization, wording consistency, sentence restructuring, synonym replacement, or adjustment of narrative perspective, while clearly defining the protection boundaries of text fields.
[0063] Text rewriting can be accomplished simultaneously with normalization and differentiation in a single inference process using a single language model; it can also be done by first performing colloquialization correction through a rule engine and then performing sentence reconstruction by a language model; or it can be accomplished by combining a template library, an industry terminology lexicon, and a language model.
[0064] The following describes step 209, namely "obtaining target structured data by using locked text fields and rewritten text," in detail with reference to an embodiment.
[0065] In this application, the target structured data may include locked text fields that retain the original factual semantics and rewritten descriptive text that has been rewritten using language normalization and verified for existing similarity. This target structured data can be directly used for subsequent form filling, page display, compliance testing, and information dissemination.
[0066] After rewriting the rewritten text fields in the initial structured data, the set of locked text fields and the set of rewritten descriptive text can be obtained respectively. Then, based on the preset target structured data pattern, the original value of the locked text fields is directly written into the corresponding business field position, and the rewritten descriptive text is written into the corresponding descriptive class field position. At the same time, other meta-information in the initial structured data (such as task identifier, timestamp, block source, etc.) is retained, and finally a complete target structured data that can be directly consumed by downstream users is generated.
[0067] In one embodiment, information extraction from the target image to generate initial structured data includes: parsing the layout of the target image to obtain the position information of multiple blocks in the target image; performing text recognition on each block to obtain the text content of each block; using a first language model to extract information from preset fields for each block based on the text content of each block and the position information of the block to which the text content belongs in the target image to obtain structured data for each block; and using the structured data of each block to obtain initial structured data.
[0068] In this application, layout analysis can be a process of visually analyzing a target image, automatically dividing the image into multiple blocks with independent information boundaries according to semantic integrity, and obtaining the spatial coordinates of each block in the image. Its purpose is not to identify specific text content in the target image, but rather to understand the physical layout structure of the target image, that is, to identify the location of each block in the target image and the type information of the block.
[0069] The multiple blocks obtained through layout analysis can be content blocks with independent semantics that are divided from the target image, each carrying a relatively independent and complete type of information. For example, in a recruitment scenario, a recruitment poster can be divided into a job promotion area and a job information area. The job information area carries information such as the job title, job responsibilities, and job requirements; the job promotion area carries promotional information about the recruitment for that job.
[0070] The location information of each block in the target image can refer to the data used to describe the spatial location of each block in the target image, including but not limited to the coordinate frame of the block (such as the pixel coordinates of the upper left and lower right corners of the rectangle boundary), the area ratio of the block, or the relative positional relationship between blocks (such as vertical arrangement, horizontal arrangement, etc.).
[0071] After the aforementioned page layout analysis yields multiple blocks, text recognition processing can be performed on each block to obtain the text content contained within that block. This is not merely a sequence of recognized pure characters, but a composite information set integrating semantic and visual spatial attributes. Specifically, the text content can include the recognized text content, its spatial coordinates, block identifiers, and visual attribute information. The spatial coordinates can be the specific location coordinates of the text content within the target image; the visual attribute information can be the font size weights, used to distinguish different visual levels of text, such as main titles, body text, and smaller notes.
[0072] Predefined fields can be a predefined set of standardized information units required for information dissemination in a target business domain. They can be represented as a series of predefined business field keys, each corresponding to a specific business information type, collectively forming a structured data output framework. The specific business information type refers to the semantic category represented by the key of that business field, i.e., which type of business meaning that the business field is specifically responsible for carrying. For example, in a recruitment scenario, if the key of a business field in the predefined fields is "job title," its corresponding specific business information type is "description of information related to job title and position name."
[0073] The first type of language model can be a language model used to perform structured extraction tasks. It can be a model fine-tuned from the target business domain corpus, or it can be a model with a business scenario adaptation layer for the target business domain added on top of a general model.
[0074] For each block, the input to the first language model can be a multimodal joint input prompt that combines business constraint instructions, the text content of the block, and the location information of the block to which the text content belongs in the target image. Business constraint instructions can be used to instruct the first language model to output initial structured data according to rules that include at least preset business fields, preset output data formats, and value constraints on preset fact-type parameter fields. Business constraint instructions can include at least preset business fields, preset output data formats, and value constraints on preset fact-type parameter fields.
[0075] The first major language model utilizes the context of multimodal joint input prompts, combining the text content of the block, the position coordinates of the text content in the target image, and the position information of the block to which the text content belongs for comprehensive judgment. This maps the text content of the block to corresponding preset fields, thus distinguishing the business fields to which each piece of text content belongs. For example, in a recruitment scenario, text content such as "monthly salary xxx-xxx yuan / month" can be categorized into the preset field "salary range," and "xx-xx years old" can be categorized into the preset field "age requirement." Simultaneously, for complex information containing multiple semantic elements in the text content, semantic decomposition is performed, breaking it down into independent sub-fields, and mapping the corresponding information to sub-fields composed of preset fields. For example, the preset field corresponding to the complex information "base salary xx + performance xx + full attendance xx" in the text content is "salary composition," which is broken down into independent sub-fields such as "basic salary," "performance salary," and "full attendance bonus," and the corresponding values are mapped to the predefined sub-fields of the preset field "salary composition."
