An education contract text performance verification method and device and electronic equipment

Abstract

The application discloses an education contract text performance verification method and device and electronic equipment, and relates to the technical field of education, which comprises the following steps: obtaining an education contract text to be verified and a multi-dimensional performance file corresponding to the education contract text to be verified in an education service project; uniformly representing at least one multimedia form content in the multi-dimensional performance file to obtain at least one performance content segment corresponding to the multi-dimensional performance file; selecting a target performance content segment meeting a verification condition from the at least one performance content segment according to index information in the education contract text to be verified, and associating the index information with the target performance content segment to obtain a performance content framework corresponding to the education contract text to be verified; and verifying the execution of the education contract text to be verified based on the performance content framework to obtain performance state information of the education contract text to be verified. The application reduces the subjective intervention of manual selection of verification content and improves the accuracy of the verification result.

Description

Technical Field

[0001] This application relates to the field of educational technology, and in particular to a method, apparatus, and electronic device for verifying the performance of educational contract texts. Background Technology

[0002] Educational service projects cover a variety of types, including operation and management, and teacher training. Contract performance verification is an important management link to ensure the standardized progress of the project and protect the rights and interests of both parties.

[0003] Currently, the verification of the performance of existing education service contracts mainly relies on manual operation. Staff need to first interpret the non-standardized clauses in the contract, manually create performance indicators in general management software, extract the text content of the performance documents through basic optical character recognition (OCR), and manually compare and verify the extraction results with the manually created performance indicators.

[0004] However, while existing technologies exist for intelligent contract processing, they primarily involve simple generation, editing, and processing of contract text. These technologies struggle to understand the specific content relevant to the education sector, leading to misunderstandings of technical terms and consequently lower accuracy in content recognition for education-related contracts. Furthermore, current contract performance verification methods cannot fully recognize all multimedia content, including text, images, and charts, resulting in incomplete or missing key information. This leads to high levels of content deficiencies and incompleteness, further complicating the accuracy of content recognition for education-related contracts. Finally, manual verification of the identified content can result in significant discrepancies between different reviewers, indicating subjective human influence and ultimately reducing the accuracy of the verification results. Summary of the Invention

[0005] In view of this, this application provides a method, apparatus, and electronic device for verifying the performance of educational contract texts. The main purpose is to improve the technical problems of existing technologies, which have poor understanding of the specific content involved in the education field during the content recognition process, low recognition completeness of different multimedia content such as text, images, and charts in the document content, and the use of manual verification in the process of verifying the identified content, which can lead to the verification results being affected by subjective human factors, resulting in poor accuracy of the verification results.

[0006] Firstly, this application provides a method for verifying the performance of an education contract, including:

[0007] Obtain the text of the education contract to be verified in the education service project and the multi-dimensional performance documents corresponding to the text of the education contract to be verified;

[0008] The content of at least one multimedia form in the multi-dimensional performance document is uniformly represented to obtain at least one performance content fragment corresponding to the multi-dimensional performance document. The at least one multimedia form includes at least one of text content, image content and chart content.

[0009] Based on the indicator information in the education contract text to be verified, a target performance content segment that meets the verification conditions is selected from the at least one performance content segment, and the indicator information is associated with the target performance content segment to obtain the performance content framework corresponding to the education contract text to be verified.

[0010] The execution status of the education contract text to be verified is verified based on the aforementioned performance content framework to obtain the performance status information corresponding to the education contract text to be verified.

[0011] Optionally, at least one multimedia content in the multi-dimensional performance document is uniformly represented to obtain at least one performance content fragment corresponding to the multi-dimensional performance document, including:

[0012] The text content of the multi-dimensional performance document is recognized to obtain at least one text block, and the first performance content segment corresponding to the multi-dimensional performance document is determined based on the at least one text block;

[0013] Image content recognition is performed on the multi-dimensional performance document to obtain at least one color-marked region, and a second performance content segment corresponding to the multi-dimensional performance document is determined based on the at least one color-marked region;

[0014] The table structure of the multi-dimensional performance document is identified to obtain at least one table data, and the table data is structurally transformed to obtain the third performance content fragment corresponding to the multi-dimensional performance document.

[0015] Optionally, the step of performing image content recognition on the multi-dimensional performance document to obtain at least one color-marked region, and determining the second performance content segment corresponding to the multi-dimensional performance document based on the at least one color-marked region, includes:

[0016] Identify the image content in the multi-dimensional performance document, perform pixel analysis on the image content, and segment the image content into at least one color-marked region;

[0017] Extract the visual features corresponding to the at least one color-marked region;

[0018] Determine the target text content corresponding to the visual features from the multi-dimensional performance documents, and extract text semantic features from the target text content;

[0019] The text semantic features and the visual features are concatenated, and the execution status information corresponding to the image content is identified based on the concatenated features. The second performance content fragment is then determined from the image content based on the execution status information.

[0020] Optionally, the step of performing table structure recognition on the multi-dimensional performance document to obtain at least one table data, and performing structure transformation on the table data to obtain a third performance content fragment corresponding to the multi-dimensional performance document, includes:

[0021] Identify the table content in the multi-dimensional performance document, and determine at least one cell in the table content and the cell position corresponding to the at least one cell;

[0022] Based on the cell position, at least one cell is used as a graph node, and connecting edges are established between the at least one cell to generate a table structure graph corresponding to the at least one cell;

[0023] Based on the table structure diagram, the row and column relationships between the at least one cell are analyzed, and the structure of the at least one cell is transformed based on the row and column relationships to obtain the third performance content fragment.

[0024] Optionally, the step of selecting target performance content segments that meet the verification conditions from the at least one performance content segment based on the indicator information in the education contract text to be verified, and associating the indicator information with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified, includes:

[0025] The target language model is used to perform deep semantic analysis on the educational contract text to be verified to obtain the indicator information in the educational contract text. The target language model is obtained by adaptive training on educational corpus.

[0026] The indicator information and the at least one performance content fragment are encoded in the same semantic vector space to obtain the indicator vector corresponding to the indicator information and the at least one content vector corresponding to the at least one performance content fragment.

[0027] Determine the semantic similarity data between the indicator vector and the at least one content vector;

[0028] Based on the semantic similarity data, segments with a similarity greater than a similarity threshold are selected from the at least one performance content segment, and the selected segments are used as candidate performance content segments.

[0029] Information alignment and segment extraction operations are performed on the candidate performance content segments to select target performance content segments that meet the verification conditions from the candidate performance content segments;

[0030] By associating the indicator information with the target performance content fragment, a performance content framework corresponding to the education contract text to be verified is obtained.

[0031] Optionally, the step of performing information alignment and segment extraction operations on the candidate performance content segments to select target performance content segments that meet the verification conditions from the candidate performance content segments includes:

[0032] The verification information corresponding to the indicator vector is extracted from the candidate performance content fragment, and the verification information is associated with the indicator vector to obtain at least one set of verification association information;

[0033] Verification element information is extracted from the at least one set of verification association information, and verification parameter information is extracted from the multi-dimensional performance documents. Based on the verification element information and the verification parameter information, a target performance content segment that meets the verification conditions is selected from the candidate performance content segments.

[0034] Optionally, associating the indicator information with the target performance content fragment to verify the execution status of the education contract text to be verified, and obtaining the performance status information corresponding to the education contract text to be verified, includes:

[0035] Based on the target performance content fragment, generate at least one verification unit group corresponding to the indicator vector, and generate at least one verification feature vector corresponding to the at least one verification unit group;

[0036] The indicator vector and the target performance content fragment are used as the verification premises for the at least one verification feature vector. The at least one verification feature vector is verified to obtain the probability distribution information corresponding to the at least one verification feature vector.

[0037] Based on the probability distribution information, the performance status information corresponding to the education contract text to be verified is generated.

[0038] Secondly, this application provides a device for verifying the performance of an education contract, comprising:

[0039] The acquisition module is configured to acquire the text of the education contract to be verified and the multi-dimensional performance documents corresponding to the text of the education contract to be verified in the education service project.

[0040] The characterization module is configured to uniformly characterize at least one multimedia content in the multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document. The at least one multimedia content includes at least one of text content, image content, and chart content.

[0041] The association module is configured to select a target performance content segment that meets the verification conditions from the at least one performance content segment based on the indicator information in the education contract text to be verified, and associate the indicator information with the target performance content segment to obtain the performance content framework corresponding to the education contract text to be verified.

[0042] The verification module is configured to verify the execution status of the education contract text to be verified based on the performance content framework, and obtain the performance status information corresponding to the education contract text to be verified.

[0043] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the performance verification method for the educational contract text described in the first aspect.

[0044] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the performance verification method for the educational contract text described in the first aspect.

[0045] Fifthly, this application provides a computer program product, which includes a computer program that, when executed by a processor, implements the performance verification method for the educational contract text described in the first aspect.

[0046] Using the above technical solution, this application provides a method, apparatus, and electronic device for verifying the performance of an education contract text, comprising: acquiring an education contract text to be verified and a multi-dimensional performance document corresponding to the education contract text to be verified in an education service project; uniformly representing at least one multimedia content in the multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document; selecting a target performance content fragment that meets the verification conditions from the at least one performance content fragment based on indicator information in the education contract text to be verified, and associating the indicator information with the target performance content fragment to obtain a performance content framework corresponding to the education contract text to be verified; verifying the execution status of the education contract text to be verified based on the performance content framework to obtain performance status information corresponding to the education contract text to be verified. Compared with existing technologies, this application achieves complete collection of basic information required for verification by acquiring the educational service contract text to be verified and the corresponding multi-dimensional performance documents of the educational service project; by uniformly representing at least one multimedia content in the multi-dimensional performance documents, at least one performance content fragment corresponding to the multi-dimensional performance documents is obtained, improving the recognition and integration of different multimedia content; by identifying indicator information in the educational contract text to be verified and associating the indicator information with the target performance content fragment, the performance content framework corresponding to the educational contract text to be verified is obtained, reducing the subjective intervention of manual selection of verification content; and by verifying the execution status of the educational contract text to be verified based on the performance content framework, the performance status information corresponding to the educational contract text to be verified is obtained, improving the accuracy of the verification results. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart illustrating a method for verifying the performance of an education contract text according to an embodiment of this application is shown.

[0050] Figure 2 A flowchart illustrating a method for verifying the performance of an education contract text according to an embodiment of this application is shown.

[0051] Figure 3a and Figure 3b This illustration shows a schematic diagram of an example of intelligent verification of educational service contract performance based on multimodal semantic understanding provided in an embodiment of this application;

[0052] Figure 4 This illustration shows a structural schematic diagram of a performance verification device for an education contract text provided in an embodiment of this application;

[0053] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0054] The embodiments of this application will now be described in more detail with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0055] To address the shortcomings of existing technologies in content recognition, such as poor understanding of educational content and low accuracy in recognizing various multimedia formats like text, images, and charts, and the inherent inaccuracies caused by subjective human factors in manual verification, this embodiment provides a method for verifying the performance of educational contracts. Figure 1 As shown, the method includes:

[0056] Step 101: Obtain the text of the education contract to be verified and the corresponding multi-dimensional performance documents in the education service project.

[0057] In this embodiment of the application, the education contract text to be verified may be an education service contract, which may include various education service-related clauses such as operation and management, teacher training, etc.

[0058] In this embodiment of the application, the multi-dimensional performance documents can be documents generated during the performance of an education service project, used to prove the execution of the terms of the education contract to be verified. For example, the multi-dimensional performance documents in this embodiment of the application may include, but are not limited to, supervision reports, achievement reports, teacher development evaluation forms, etc., and the multi-dimensional performance documents may be presented in various multimedia formats such as text, images, and tables.

[0059] Step 102: Perform a unified representation of at least one multimedia content in the multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document.

[0060] Among them, at least one multimedia content includes at least one of text content, image content, and chart content.

[0061] In this application embodiment, unified representation can be the conversion of the content of different multimedia formats (text, images, tables) in multi-dimensional performance documents into a standardized information format that can be processed and associated by machines.

[0062] In this embodiment of the application, the performance content fragment may be a partial information fragment extracted from multi-dimensional performance documents that is related to the execution of the text terms of the education contract to be verified. Each performance content fragment may correspond to a multimedia form of partial content or processed structured information.

[0063] For example, if the multi-dimensional performance document is a text file, the unified representation can be converted into a vector representation or a structured text block through text encoding; if the multi-dimensional performance document is an image file, the unified representation can be a fusion vector of visual features and corresponding text semantic features; if the multi-dimensional performance document is a tabular file, the unified representation can be structured data or a data vector after structural reconstruction.