[0076] After completing the semantic reasoning described above, the first major language model will organize the categorized and decomposed information into a standardized output according to the business constraints in the input prompts. This output is the initial structured data corresponding to that block. For example, taking a recruitment poster as an example, the initial structured data output would look like this: { Job Title: General Workshop Worker Salary range: "xxx-xxx yuan / month" "Salary Structure": Base salary: xxx Performance-based salary: xxx Full attendance bonus: xxx }” Age requirement: "xx-xx years old" Work Location: "XX Industrial Park, XX District, XX City" "Shift": "Day shift" Job Responsibilities: "Product assembly and material handling on the assembly line". "Benefits": "Free room and board, monthly provision of daily necessities" "Employment period": "Long-term employees" } It can be seen that by dividing the target image into multiple independent blocks and establishing the attribution relationship between text content and blocks, and by jointly inputting the text content of each block and the position information of the block to which the text content belongs in the image into the large language model, the large language model can use the layout space to assist semantic reasoning, accurately classify the text content into preset fields, and independently generate initial structured data at the block level. This solves the problems of text attribution confusion, field classification errors, and unclear semantic boundaries that exist when extracting structured information from complex layout images. It achieves high-precision and automated structured information extraction from images with multiple regions and multiple information types, and improves the accuracy and structure of information extraction.
[0077] In one embodiment, the target image is parsed to obtain the position information of multiple blocks in the target image, including: parsing the target image using a page segmentation model to obtain multiple blocks included in the target image, the position information of the multiple blocks in the target image, and the type information of the multiple blocks; determining one or more blocks associated with the target service based on the type information of the multiple blocks; and cropping the target image based on the position information of the one or more blocks associated with the target service in the target image to obtain the tile corresponding to each block associated with the target service.
[0078] In one embodiment, text recognition is performed on each block to obtain the text content of each block, including: optical character recognition is performed on the image blocks corresponding to each target service-related block to obtain the text content of each target service-related block.
[0079] In this application, the layout segmentation model can be a lightweight YOLO layout segmentation model, or other detection or segmentation models suitable for outputting block coordinates and block types. The block type information consists of semantic category labels learned during training, used to label the layout attributes of each block. For example, in a recruitment scenario, after parsing the layout of a recruitment poster, the type information of each detected block can include, but is not limited to, job information area, job promotion area, and explanatory notes area.
[0080] The target business can be a business area pre-specified by the designer or configurer based on the actual application scenario. For example, on a recruitment information publishing platform, the target business can be pre-specified as recruitment business, labor recruitment business, etc.
[0081] It can traverse multiple detected blocks and read the type label of each block, and match the type label of each block with a pre-set type configuration table. If the type label of the block exists in the configuration table, it is determined that the block is associated with the target service, the block is retained, and the location information of the block in the target image is recorded. If the type label is not in the configuration table, it is determined that the block is unrelated to the target service, the block is not retained, thus obtaining more than one block associated with the target service.
[0082] Each block associated with the target service can be traversed, and based on the coordinate range defined by the block's location information in the target image, a sub-region of the image within that coordinate range can be cropped from the target image. The main purpose of this cropping operation is to retain the pixel content within the boundary of the block associated with the target service, forming a new independent image unit (i.e., a tile corresponding to the block associated with the target service), while the pixel content outside the boundary of the associated block is not included in the tile. For the cropping operation, rectangular cropping, polygonal cropping, or cropping with edge extension can be used to avoid truncating text edges.
[0083] For each block associated with the target business, a corresponding tile is generated. Each tile contains only the pixel content of its corresponding target business-associated block within the target image. For example, taking a recruitment poster as an example, when there are two target business-associated blocks (such as two side-by-side job information areas), after cropping, two independent tiles will be obtained. Areas not associated with the target business, such as the job promotion area at the top and the explanatory notes area at the bottom of the target image, will not be included in any tile. For each tile, optical character recognition (OCR) is performed to identify the text content of the tile, i.e., the text content of the block associated with the target business.
[0084] As can be seen, after parsing the target image and outputting block type information through the page segmentation model, blocks related to the target business are filtered out before text recognition, and optical character recognition is performed only on the blocks corresponding to those blocks. On the one hand, this limits the recognition range and reduces the computational overhead of invalid areas; on the other hand, it filters out irrelevant text that is not related to the target business from the source, improving the purity and signal-to-noise ratio of the text set input to the subsequent structured extraction stage.
[0085] To further suppress text content unrelated to the target business, in one embodiment, text recognition is performed on each block to obtain the text content of each block, including: performing text recognition on each block to obtain multiple candidate texts for each block and font size weight labels for each candidate text; and filtering out candidate texts that exceed the preset font size weight range from the multiple candidate texts based on the font size weight labels to obtain the text content of each block.
[0086] In this application, candidate text can be a set of texts composed of all the text recognition results contained within a single block after performing text recognition on that block. It covers the character sequences corresponding to each detected text line or text region within the block. For example, taking the job information area of a recruitment poster as an example, after performing text recognition on that block, all the text regions contained within the block can be obtained, outputting the following candidate text set: {“High-paying urgent recruitment,” “Large-scale recruitment,” “Salary range xxx-xxx yuan / month,” “Age requirement xx-xx years old,” “Work location: Area A, Industrial Park, a certain city,” “Contact person: xxx”}. This candidate text set is the complete text content obtained after text recognition of that block, covering both valid recruitment information texts such as “Salary range xxx-xxx yuan / month” and “Age requirement xx-xx years old,” as well as promotional texts such as “High-paying urgent recruitment” and “Large-scale recruitment.”