[0064] For example, if the multi-dimensional performance document is a student's academic achievement report, and the achievement report includes text, images, and tables, then the text paragraphs in the achievement report can be BERT encoded to obtain text vectors. The HSV color histogram features and the corresponding semantic features of the text can be extracted from the completed areas marked in green in the achievement report and concatenated into a fusion vector. The teacher development assessment form can be structurally reconstructed to obtain JSON format structured data. All the processed information can be used as performance content fragments.

[0065] Step 103: Identify the indicator information in the education contract text to be verified, and select the target performance content segment that meets the verification conditions from at least one performance content segment based on the indicator information.

[0066] In this embodiment of the application, the indicator information can be structured information extracted from the text of the education contract to be verified, which can be used to measure the implementation of the clauses. For example, the indicator information in this embodiment of the application may specifically include target entities, quantitative targets, time nodes, operators, etc.

[0067] In the embodiments of this application, the target performance content fragment that meets the verification conditions can be a performance content fragment that is semantically related to the target information and can provide execution data or status description. The target performance content fragment can be used for subsequent verification associated with the indicator information.

[0068] In the embodiments of this application, the performance content framework can be a structured framework formed by associating indicator information with corresponding target performance content fragments. The performance content framework can be used to integrate key information of indicator requirements and corresponding performance evidence.

[0069] In this embodiment, target performance content segments that meet the verification conditions are selected from at least one performance content segment based on the indicator information in the education contract text to be verified. The indicator information is then associated with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified. The indicator information can be clarified by first parsing the education contract text to be verified through a language model adapted to the education field. Then, relevant target performance content segments are selected from the performance content segments through semantic matching. Finally, each indicator information can be bound to the corresponding target performance content segment according to a preset structure to form a performance content framework that includes the correspondence between indicators and evidence.

[0070] For example, if the indicator information identified from the education contract text to be verified is that the number of students admitted by School A at the end of the first academic year is greater than or equal to 5, then the segments containing the number of students admitted by School A and the relevant information at the end of the first academic year can be selected from the performance content segments. Then, the indicator information can be associated with the selected segments to form a structured performance content framework of indicator (the number of students admitted by School A at the end of the first academic year is greater than or equal to 5) and target performance content segment (School A admitted a total of 3 students at the end of the first academic year).

[0071] Step 104: Verify the execution status of the education contract text to be verified based on the performance content framework, and obtain the performance status information corresponding to the education contract text to be verified.

[0072] In this embodiment of the application, the verification of the execution status can be a process of determining whether the terms of the educational contract to be verified meet the preset requirements of the indicator information based on the corresponding information between the indicators and evidence integrated in the performance content framework.

[0073] In this embodiment of the application, the execution verification can be a process of determining whether the terms of the education contract text to be verified meet the requirements based on the associated information.

[0074] In this embodiment of the application, the performance status information can be the verified output of the terms execution result information. For example, the performance status information in this embodiment of the application may specifically include three status information: fulfilled, partially fulfilled, and not fulfilled.

[0075] In this embodiment of the application, the execution status of the education contract text to be verified is verified based on the performance content framework to obtain the performance status information corresponding to the education contract text to be verified. The verification elements (such as target entities, target values, operators, etc.) of each indicator information and the verification parameters (such as actual values, status codes, etc.) of the corresponding target performance content segments can be extracted from the performance content framework first. Then, through multi-dimensional fusion analysis, it is determined whether the verification parameters meet the requirements of the verification elements, and clear performance status information can be output.

[0076] Compared with existing technologies, this embodiment achieves complete collection of basic information required for verification by acquiring the educational service project's contract text to be verified and the corresponding multi-dimensional performance documents. By uniformly representing at least one multimedia content format in the multi-dimensional performance documents, at least one performance content fragment corresponding to the multi-dimensional performance documents is obtained, improving the recognition and integration of different multimedia content formats. By identifying indicator information in the contract text to be verified and associating the indicator information with the target performance content fragment, a performance content framework corresponding to the contract text to be verified is obtained, reducing subjective intervention in manually selecting verification content. Based on the performance content framework, the execution status of the contract text to be verified is obtained to acquire performance status information corresponding to the contract text, improving the accuracy of the verification results.

[0077] As an optional approach, when performing the task of "unified representation of at least one multimedia content in a multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document," the following methods can be used, but are not limited to them: Figure 2 As shown, the method includes:

[0078] Step 201: Perform text content recognition on the multi-dimensional performance documents to obtain at least one text block, and determine the first performance content segment corresponding to the multi-dimensional performance documents based on at least one text block.

[0079] In the embodiments of this application, a text block can be a local text region obtained by dividing the text content in a multi-dimensional performance document into page layouts, and each text block can have a clear semantic boundary. For example, the text blocks in the embodiments of this application may specifically include title blocks, paragraph blocks, chart description blocks, etc.

[0080] In the embodiments of this application, the first performance content fragment may be a text fragment extracted from a text block that is related to indicator information or text information that has been semantically encoded.

[0081] In this embodiment of the application, text content recognition is performed on multi-dimensional performance documents to obtain at least one text block, and the first performance content segment corresponding to the multi-dimensional performance document is determined based on at least one text block. This can be achieved by first using a document layout analysis tool based on deep learning, such as the LayoutLM series model, to segment the layout of multi-dimensional performance documents in PDF or Word format and identify different types of text blocks such as titles, paragraphs, and chart areas. Subsequently, a pre-trained model (such as the BERT model) fine-tuned on educational reports can be used to sequence encode each text block. The contextual dependencies between text blocks are captured through a cross-sentence attention mechanism (i.e., the self-attention layer in Transformer) to understand the overall semantics of the text. Natural language questions can be automatically constructed based on the quantitative indicators identified in subsequent steps (e.g., when the quantitative indicator is the number of teacher training sessions, the question could be how many teacher training sessions have been completed). Reading comprehension (MRC) technology can be used to input the question and text block into the MRC model (such as a question-answering model based on BERT). The answer span is accurately extracted through a pointer network. Finally, the text block including the answer span or the extracted answer information can be determined as the first performance content segment.

[0082] For example, when the multi-dimensional performance document is a record of the teacher training process, the training implementation status can be divided into paragraph blocks through page segmentation. Text information of a total of 8 teacher training sessions completed can be extracted through coding and MRC technology. The completion of 8 teacher training sessions constitutes the first performance content segment.

[0083] Step 202: Perform image content recognition on the multi-dimensional performance document to obtain at least one color-marked region, and determine the second performance content segment corresponding to the multi-dimensional performance document based on the at least one color-marked region.

[0084] In this embodiment, the color-marked area can be a region in the image content of a multi-dimensional performance document marked with a specific color (such as red, green, or yellow), and the color-marked area can be used to represent specific business semantics. For example, the color-marked area in this embodiment may specifically include red marking "not achieved," green marking "achieved," and yellow marking "in progress."

[0085] In this embodiment of the application, the second performance content fragment may be a structured fragment containing visual features, text semantic features and execution status information corresponding to the color-marked area.

[0086] In the embodiments of this application, image content recognition is performed on the multi-dimensional performance document to obtain at least one color-marked region, and the second performance content segment corresponding to the multi-dimensional performance document is determined based on the at least one color-marked region. The color-marked region can be segmented by image content recognition, relevant visual features and corresponding text semantic features can be extracted and fused, and then the execution status information can be identified to determine the second performance content segment.

[0087] Step 203: Perform table structure recognition on the multi-dimensional performance documents to obtain at least one table data, and perform structure transformation on the table data to obtain the third performance content fragment corresponding to the multi-dimensional performance documents.

[0088] In the embodiments of this application, the tabular data can be the raw data contained in the form of tables in multi-dimensional performance documents. For example, the tabular data in the embodiments of this application may specifically include text, numerical information, etc. within cells.

[0089] In this embodiment of the application, the third performance content fragment can be structured data that has been structurally transformed and correctly associated with the table header.

[0090] In the embodiments of this application, at least one table data is obtained by performing table structure recognition on the multi-dimensional performance documents, and the table data is transformed to obtain the third performance content fragment corresponding to the multi-dimensional performance documents. The cell and position can be determined by table structure recognition, a table structure diagram is constructed to analyze the row and column belonging relationship, and then the third performance content fragment is obtained through structure transformation.

[0091] Optionally, when performing the action of "recognizing image content of a multi-dimensional performance document to obtain at least one color-marked region, and determining the second performance content segment corresponding to the multi-dimensional performance document based on the at least one color-marked region", the following methods may be used, but are not limited to: recognizing image content in the multi-dimensional performance document; performing pixel analysis on the image content to segment the image content into at least one color-marked region; extracting visual features corresponding to the at least one color-marked region; determining the target text content corresponding to the visual features from the multi-dimensional performance document, and extracting text semantic features from the target text content; concatenating the text semantic features and visual features, and recognizing the execution status information corresponding to the image content based on the concatenated features, and determining the second performance content segment from the image content based on the execution status information.

[0092] In this embodiment, pixel analysis can be an analysis process that classifies image pixels point by point using an instance segmentation model to distinguish between color-marked regions and background regions.

[0093] In the embodiments of this application, visual features can be feature parameters that characterize the visual attributes of the color-marked area. Visual features may include, but are not limited to, color histograms, RGB mean, and luminance variance.

[0094] In this embodiment of the application, the target text content can be the text content that corresponds to the color-marked area in spatial location, and the target text content can be extracted using OCR technology.

[0095] In the embodiments of this application, the execution status information may be the business semantics carried by the color-marked areas. For example, the execution status information in the embodiments of this application may specifically include "achieved," "not achieved," and "in progress."

[0096] In this embodiment of the application, identifying the image content in a multi-dimensional performance document and performing pixel analysis on the image content to segment the image content into at least one color-marked region can be used to identify the image content in a multi-dimensional performance document. This can be achieved by using an instance segmentation model in computer vision (such as the Mask R-CNN model) to perform pixel-level analysis on the image content, which can segment out color-marked regions (including highlighted areas, color blocks, colored fonts, etc.).

[0097] In this embodiment, extracting visual features corresponding to at least one color-marked region can be done by extracting visual features of the color-marked region. Visual features may include color histograms in HSV space, RGB mean, etc. For example, extracting a 256-dimensional color histogram vector of the green-marked region; then, associating the segmented color-marked region with the text content extracted by OCR technology through spatial coordinate alignment to determine the target text content corresponding to the visual features, and then extracting text semantic features from the target text content (a 768-dimensional vector can be obtained by embedding through a BERT model).

[0098] In this embodiment, textual semantic features and visual features are concatenated, and execution state information corresponding to the image content is identified based on the concatenated features. The second performance content segment is then determined from the image content based on the execution state information. First, the textual semantic features and visual features are concatenated to form a 1024-dimensional input vector, which is then input into a constructed lightweight classification network. This lightweight classification network can be a multilayer perceptron (MLP) classification network, which may include an input layer, two hidden layers (dimensions 512 and 256 respectively), and an output layer. The hidden layers can use the ReLU activation function and Dropout (ratio 0.3) to prevent overfitting, and the output layer can use a Softmax classifier to output a predefined business state probability distribution. Training the lightweight classification network can use cross-entropy loss and the Adam optimizer to achieve end-to-end mapping from pixels to business semantics. The second performance content segment is then determined from the image content based on the execution state information identified by the network.

[0099] For example, to determine the second fulfillment content segment of the achievement report, we can first perform pixel analysis on the image content of the achievement report to segment out the unfulfilled areas marked in red; we can extract the HSV color histogram and RGB mean of the unfulfilled areas as visual features, and find the corresponding text "average exam score improvement of 3 points" by spatial coordinate alignment; we can extract the BERT embedding vector of the text as text semantic features, concatenate the two types of features and input them into a classification network to identify the unfulfilled execution status information; we can then determine the red-marked areas and the corresponding text and execution status information as the second fulfillment content segment.

[0100] Optionally, when performing the task of "identifying the table structure of a multi-dimensional performance document to obtain at least one table data, and performing structural transformation on the table data to obtain a third performance content fragment corresponding to the multi-dimensional performance document", the following methods may be used, but are not limited to: identifying the table content in the multi-dimensional performance document, determining at least one cell in the table content and the cell position corresponding to the at least one cell; using the at least one cell as a graph node based on the cell position, and establishing connecting edges between the at least one cell to generate a table structure graph corresponding to the at least one cell; analyzing the row and column relationships between the at least one cell based on the table structure graph, and performing structural transformation on the at least one cell based on the row and column relationships to obtain a third performance content fragment.

[0101] In this embodiment, the cell position can be the coordinate information of the cell in the table image, and the cell position can include the x-coordinate of the top left corner, the y-coordinate of the top left corner, the cell width, and the cell height;

[0102] In this embodiment of the application, the connecting edge can be an edge established based on the spatial overlap and semantic similarity of the cell and used to represent the cell association relationship. The edge weight of the connecting edge can be calculated by weighted sum of spatial overlap rate and semantic similarity.