[0087] Font size weight tags can be identifiers associated with candidate text, used to characterize the visual salience and importance level of the candidate text in the target image, distinguishing between "large-font promotional slogans" and "small-font key information." Font size weight tags can be calculated and generated based on visual features such as font height, character box pixel area, line width, stroke thickness, or relative screen proportion, without specific limitations here. For example, in the job information area of a recruitment poster, the large font size and large character box area of "High Salary Urgent Hiring" would assign it a low font size weight tag (e.g., a weight value of "low" or 0.2), indicating that it is promotional text with limited help in information extraction; while the smaller font size of "xx-xx years old" carries core business information, it would be assigned a higher font size weight tag (e.g., a weight value of "high" or 0.9), indicating that it is text related to the target business and helpful in information extraction.
[0088] The preset font size weight range can be adaptively adjusted based on the resolution, scaling ratio, or image templates of the actual target image or the business domain. The candidate text set of the current block is traversed, and the font size weight label of each candidate text is compared with the preset font size weight range. If the font size weight label of a candidate text is within the preset range, the candidate text is determined to be valid text and retained; if the font size weight label of a candidate text exceeds the preset range, the candidate text is determined to be interference text and filtered out. The retained valid text constitutes the final output text content of the block.
[0089] It can be seen that by assigning font size weight labels to the output candidate texts to represent their visual salience and importance level, and filtering out candidate texts based on the preset font size weight range, it is possible to automatically distinguish and eliminate interfering texts that are too large, visually impactful but have low information value during the text recognition process, while retaining effective texts that are small in size but carry key information. Thus, text filtering is completed before the text content enters the subsequent processing stage, which effectively improves the purity and information density of the output block text content and reduces the interference of irrelevant texts on information extraction.
[0090] To further improve the determinism of text field recognition, in one embodiment, text field recognition is performed on the initial structured data to obtain locked text fields and rewritten text fields. This includes: querying a preset first mapping relationship configuration table based on the keys of each business field in the initial structured data to determine the field attributes corresponding to each business field. The first mapping relationship configuration table records the correspondence between the keys of the business fields and the field attributes, and the field attributes include locked field attributes or rewritten field attributes; and determining the locked text fields and rewritten text fields based on the field attributes corresponding to each business field.
[0091] In this application, the first mapping configuration table can be a pre-built data structure used to record the correspondence between the keys and field attributes of each business field in the initial structured data. The key corresponds to the business field name in the structured data (e.g., job title, salary range, etc.); the field attribute indicates which processing category the business field should be assigned to in subsequent processing, that is, whether to lock the field attribute or modify the field attribute; locking the field attribute means that the business field carries factual parameters and cannot be modified, while modifying the field attribute means that the field carries descriptive text and can be modified.
[0092] Once the initial structured data is generated, a traversal operation can be performed on each business field in the initial structured data to extract the key (i.e., business field name) of each business field in the initial structured data one by one. Then, using the key as the query condition, the field attribute corresponding to the business field name can be retrieved from the first mapping relationship configuration table.
[0093] Based on the queried field attributes, the currently traversed business field is classified into the corresponding category. If the queried field attribute is a locked field attribute, the business field (including its key and the corresponding business field value) is determined as a locked text field; if the queried field attribute is a rewrite field attribute, the business field (including its key and the corresponding business field value) is determined as a rewrite text field.
[0094] As can be seen, by pre-setting a first mapping relationship configuration table that records the correspondence between the keys and attributes of business fields, when performing text field recognition on the initial structured data, a deterministic table lookup operation is performed using the keys of each business field as an index. This directly determines whether the business field belongs to a locked field attribute or a rewritten field attribute, thus transforming the field classification process from semantic judgment to deterministic mapping matching. This eliminates attribute misjudgments caused by changes in text expression, linguistic ambiguity, or contextual differences. While improving recognition efficiency, it ensures the consistency and traceability of the results of locking and rewritten text field classification in each processing, and provides a clear, stable, and traceable field processing boundary for subsequent text rewriting stages.
[0095] To further improve the automation of rewriting text to meet differentiation requirements, in one embodiment, the rewriting text field in the initial structured data is rewritten to obtain rewritten descriptive text that conforms to expression specifications and / or differentiation requirements. This includes: rewriting the rewriting text field using a second language model based on expression specifications and synonym conversion rules; calculating the similarity between the rewritten descriptive text and the existing descriptive text; and rewriting the rewritten descriptive text with a similarity greater than a similarity threshold until all rewritten descriptive texts meet the differentiation requirements.
[0096] In this application, the second major language model can be a language model used to perform text rewriting tasks on rewritten text fields in the initial structured data. Physically, it and the aforementioned first major language model can be different invocation channels of the same model instance, or they can be two completely independently trained or fine-tuned model instances.
[0097] After receiving the descriptive text to be rewritten, the second language model optimizes the text at the sentence level according to preset expression norms and synonym conversion rules. At the same time, it ensures that the locked text fields in the initial structured data are not modified during the rewriting process by obtaining the locking control information.
[0098] Synonym substitution rules are a set of rules used during text rewriting to guide the second language model in changing wording and restructuring sentences in descriptive text while preserving the business meaning. These rules can include, but are not limited to, word-level synonym substitution rules, sentence restructuring rules, narrative perspective shifting rules, and language style standardization rules. Synonym substitution rules can be automatically updated based on a pre-built system rule library, custom rules added by business operations personnel according to their job roles, and feature feedback from existing platform text.
[0099] The rewritten text field content can be extracted from the initial structured data, and the rewritten text field content, expression norms, and synonym conversion rules are assembled into input prompts for the second language model. Based on the input prompts, the second language model performs word-level replacement and sentence reconstruction on the text according to the synonym conversion rules, standardizes and corrects the language style according to the expression norms, and outputs the rewritten descriptive text.