[0103] In the embodiments of this application, the table structure diagram can be a structure diagram formed by taking the cells in the table as nodes and establishing connecting edges based on the relationships between cells. The table structure diagram can be used to represent the logical relationships between the rows and columns of the table.

[0104] In the embodiments of this application, the row and column affiliation relationship can be the subordinate relationship between a cell and the rows and columns of a table.

[0105] In this embodiment of the application, identifying the table content in the multi-dimensional performance document and determining at least one cell in the table content and the cell position corresponding to at least one cell can be done by first identifying the table content in the multi-dimensional performance document, using an object detection model (which can be the YOLO model) to locate each cell in the table image, and determining the spatial coordinates (x, y, width, height) of each cell.

[0106] In the embodiments of this application, at least one cell is used as a graph node based on its position, and connecting edges are established between at least one cell to generate a table structure graph corresponding to at least one cell. This can be achieved by using each cell as a graph node based on its coordinates. The graph node features can include the spatial coordinates and semantic content of the cell (vectors obtained by BERT embedding after OCR extraction of text). Edge weights can be calculated based on the spatial overlap and semantic similarity of the cells (calculated by cosine similarity) to construct an adjacency matrix. For example, connecting edges are established between spatially adjacent and content-related cells. Then, a graph convolutional network (GCN) layer can be used for message passing. Each node aggregates the features of its neighboring nodes to update its own representation. The row and column affiliation relationships between cells can be inferred through multiple layers of GCN (e.g., 2 layers). Finally, a clustering algorithm (e.g., DBSCAN algorithm) can be used to group the nodes by rows and columns, parse cross-row or cross-column merging relationships, and convert the table data into structured data in JSON format, where each data point is correctly associated with the table header. The structured data in JSON format can be a third-party content fragment.

[0107] Optionally, when performing the task of "selecting target performance content segments that meet the verification conditions from at least one performance content segment based on the indicator information in the education contract text to be verified, and associating the indicator information with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified," the following methods may be used, but are not limited to: using a target language model to perform deep semantic parsing on the education contract text to be verified to obtain the indicator information in the education contract text to be verified, wherein the target language model is obtained through adaptive training on corpora in the education domain; and associating the indicator information and at least one performance content segment in the same semantic vector space. The process involves encoding the data to obtain indicator vectors corresponding to indicator information and at least one content vector corresponding to at least one performance content segment; determining the semantic similarity data between the indicator vectors and at least one content vector; selecting segments with similarity greater than a similarity threshold from at least one performance content segment based on the semantic similarity data, and using the selected segments as candidate performance content segments; performing information alignment and segment extraction operations on the candidate performance content segments to select target performance content segments that meet the verification conditions; and associating the indicator information with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified.

[0108] In this embodiment, the target language model can be a semantic understanding model formed by adaptive pre-training and supervised fine-tuning on educational corpora based on a basic pre-trained model. The target language model can identify entities unique to the educational field and the relationships between entities, and convert non-standard contract terms into structured indicators.

[0109] In this embodiment of the application, the corpus of education domain can include text data such as education industry contracts, policy documents, and education reports. The corpus of education domain can be used to enable the model to master professional terms and expression styles such as student-teacher ratio, college admissions, and teaching and research project completion rate.

[0110] In the embodiments of this application, deep semantic parsing can be a process of entity recognition, relationship classification, and descriptive indicator quantification of the educational contract text to be verified. Deep semantic parsing can be used to extract structured indicator information.

[0111] In the embodiments of this application, the semantic vector space can be a high-dimensional vector space. After the text information in the semantic vector space is mapped into vectors, the cosine distance between the vectors can reflect the semantic similarity.

[0112] In this embodiment of the application, the semantic similarity data can be the similarity score between the index vector and the content vector calculated by the cosine similarity algorithm. The score range of the semantic similarity data can be between [-1, 1]. The closer the score of the semantic similarity data is to 1, the more related the semantics are.

[0113] In this embodiment of the application, the similarity threshold can be a value dynamically determined based on the mean and standard deviation of the similarity scores. The similarity threshold can be used to filter semantically related performance content fragments.

[0114] In the embodiments of this application, the candidate performance content fragment can be a performance content fragment with a similarity score exceeding a threshold.

[0115] In this embodiment of the application, information alignment can be a process of matching information in candidate performance content fragments with elements of indicator information to ensure that the extracted information can be used for verification.

[0116] In the embodiments of this application, the fragment extraction operation can be the process of extracting core information fragments that are directly related to indicator information from candidate performance content fragments.

[0117] In this embodiment, using a target language model to perform deep semantic parsing of the educational contract text to be verified to obtain the indicator information in the educational contract text can be achieved by first training the target language model: the BERT-base Chinese model can be selected as the basic pre-training model, and domain-adaptive pre-training can be performed on the educational domain corpus. The training parameters can be set to a learning rate of 2e-5, a batch size of 32, and training for 3 epochs. Then, manually annotated contract texts and structured indicator data from the historical project database are input into the model for supervised fine-tuning, and the model is trained to perform specific information extraction tasks. The model can be used to parse the educational contract text to be verified by using the BIO annotation architecture to segment the text into word sequences, with each word labeled as B (entity start), I (entity interior), or O (entity exterior), to identify deliverables, quantified targets, and time nodes. Entity boundaries are defined; word sequences are encoded using a shared Transformer encoder BERT to generate context vectors, which are then connected in parallel between an entity recognition head (Softmax classifier predicts BIO labels) and a relation classification head (predicts relation types based on entity pairs). The model is optimized using a weighted loss function (a weighted sum of entity recognition loss and relation classification loss), outputting (entity, relation, entity) triples. For descriptive clauses that significantly improve the average exam score, semantic role labeling can be used to identify core actions that improve performance, significantly enhance the level of modification, and outperform the benchmark. The k-nearest neighbor algorithm (k value can be dynamically adjusted, default k=5) is used to calculate the semantic similarity with historical descriptions (based on word vector cosine similarity), find the numerical distribution of the most similar descriptions, determine the quantification threshold range (such as mean plus or minus standard deviation), and form complete indicator information.

[0118] In this embodiment, the indicator information and at least one performance content fragment are encoded in the same semantic vector space to obtain the indicator vector corresponding to the indicator information and at least one content vector corresponding to the at least one performance content fragment. This can be achieved by using a semantic encoder (such as Sentence-BERT) that shares the inputs of the indicator information and the performance content fragment. The two are then converted into 768-dimensional high-dimensional vectors (indicator vector and content vector) through a Siamese network structure and an average pooling layer, thus achieving encoding in the same semantic vector space.

[0119] In this embodiment of the application, the semantic similarity data between the indicator vector and at least one content vector can be obtained by calculating the similarity between the indicator vector and each content vector using the cosine similarity formula.

[0120] In the embodiments of this application, based on semantic similarity data, segments with similarity greater than a similarity threshold are selected from at least one performance content segment, and the selected segments are used as candidate performance content segments. This can be done by selecting the top K (e.g., K=3) performance content segments with similarity as initial candidate performance content segments, calculating the mean and standard deviation of all similarity scores, and the dynamic threshold can be the mean + N × standard deviation. Segments with scores exceeding the dynamic threshold in the initial candidates can be determined as candidate performance content segments.

[0121] In this embodiment of the application, information alignment and segment extraction operations are performed on candidate performance content segments to select target performance content segments that meet the verification conditions. This can be achieved by activating the information alignment module and parsing verification elements (target entity, operator, target value, time node, etc.) from the indicator information. For example, the verification elements for the indicator information that the number of students admitted to school A at the end of the first academic year is greater than or equal to 5 are the target entity - the number of students admitted to school A, the operator - greater than or equal to, the target value - 5, and the time node - the end of the first academic year. Verification parameters can be extracted from the candidate performance content segments (text segments extract actual values ​​through the MRC model, table segments read actual values ​​through JSON data, and image segments obtain visual state codes). If the verification parameters can completely match the verification elements, the corresponding candidate segment can be determined as the target performance content segment.

[0122] In this embodiment of the application, the indicator information is associated with the target performance content fragments to obtain the performance content framework corresponding to the education contract text to be verified. This can be done by associating each indicator information with one or more target performance content fragments in a structured format. During the association process, the verification elements of the indicator information (target entity, operator, target value, time node, etc.) can be matched with the verification parameters of the target performance content fragments (actual value of evidence, confidence level of evidence value, visual state coding, text implication score, etc.) to construct a standardized indicator and evidence association structure.

[0123] For example, the structure of the performance content framework may include: KPI_ID, target entity, operator, target value, time node, target performance content fragment identifier, actual evidence value, evidence value confidence, visual state encoding, text implication score, etc., where each field may come from the indicator information parsing result and the target performance content fragment extraction result.

[0124] Optionally, when performing the "information alignment and segment extraction operation on candidate performance content segments to select target performance content segments that meet the verification conditions from the candidate performance content segments", the following methods may be used, but are not limited to: obtaining student teaching feedback data on the target teaching content; extracting verification information corresponding to indicator vectors from candidate performance content segments, and associating the verification information with indicator vectors to obtain at least one set of verification association information; extracting verification element information from at least one set of verification association information, and extracting verification parameter information from multi-dimensional performance documents; and selecting target performance content segments that meet the verification conditions from candidate performance content segments based on the verification element information and verification parameter information.

[0125] In this embodiment, the verification information may be information related to the semantics of the indicator vector within the candidate performance content fragment. For example, the verification information in this embodiment may specifically include text descriptions, numerical data, status markers, etc.

[0126] In this embodiment of the application, the verification association information can be information formed by binding verification information with the corresponding indicator vector.

[0127] In the embodiments of this application, the verification element information can be the core elements used for verification in the indicator vector. For example, the verification element information in the embodiments of this application may specifically include target entities, operators, target values, time nodes, etc.

[0128] In this embodiment of the application, the verification parameter information may be the actual data, status code, etc., corresponding to the verification element information in the candidate performance content segment, and the verification parameter information may be used to prove the execution status of the terms.

[0129] Optionally, when performing the step of "associating indicator information with target performance content fragments to verify the execution status of the education contract text to be verified and obtaining the performance status information corresponding to the education contract text to be verified", the following methods may be used, but are not limited to: generating at least one verification unit group corresponding to the indicator vector based on the target performance content fragments, and generating at least one verification feature vector corresponding to the at least one verification unit group; using the indicator vector and the target performance content fragments as verification premises for at least one verification feature vector, verifying at least one verification feature vector, and obtaining the probability distribution information corresponding to at least one verification feature vector; and generating the performance status information corresponding to the education contract text to be verified based on the probability distribution information.

[0130] In this embodiment of the application, the verification unit group can be a unit formed by structurally combining the verification elements of indicator information with the verification parameters of the target performance content fragment. Its structure can be (KPI_ID, target entity, operator, target value, actual value of evidence, confidence level of evidence value, visual state code, text implication score).

[0131] In this embodiment of the application, the verification feature vector can be a numerical feature vector that can be processed by a classifier by converting heterogeneous information in the verification unit group, so as to comprehensively reflect the matching of verification elements and verification parameters.

[0132] In this application embodiment, the verification premise can be the basis for judging the execution status, namely the requirements of the indicator information and the actual situation of the target performance content segment.

[0133] In the embodiments of this application, the probability distribution information can be the probability values ​​output by the classifier corresponding to the verified feature vectors of achieved, in progress, and not achieved states.

[0134] In the embodiments of this application, the performance status information can be the status information with the highest probability, and the performance status information can be the final result of the execution of the terms.

[0135] For embodiments of this application, generating at least one verification unit group corresponding to the indicator vector based on the target performance content fragment may include: KPI_ID can be a unique identifier of indicator information; target entity, operator, and target value can be extracted from indicator information; actual evidence value can be extracted from the target performance content fragment (text fragments are generated by the MRC model for question extraction, table fragments are read by JSON data, and image fragments are converted by visual state encoding); the confidence level of the evidence value can be the reliability of the actual value; the visual state encoding is the business state one-hot vector of the image fragment (e.g., green corresponds to [0,1,0]); and the text implication score is the implication probability of the indicator text and evidence text output by the natural language inference model.

[0136] For the embodiments of this application, generating at least one verification feature vector corresponding to at least one verification unit group can first convert the heterogeneous information of the verification unit group into numerical features: calculate the difference between the actual value of the evidence and the target value and normalize it to obtain numerical comparison features; input the indicator text and the evidence text into the natural language reasoning model, and output the probability distribution of implication, neutrality and contradiction, and take the implication probability as the text implication score feature; visual state encoding can be used as visual state feature; embed the features into 64 dimensions through independent fully connected layers to form different modal feature vectors.