[0100] The rewritten descriptive text is compared with existing descriptive text of the same type. For example, in a recruitment scenario, for a text field where the rewritten text field is "job responsibilities," similarity calculations can be performed with existing descriptive text of job responsibilities. If the similarity exceeds a similarity threshold, the second language model further applies expression norms and synonym conversion rules within the same semantic framework for a second round of rewriting until the similarity is less than or equal to the similarity threshold, ensuring that the rewritten descriptive text meets the differentiation requirements. Alternatively, a predetermined number of iterations can be scheduled. Once the predetermined number of iterations is reached, the rewritten descriptive text meets the differentiation requirements. The maximum number of iterations can be preset in the system configuration.
[0101] It can be seen that the technical solution of achieving differentiation requirements for rewritten descriptive text by performing text rewriting based on expression norms and synonym conversion rules using the second largest language model, and calculating similarity and threshold determination between the rewritten descriptive text and similar existing descriptive text, enables iterative rewriting to be automatically triggered when the rewritten text field fails to meet the differentiation requirements in a single rewriting, until the rewritten descriptive text reaches the expected similarity condition with similar existing descriptive text without human intervention. Thus, without human intervention, it ensures the standardization of descriptive text in terms of language style and terminology usage, and ensures that the degree of differentiation of the rewriting results is controllable, stable and predictable. It achieves an automated balance between text rewriting quality and content differentiation, while suppressing the homogenization of information content and improving content diversity.
[0102] To further reduce the risk of published content, in one embodiment, the method further includes: performing compliance semantic detection on the target structured data; if the detection result indicates that there is a compliance risk in the target structured data, marking the text fields in the target structured data that have compliance risks; and sending the risk-marked target structured data to the user terminal.
[0103] In this application, compliance semantic detection can be a process of analyzing the semantic content carried by each text field in the target structured data to determine whether the target structured data contains expressions that violate laws, regulations, industry standards, or platform publishing rules. Its purpose is to automatically identify and mark potentially non-compliant fields in the target structured data before the information is officially released, so that subsequent steps can guide users to correct them through explicit prompts.
[0104] Risk markers can include a risk key (risk_key) and a risk description (risk_desc). The risk key can be a specific text field indicating a compliance risk, specifying the data location of the risk; the risk description can be a description explaining the specific type or cause of the risk. For example, if the text contains content indicating that a company only hires men, the risk key could point to the corresponding field, and the risk description could be a prompt such as "Gender restriction statement exists, please modify." Similarly, if the text contains clearly false high-salary advertisements, the risk key could point to the corresponding field, and the risk description could be a prompt such as "Salary description is abnormal, please verify its authenticity."
[0105] The risk-marked target structured data can be sent from the server to the user terminal via WebSocket push, HTTP response return, or message queue forwarding.
[0106] As can be seen, by automatically performing compliance semantic detection on the generated target structured data, and directly adding targeted risk marker fields to the target structured data when compliance risks are detected, potential non-compliant content is transformed into a parsable risk identifier before entering the subsequent display or publication process. This eliminates the need for downstream systems or users to perform semantic interpretation of the original text line by line again. They can directly perform differentiated rendering, interception, or prompts on the corresponding fields based on the risk markers, reducing the probability of non-compliant content flowing into the online platform and improving the quality of information publication.
[0107] Figure 3 This is a flowchart of a data processing method executed by a client according to an embodiment of this application. This method can be performed by… Figure 1 The client in the system shown executes this. For example... Figure 3 As shown, the method may include the following steps: Step 201: Step 301: Obtain the target image specified by the user and provide the target image to the server.
[0108] Step 303: Obtain target structured data based on the target image from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold.
[0109] Step 305: Display the information publishing page and populate the target structured data into the multiple fields contained in the information publishing page.
[0110] Step 307: In response to the detection of an operation that triggers information publishing, publish the information based on the information publishing page.
[0111] As can be seen from the above process, by centrally processing the target image containing the information to be published through the server, including information extraction, factual parameter information locking, and description text standardization and / or differentiation, the processing results are returned to the client in the form of target structured data and automatically filled into the corresponding fields on the information publishing page. This achieves the automatic conversion from unstructured target images to editable and publishable target structured data on the information publishing page. This integrates the traditional multi-step operations of image viewing, information identification, manual field-by-field entry, and text modification into a continuous automatic link of one trigger and one backfill, simplifying the user's operation process from image to information publishing. At the same time, it avoids information distortion and inconsistent formatting caused by visual omissions, input errors, or arbitrary expression in the manual process, thereby improving the quality and diversity of information publishing.
[0112] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments.
[0113] First, the above step 301, namely "acquiring the target image determined by the user and providing the target image to the server", will be described in detail with reference to the embodiments.
[0114] In this application, users can click the "Upload Image" button on the job posting page of the client and select a target image from their local photo album; they can also directly capture a target image by calling the camera; or they can select a target image from chat history, email attachments, or in-app screenshot cache.
[0115] After obtaining the target image, the client performs image compression, format unification, and other processing, and sends it to the server as a binary image stream via HTTP request or file upload interface, so that the server can perform subsequent information extraction, text rewriting, and structured processing based on the target image.
[0116] The following describes step 303, namely "obtaining target structured data based on the target image from the server", in detail with reference to the embodiments.
[0117] In this application, the target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold.
[0118] The target structured data is the final data object returned by the server after performing all processing on the target image. It contains at least two types of information: one is factual parameter information extracted from the target image, i.e., content with objective constraint attributes; the other is descriptive text after text rewriting, which conforms to expression norms and / or meets differentiation requirements. The specific implementation process of the target structured data obtained by the server performing all processing on the target image can refer to the processing process of steps 201 to 209 mentioned above, and will not be repeated here.