[0137] In this embodiment, the indicator vector and the target performance content fragment are used as verification premises for at least one verification feature vector. The at least one verification feature vector is verified to obtain the probability distribution information corresponding to the at least one verification feature vector. This can be done by using the indicator vector and the target performance content fragment as verification premises and employing a multimodal fusion classifier for verification. Numerical comparison features can be used as query vectors, textual implied features as key vectors, and visual state features as value vectors. Feature weight allocation can be learned through a multi-head attention mechanism to automatically determine feature dependencies in different scenarios (e.g., performance indicators depend on numerical comparison features, while qualitative evaluation indicators depend on text or visual features). The fusion vector output from the attention layer can be input into a multilayer perceptron, and the probability distribution information of the state can be output through the Softmax function. Based on the probability distribution information, the performance status information corresponding to the educational contract text to be verified can be generated.

[0138] Optionally, embodiments of this application also provide an example of intelligent verification of educational service contract performance based on multimodal semantic understanding, such as... Figure 3a and Figure 3b As shown, Figure 3a It includes a unified representation module for multi-dimensional performance documents (i.e., the processing module in this application embodiment that performs unified representation of at least one multimedia content in multi-dimensional performance documents). Figure 3b The system includes an indicator analysis and association verification module for the education contract text to be verified (i.e., the processing module in this application embodiment that identifies indicator information in the education contract text to be verified, selects target performance content segments, and associates them for verification). The two modules work together to verify the text content.

[0139] Optional, Figure 3aThe unified representation module for multi-dimensional performance documents can take multi-dimensional performance documents (text, images, or tables) as input. Through text-based evidence processing (LayoutLM, BERT, and MRC), it completes document layout segmentation (title / paragraph / chart area), text block sequence encoding (across attention mechanism context), and text information extraction. MRC technology can automatically construct questions and accurately extract the answer span through a pointer network. Image-based evidence processing (MaskR-CNN+MLP) achieves pixel-level segmentation of color-marked regions (highlights / color blocks / colored fonts), visual feature extraction (color histogram (HSV space), RGB mean), and completes visual-text feature fusion by aligning OCR text and visual features using spatial coordinates. The lightweight MLP classification network takes 256-dimensional visual features and 768-dimensional text features as input, processes them through a hidden layer (512→256, ReLU+Dropout0.3), and outputs business status probabilities (e.g., green: achieved, red: not achieved). After cell localization through tabular evidence processing (YOLO, GCN, and DBSCAN), graph nodes are constructed (spatial coordinates + text BERT embedding). Based on spatial overlap and cosine weights, an adjacency matrix is ​​generated to construct edge relationships. After row and column attribution through two layers of GCN message passing inference units and DBSCAN clustering to complete row and column grouping and cell parsing and merging, the final output is JSON structured data, forming multimodal structured evidence fragments of multidimensional performance documents.

[0140] Optional, Figure 3b The indicator analysis and correlation verification module of the education service contract text to be verified can take the education service contract text (education contract text to be verified) as input, extract key entities and relationships, quantify and map descriptive indicators to obtain structured verification indicators (KPIs), and then associate the output multimodal structured evidence fragments through semantic association addressing (unified semantic representation and deep semantic matching). After multimodal fusion verification (including operations such as parsing verification elements from KPIs, extracting verification parameters from evidence, information alignment, multimodal feature attention fusion, classification decision, etc.), the final performance status verification result is output after constructing the structured verification tuple.

[0141] Compared with existing technologies, this embodiment achieves the classification and extraction of multiple types of performance content by performing text content recognition, image content recognition, and table structure recognition on multi-dimensional performance documents to obtain corresponding performance content fragments; it improves the effectiveness of identifying business information in image-based content by performing pixel analysis and feature stitching on the image content of multi-dimensional performance documents to determine the second performance content fragment; it achieves the structured transformation of table data by constructing a table structure diagram and analyzing the row and column relationships of the table content of multi-dimensional performance documents to obtain the third performance content fragment; it improves the matching accuracy between indicator information and performance content by parsing the indicator information of the education contract text to be verified through a target language model, encoding the indicator information and performance content fragments, and selecting the target performance content fragment; and it improves the alignment of verification content by extracting verification information from candidate performance content fragments and associating it with indicator information to select the target performance content fragment.

[0142] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a device for verifying the performance of an education contract text, such as... Figure 4 As shown, the device includes: an acquisition module 31, a characterization module 32, an association module 33, and a verification module 34.

[0143] Module 31 is configured to acquire the text of the education contract to be verified for the education service project and the multi-dimensional performance documents corresponding to the text of the education contract to be verified.

[0144] The characterization module 32 is configured to uniformly characterize at least one multimedia content in the multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document. The at least one multimedia content includes at least one of text content, image content and chart content.

[0145] The association module 33 is configured to select a target performance content segment that meets the verification conditions from at least one performance content segment based on the indicator information in the education contract text to be verified, and associate the indicator information with the target performance content segment to obtain the performance content framework corresponding to the education contract text to be verified.

[0146] The verification module 34 is configured to verify the execution status of the education contract text to be verified based on the performance content framework, and obtain the performance status information corresponding to the education contract text to be verified.

[0147] In some examples of this embodiment, the association module 33 is specifically configured to perform text content recognition on the multi-dimensional performance document to obtain at least one text block, and determine a first performance content segment corresponding to the multi-dimensional performance document based on the at least one text block; perform image content recognition on the multi-dimensional performance document to obtain at least one color-marked region, and determine a second performance content segment corresponding to the multi-dimensional performance document based on the at least one color-marked region; perform table structure recognition on the multi-dimensional performance document to obtain at least one table data, and perform structure transformation on the table data to obtain a third performance content segment corresponding to the multi-dimensional performance document.

[0148] In some examples of this embodiment, the association module 33 is further configured to identify image content in a multi-dimensional performance document, perform pixel analysis on the image content to segment the image content into at least one color-marked region; extract visual features corresponding to at least one color-marked region; determine the target text content corresponding to the visual features from the multi-dimensional performance document, and extract text semantic features from the target text content; concatenate the text semantic features and visual features, and identify the execution status information corresponding to the image content based on the concatenated features, and determine the second performance content fragment from the image content based on the execution status information.

[0149] In some examples of this embodiment, the association module 33 is further configured to identify the table content in the multi-dimensional performance document, determine at least one cell in the table content and the cell position corresponding to the at least one cell; take the at least one cell as a graph node based on the cell position, and establish connecting edges between the at least one cell to generate a table structure graph corresponding to the at least one cell; analyze the row and column belonging relationship between the at least one cell based on the table structure graph, and perform structural transformation on the at least one cell based on the row and column belonging relationship to obtain a third performance content fragment.