[0119] The client and server can agree to communicate using the HTTP protocol. The server will process and generate the target structured data, write it into the HTTP response body in JSON format, and return it to the client. The client will parse the JSON data from the response body to obtain the target structured data, thus completing a complete data request and response interaction.
[0120] The following describes in detail step 305, namely "displaying the information publishing page and filling the target structured data into the multiple fields contained in the information publishing page", with reference to the embodiments.
[0121] In this application, the information publishing page can be an interactive interface on the client side used to ultimately host and submit the information to be published. After confirming that the content of each field on the information publishing page is correct, the user can trigger the final information publishing operation through the publish button on the information publishing page.
[0122] After receiving the target structured data, the client can automatically populate the corresponding target fields according to the field mapping relationship between each business field in the target structured data and multiple target fields on the information publishing page. At this point, the user can directly see the automatically populated draft of the published information on the information publishing page.
[0123] The following describes step 307, namely "in response to detecting an operation that triggers information publishing, publishing information based on the information publishing page," in detail with reference to an embodiment.
[0124] Users can check the draft of the automatically populated information. If the information is correct, users can click the "Publish," "Submit," or "Confirm Publish" button on the information publishing page to trigger the information publishing operation. After detecting the operation, the client collects the current values of each field on the information publishing page (which may include minor adjustments made by the user after autofilling), assembles the publishing data according to the format required by the publishing interface, and submits it to the server's publishing processing interface to complete the formal publishing of the information.
[0125] In one embodiment, acquiring a target image determined by the user and providing the target image to the server includes: displaying the target image determined by the user and a first component for triggering information recognition on a job editing page; and providing the target image to the server in response to the operation of triggering the first component.
[0126] In this application, the information publishing page and the job posting page can be different functional views of the same page. During the stage where the user uploads and edits an image, this page is referred to as the "job posting page"; during the stage where data is filled in and submission is ready, this page is referred to as the "information publishing page." The difference between the two lies only in the current operational stage and functional focus, rather than being two independent and different pages.
[0127] The job posting page includes an image upload area, a first component, and at least one editable text box. The image preview area displays the target image selected by the user, the first component triggers information recognition processing for that target image, and the editable text box contains descriptive text from the target structured data returned by the server, after text rewriting.
[0128] The first component can be placed below, to the side of, or at the bottom of the job editing page. It can be a button, icon, floating control, or other interactive element that can be triggered by the user, and its displayed content can be text such as "AI Intelligent Recognition" or "One-Click Extraction" to indicate the recognition function.
[0129] In one implementation, after detecting a user's action on the first component, the target image displayed on the current job editing page is provided to the server. This provision process may include formatting the target image, compressing it, or encapsulating upload parameters, and then sending the target image to the server via a network request, enabling the server to generate target structured data based on the target image.
[0130] In another implementation, the provision process may include formatting and compressing the target image, or encapsulating upload parameters, and sending the target image to the server via a network request. The server performs layout analysis on the target image, identifies the type information of each block on the target image, and returns the identified job names from each job information area to the client. The client displays each job name on the job editing page and associates each job name with its corresponding second component. After detecting a user's triggering of an operation on the second component, the server processes the cropped blocks in the target image corresponding to the job names bound to the second component to obtain target structured data for that job name.
[0131] It should be noted that the second component can be implemented as a button, icon, floating control, or other interactive element that can be triggered by the user, and its displayed content can be text such as "AI intelligent recognition" or "one-click extraction" to prompt the recognition function.
[0132] In one embodiment, obtaining target structured data based on a target image from a server includes: obtaining target structured data based on a target image from a server and displaying descriptive text after text rewriting in an editable text box on the job editing page.
[0133] After the client obtains the target structured data from the server, it parses the data to obtain the keys and values of each business field. It then determines the business fields of the descriptive text after text rewriting and fills the corresponding business field values into the editable text boxes on the job posting page, displaying them to the user. These editable text boxes allow users to view, add, delete, or modify the automatically filled content, ensuring that the automatic recognition and text rewriting results are confirmed by the user before publication.
[0134] As can be seen, by integrating the target image display, recognition trigger entry, and text editing area into the job editing page, users can continuously complete image confirmation, recognition initiation, and text review and modification on the same page. The upload trigger of image materials and the presentation and editing of text optimization results are concentrated in a single operation interface, eliminating the need to switch back and forth between different views such as image browsing, recognition waiting, and text editing. This forms a continuous interactive closed loop from visual material input to publishable text output, reducing user operation interruption and cognitive burden.
[0135] To further illustrate the method of filling in the target structured data, in one embodiment, the target structured data is filled into multiple fields contained in the information publishing page, including: based on a preset second mapping relationship configuration table, the text fields in the target structured data are filled into multiple target fields contained in the information publishing page, and the text fields with compliance risks are displayed with a style or mark that is different from other text fields to prompt the user to modify the text fields with compliance risks; wherein, the second mapping relationship configuration table includes the correspondence between text fields and target fields on the information publishing page.
[0136] In this application, text fields in the target structured data that pose compliance risks are marked with risk tags. After generating the target structured data, the server also performs compliance semantic checks on the target structured data. When a text field in the target structured data that poses a compliance risk is detected, the server adds a risk tag to the corresponding text field in the target structured data. The risk tag may include a risk key and a risk description; the risk key points to the specific text field that poses a compliance risk, and the risk description explains the specific type or cause of the risk.
[0137] The second mapping configuration table is a pre-set table used to record the correspondence between text fields in the target structured data and target fields on the information publishing page. For example, in a recruitment scenario, the "salary range" field in the target structured data corresponds to the "salary input box" control on the information publishing page, and the "job responsibilities" field in the target structured data corresponds to the "job responsibilities multi-line text area" control on the information publishing page, etc.
[0138] During the backfilling process, the client iterates through each text field in the target structured data, reads the key of each text field, finds the corresponding target field on the information publishing page based on the second mapping relationship configuration table, and fills the field value of that text field into the corresponding input control. During the backfilling process, the client simultaneously checks whether each text field carries a risk marker. For text fields carrying risk markers, the client displays their corresponding input controls using a style or marker that distinguishes them from other text fields. For example, the border of the input control can be highlighted in red, a warning icon can be displayed next to the control, or a red warning message can be displayed above the control (such as "This field may have compliance risks; please pay attention and modify"), making risky fields visually distinct from other normal fields.
[0139] As can be seen, the text fields in the target structured data can not only be automatically backfilled into multiple target fields on the information publishing page based on the second mapping relationship configuration table, but also provide visual prompts for text fields with risk markers. This method allows the risk identification results from the server to be directly transmitted to the user's editing interface, enabling users to correct relevant content based on differentiated display results before publishing, thereby improving the reviewability and controllability of the backfilled results of the target structured data.
[0140] Figure 4 This is an interactive schematic diagram of the data processing method according to an embodiment of this application.
[0141] like Figure 4 As shown, taking job recruitment as the target business and a recruitment poster as the target image as an example, the recruitment poster contains at least one job recruitment information. Figure 4 This illustration schematically demonstrates the interactive process of generating target structured data from a recruitment poster as the target image and populating it onto an information publishing page. In this embodiment, the method can be executed collaboratively by a client and a server. The server is used to generate target structured data based on the target image, while the client is used to provide the target image, display the results, populate fields, and trigger the publishing process.
[0142] The client first displays a job posting editing page, which can include an image upload area and a first component to trigger information recognition. For example, the first component can display "AI intelligent recognition," and at this time, the first component is unavailable. Figure 4 (a)). After the user clicks on the image upload area and obtains the target image determined by the user through methods such as selecting from the local album, taking a picture, or selecting from the cache, the target image is displayed on the job editing page. At this time, the first component that triggers information recognition switches from an unusable state to an available state, so that the user can confirm the image content to be processed. Figure 4 In (b) of the diagram, the dashed line of the first component indicates an unusable state, and the implementation indicates an usable state.
[0143] In response to detecting a user's triggering action on the first component, after obtaining the job titles in the job information areas of each position in the target image, the client pops up a job title display interface on the job editing page to display each job title and the corresponding second component used to trigger job information recognition. For example, the display content of the second component can be "AI Intelligent Recognition" (…). Figure 4 (c) in the middle.
[0144] In response to a detected user interaction with the second component, the client retrieves the target structured data related to the job title corresponding to the second component from the server. It then parses the target structured data, identifies fields belonging to descriptive text, and fills the corresponding editable text boxes on the job editing page with the values of these fields. This displays the rewritten descriptive text in the editable text boxes on the job editing page. For example, if the identified descriptive text field is "job description," the rewritten descriptive text corresponding to this job description is filled into the editable text box for the job description and displayed to the user. The user can view, modify, add to, or delete the text content in the editable text box for the job description. At this time, the parameter fields corresponding to factual parameter information on the job editing page, such as "salary range" and "salary composition," can be collapsed. Figure 4 (d) In this context, when the user clicks the "Next" button on the job editing page, the client displays the information publishing page and fills the factual parameter information from the target structured data into the corresponding fields on the information publishing page. For example, the "Job 1" field is filled into the corresponding job name field, and the salary range value is filled into the salary range field on the information publishing page. This, along with the rewritten descriptive text, forms a publishing draft that the user can check and modify. When the user checks the publishing draft and finds it correct and ready to publish, the client detects that the user has triggered the information publishing operation, i.e., the user clicks the "Publish" button to publish the information. Figure 4 (e) in the middle.
[0145] When the information is successfully published, a publishing results interface pops up on the information publishing page. This interface displays "Published successfully" and also asks whether to continue publishing job postings. "Yes" and "No" controls are available on this interface. Figure 4 (f) If the user clicks "Yes", the page will redirect to... Figure 4 In step (c), the second component corresponding to the job that has already been published is unavailable. The user clicks on the second component corresponding to the job to be published to continue the subsequent interaction process; if the user clicks "no", the process ends.
[0146] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0147] According to another embodiment, a data processing apparatus is provided. Figure 5This is a schematic block diagram of a server-side data processing apparatus provided in an embodiment of this application. Figure 5 As shown, the device 500 includes: an image acquisition unit 510, an information extraction unit 520, a field recognition unit 530, a text rewriting unit 540, and a data acquisition unit 550. The main functions of each component are as follows: The image acquisition unit 510 is configured to acquire a target image.
[0148] The information extraction unit 520 is configured to extract information from the target image and generate initial structured data.
[0149] The field recognition unit 530 is configured to perform text field recognition on the initial structured data to obtain locked text fields and rewritten text fields. The locked text fields carry factual parameter information, and the rewritten text fields carry descriptive text.
[0150] The text rewriting unit 540 is configured to rewrite the rewriting text field in the initial structured data to obtain rewritten descriptive text that meets the expression specifications and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than a similarity threshold.
[0151] Data acquisition unit 550 is configured to obtain target structured data using locked text fields and rewritten text.
[0152] Furthermore, the field identification unit 530 can be configured to: query a preset first mapping relationship configuration table based on the keys of each business field in the initial structured data to determine the field attributes corresponding to each business field. The first mapping relationship configuration table records the correspondence between the keys of the business fields and the field attributes, and the field attributes include locked field attributes or rewritten field attributes; and determine locked text fields and rewritten text fields based on the field attributes corresponding to each business field.
[0153] Furthermore, the text rewriting unit 540 can be configured to rewrite the rewritten text field based on the expression norms and synonym conversion rules using the second language model; calculate the similarity between the rewritten descriptive text and the existing descriptive text; and rewrite the rewritten descriptive text with a similarity greater than the similarity threshold until all rewritten descriptive texts meet the differentiation requirements.
[0154] As one implementation method, the information extraction unit 520 can be configured to perform layout parsing on the target image to obtain the position information of multiple blocks in the target image; perform text recognition on each block to obtain the text content of each block; use the first language model to extract information from preset fields for each block based on the text content of each block and the position information of the block to which the text content belongs in the target image to obtain the structured data of each block; and use the structured data of each block to obtain the initial structured data.
[0155] Furthermore, the information extraction unit 520 can be configured to perform layout analysis on the target image using a layout segmentation model to obtain multiple blocks included in the target image, the position information of the multiple blocks in the target image, and the type information of the multiple blocks; based on the type information of the multiple blocks, determine one or more blocks associated with the target business; based on the position information of the one or more blocks associated with the target business in the target image, crop the target image to obtain the tile corresponding to each block associated with the target business.
[0156] Furthermore, the information extraction unit 520 can be configured to perform text recognition on each block to obtain the text content of each block, including: performing optical character recognition on the blocks corresponding to each target service-related block to obtain the text content of each target service-related block.
[0157] Furthermore, the information extraction unit 520 can be configured to perform text recognition for each block separately, obtain multiple candidate texts for each block and font size weight labels for each candidate text; based on the font size weight labels, filter out candidate texts that exceed the preset font size weight range from the multiple candidate texts to obtain the text content of each block.
[0158] In one implementation, the device 500 includes: a semantic detection unit, a risk marking unit, and a data transmission unit.
[0159] The semantic detection unit is configured to perform compliance semantic detection on the target structured data.
[0160] The risk labeling unit is configured to label text fields in the target structured data that have compliance risks when the detection results indicate that there are compliance risks in the target structured data.
[0161] The data sending unit is configured to send the risk-marked target structured data to the user terminal.
[0162] According to another embodiment, a data processing apparatus is provided. Figure 6 A schematic block diagram of a client-side data processing apparatus provided in another embodiment of this application. (See diagram below.) Figure 6As shown, the device 600 includes: an image providing unit 610, a data acquisition unit 620, a page display unit 630, and an information publishing unit 640. The main functions of each component are as follows: The image providing unit 610 is configured to acquire a target image determined by the user and provide the target image to the server.
[0163] The data acquisition unit 620 is configured to acquire target structured data based on the target image from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold.
[0164] Page display unit 630 is configured to display the information publishing page and populate the target structured data into the multiple fields contained in the information publishing page.
[0165] The information publishing unit 640 is configured to publish information based on the information publishing page in response to detecting an operation that triggers information publishing.
[0166] In one implementation, the image providing unit 610 can be configured to display a target image determined by the user and a first component for triggering information recognition on the job editing page; in response to the operation of triggering the first component, provide the target image to the server; obtain target structured data based on the target image from the server, and display the rewritten descriptive text in the editable text box of the job editing page.
[0167] As one implementation method, text fields in the target structured data that pose compliance risks are marked with risk indicators. Further, the page display unit 630 can be configured to, based on a preset second mapping relationship configuration table, populate the text fields in the target structured data into multiple target fields included in the information publishing page, and display the text fields with compliance risks using a style or marker different from other text fields to prompt the user to modify the text fields with compliance risks; wherein, the second mapping relationship configuration table includes the correspondence between text fields and target fields on the information publishing page.
[0168] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.
[0170] In addition, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the methods in the foregoing method embodiments.
[0171] And an electronic device, comprising: One or more processors; and A memory associated with one or more processors, the memory being used to store program instructions that, when read and executed by one or more processors, perform the steps of any of the methods in the foregoing method embodiments.
[0172] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods in the foregoing method embodiments.
[0173] in, Figure 7This is a schematic block diagram of an electronic device provided in an embodiment of this application. It exemplarily illustrates the architecture of the electronic device, which may specifically include a processor 710, a video display adapter 711, a disk drive 712, an input / output interface 713, a network interface 714, and a memory 720. The processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, and memory 720 can be communicatively connected via a communication bus 730.
[0174] The processor 710 can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs and implement the technical solution provided in this application.
[0175] The memory 720 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 720 can store the operating system 721 for controlling the operation of the electronic device 700, and the basic input / output system (BIOS) 722 for controlling the low-level operations of the electronic device 700. Additionally, it can store a web browser 723, a data storage management system 724, and data processing devices 500 / 600, etc. The aforementioned data processing devices 500 / 600 can be the application programs that specifically implement the aforementioned steps in the embodiments of this application. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 720 and is called and executed by the processor 710.
[0176] Input / output interface 713 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.
[0177] Network interface 714 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0178] Bus 730 includes a pathway for transmitting information between various components of the device, such as processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, and memory 720.
[0179] It should be noted that although the above-described device only shows the processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, memory 720, bus 730, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.
[0180] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer program product. This computer program product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of the embodiments of this application.
[0181] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A data processing method, characterized in that, The method includes: Acquire the target image; Information is extracted from the target image to generate initial structured data; Text field recognition is performed on the initial structured data to obtain locked text fields and rewritten text fields. The locked text fields carry factual parameter information, and the rewritten text fields carry descriptive text. The rewritten text field in the initial structured data is rewritten to obtain rewritten descriptive text that conforms to expression standards and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than a similarity threshold. The target structured data is obtained by using the locked text field and the rewritten text.
2. The method according to claim 1, characterized in that, The step of performing text field recognition on the initial structured data to obtain locked text fields and rewritten text fields includes: Based on the keys of each business field in the initial structured data, a preset first mapping relationship configuration table is queried to determine the field attributes corresponding to each business field. The first mapping relationship configuration table records the correspondence between the keys of the business fields and the field attributes. The field attributes include locking field attributes or rewriting field attributes. Based on the field attributes corresponding to each business field, the locked text field and the rewritten text field are determined.
3. The method according to claim 1, characterized in that, The step of rewriting the rewritten text field in the initial structured data to obtain rewritten descriptive text that conforms to expression standards and / or differentiation requirements includes: The rewritten text field is rewritten using the second largest language model based on expression norms and synonym conversion rules; The similarity between the rewritten descriptive text and the existing descriptive text is calculated. The rewritten description text with a similarity greater than the similarity threshold is rewritten again until all rewritten description texts meet the differentiation requirements.
4. The method according to claim 1, characterized in that, The step of extracting information from the target image to generate initial structured data includes: The target image is parsed to obtain the position information of multiple blocks in the target image. Text recognition is performed on each of the aforementioned blocks to obtain the text content of each block; Using the first major language model, based on the text content of each block and the location information of the block to which the text content belongs in the target image, information is extracted from preset fields for each block to obtain structured data for each block; the initial structured data is obtained using the structured data of each block.
5. The method according to claim 4, characterized in that, The step of parsing the target image to obtain the position information of multiple blocks within the target image includes: The target image is parsed using a page segmentation model to obtain multiple blocks included in the target image, the position information of the multiple blocks in the target image, and the type information of the multiple blocks; Based on the type information of the multiple blocks, determine one or more blocks associated with the target business; Based on the location information of the one or more blocks associated with the target service in the target image, the target image is cropped to obtain the image blocks corresponding to each block associated with the target service. The step of performing text recognition on each of the blocks to obtain the text content of each block includes: performing optical character recognition on the image blocks corresponding to each block associated with the target service to obtain the text content of each block associated with the target service.
6. The method according to claim 4, characterized in that, The step of performing text recognition on each of the aforementioned blocks to obtain the text content of each block includes: Text recognition is performed on each of the blocks to obtain multiple candidate texts for each block and font size weight labels for each candidate text; Based on the font size weight tags, candidate texts that exceed the preset font size weight range are filtered out from the multiple candidate texts to obtain the text content of each block.
7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Perform compliance semantic detection on the target structured data; If the detection results indicate that there is a compliance risk in the target structured data, risk marking is performed on the text fields in the target structured data that have compliance risks; The target structured data after risk labeling is sent to the user terminal.
8. A data processing method, characterized in that, The method includes: Obtain the target image specified by the user and provide the target image to the server; The target structured data obtained from the target image is obtained from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to the expression standard and / or meets the differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than the similarity threshold. Display the information publishing page and populate the target structured data into the multiple fields contained in the information publishing page accordingly; In response to the detection of an operation that triggers information publishing, information is published based on the information publishing page.
9. The method according to claim 8, characterized in that, The step of obtaining the target image determined by the user and providing the target image to the server includes: displaying the target image determined by the user and a first component for triggering information recognition on the job editing page; and providing the target image to the server in response to the operation of triggering the first component. The step of obtaining the target structured data based on the target image from the server includes: obtaining the target structured data based on the target image from the server, and displaying the rewritten descriptive text in the editable text box on the job editing page.
10. The method according to claim 8, characterized in that, The text fields in the target structured data that pose compliance risks are marked with risk indicators; The step of filling the target structured data into the multiple fields included in the information publishing page includes: based on a preset second mapping relationship configuration table, filling the text fields in the target structured data into the multiple target fields included in the information publishing page, and displaying the text fields with compliance risks using a style or mark different from other text fields to prompt users to modify the text fields with compliance risks; wherein, the second mapping relationship configuration table includes the correspondence between text fields and target fields on the information publishing page.
11. A data processing apparatus, characterized in that, The device includes: The image acquisition unit is configured to acquire the target image; An information extraction unit is configured to extract information from the target image and generate initial structured data; The field recognition unit is configured to perform text field recognition on the initial structured data to obtain a locked text field and a rewritten text field, wherein the locked text field carries factual parameter information and the rewritten text field carries descriptive text; The text rewriting unit is configured to rewrite the rewrite text field in the initial structured data to obtain rewritten descriptive text that conforms to the expression specifications and / or differentiation requirements. The differentiation requirements include that the similarity between the rewritten descriptive text and the existing descriptive text is greater than a similarity threshold. The data acquisition unit is configured to obtain target structured data using the locked text field and the rewritten text.
12. A data processing apparatus, characterized in that, The device includes: An image providing unit is configured to acquire a target image determined by the user and provide the target image to the server. The data acquisition unit is configured to acquire target structured data based on the target image from the server. The target structured data includes factual parameter information extracted from the target image and descriptive text after text rewriting. The descriptive text after text rewriting conforms to expression norms and / or meets differentiation requirements. The differentiation requirements include that the similarity between the descriptive text after text rewriting and the existing descriptive text is greater than a similarity threshold. The page display unit is configured to display an information publishing page and fill the target structured data into multiple fields contained in the information publishing page. The information publishing unit is configured to publish information based on the information publishing page in response to detecting an operation that triggers information publishing.
13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1 to 10.
14. An electronic device, characterized in that, include: One or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method according to any one of claims 1 to 10.
15. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 10.