[0150] In some examples of this embodiment, the association module 33 is further configured to perform deep semantic parsing on the educational contract text to be verified using a target language model to obtain indicator information in the educational contract text. The target language model is obtained through adaptive training on educational domain corpora. The indicator information and at least one performance content fragment are encoded in the same semantic vector space to obtain the indicator vector corresponding to the indicator information and at least one content vector corresponding to the at least one performance content fragment. The semantic similarity data between the indicator vector and the at least one content vector is determined. Based on the semantic similarity data, fragments with similarity greater than a similarity threshold are selected from the at least one performance content fragment, and the selected fragments are used as candidate performance content fragments. Information alignment and fragment extraction operations are performed on the candidate performance content fragments to select target performance content fragments that meet the verification conditions from the candidate performance content fragments. The indicator information and the target performance content fragments are associated to obtain the performance content framework corresponding to the educational contract text to be verified.

[0151] In some examples of this embodiment, the association module 33 is further configured to extract verification information corresponding to the indicator vector from the candidate performance content fragments, and associate the verification information with the indicator vector to obtain at least one set of verification association information; extract verification element information from the at least one set of verification association information, and extract verification parameter information from the multi-dimensional performance documents; and select the target performance content fragment that meets the verification conditions from the candidate performance content fragments based on the verification element information and the verification parameter information.

[0152] In some examples of this embodiment, the association module 33 is further configured to generate at least one verification unit group corresponding to the indicator vector based on the target performance content fragment, and generate at least one verification feature vector corresponding to the at least one verification unit group; use the indicator vector and the target performance content fragment as verification premises for at least one verification feature vector, verify at least one verification feature vector, and obtain probability distribution information corresponding to at least one verification feature vector; and generate performance status information corresponding to the education contract text to be verified based on the probability distribution information.

[0153] It should be noted that other corresponding descriptions of the functional units involved in the performance verification device for educational contract text provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0154] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.

[0155] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0156] like Figure 5 The diagram shown is a hardware structure schematic of an electronic device according to the present invention, comprising:

[0157] At least one processor 401; and,

[0158] Memory 402 is communicatively connected to at least one processor 401; wherein,

[0159] The memory 402 stores instructions that can be executed by at least one processor to enable the at least one processor to perform the performance verification method of the aforementioned education contract text.

[0160] Figure 5 Take a processor 401 as an example.

[0161] The electronic device may also include an input device 403 and an output device 404.

[0162] The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0163] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the performance verification method of the education contract text in the embodiments of this application, for example, Figure 1 and Figure 2 The method flow is shown. The processor 401 executes various functional applications and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing the performance verification method of the education contract text in the above embodiment.

[0164] Memory 402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the use of the performance verification method for the educational contract text. Furthermore, memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, which can be connected via a network to means of performing the performance verification method for the educational contract text. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0165] Input device 403 can receive user clicks and generate signal inputs related to user settings and function control of the performance verification method for the education contract text. Output device 404 may include display devices such as a display screen.

[0166] One or more modules are stored in memory 402, and when run by one or more processors 401, the method for verifying the performance of the education contract text in any of the above method embodiments is executed.

[0167] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0168] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0169] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0170] Through 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, or it can be implemented by hardware. By applying the solution of this embodiment, compared with the existing technology, this embodiment can achieve complete collection of the basic information required for verification by obtaining the educational contract text to be verified for the educational service project and the multi-dimensional performance documents corresponding to the educational contract text to be verified; by uniformly representing at least one multimedia content in the multi-dimensional performance documents, at least one performance content fragment corresponding to the multi-dimensional performance documents is obtained, improving the recognition and integration of different multimedia content; by identifying the indicator information in the educational contract text to be verified and associating the indicator information with the target performance content fragment, the performance content framework corresponding to the educational contract text to be verified is obtained, reducing the subjective intervention of manual selection of verification content; and by verifying the execution status of the educational contract text to be verified based on the performance content framework, the performance status information corresponding to the educational contract text to be verified is obtained, improving the accuracy of the verification results. The system achieves classification and extraction of various types of performance content by performing text content recognition, image content recognition, and table structure recognition on multi-dimensional performance documents to obtain corresponding performance content fragments. It also improves the effectiveness of identifying business information in image-based content by performing pixel analysis and feature stitching on the image content of multi-dimensional performance documents to determine the second performance content fragment. Furthermore, it achieves structured transformation of table data by constructing a table structure diagram from the table content of multi-dimensional performance documents and analyzing row and column relationships. Finally, it improves the accuracy of matching indicator information with performance content by parsing the indicator information of the education contract text to be verified using a target language model, encoding the indicator information with performance content fragments, and selecting the target performance content fragment. Finally, it improves the alignment of verification content by extracting verification information from candidate performance content fragments and associating it with indicator information to select the target performance content fragment.

[0171] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0172] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for verifying the performance of an education contract, characterized in that, include: Obtain the text of the education contract to be verified in the education service project and the multi-dimensional performance documents corresponding to the text of the education contract to be verified; The content of at least one multimedia format in the multi-dimensional performance document is uniformly represented; the uniform representation is to convert the content of different multimedia formats in the multi-dimensional performance document into a standardized information format that can be processed and associated by a machine, so as to obtain at least one performance content fragment corresponding to the multi-dimensional performance document, wherein the at least one multimedia format includes at least one of text content, image content and chart content. Based on the indicator information in the education contract text to be verified, a target performance content segment that meets the verification conditions is selected from the at least one performance content segment, and the indicator information is associated with the target performance content segment to obtain the performance content framework corresponding to the education contract text to be verified; the indicator information is structured information extracted from the education contract text to be verified that can be used to measure the performance of the clauses. The execution status of the education contract text to be verified is verified based on the aforementioned performance content framework to obtain the performance status information corresponding to the education contract text to be verified.

2. The method according to claim 1, characterized in that, The step of uniformly representing at least one multimedia content in the multi-dimensional performance document to obtain at least one performance content fragment corresponding to the multi-dimensional performance document includes: The text content of the multi-dimensional performance document is recognized to obtain at least one text block, and the first performance content segment corresponding to the multi-dimensional performance document is determined based on the at least one text block; Image content recognition is performed on the multi-dimensional performance document to obtain at least one color-marked region, and a second performance content segment corresponding to the multi-dimensional performance document is determined based on the at least one color-marked region; The table structure of the multi-dimensional performance document is identified to obtain at least one table data, and the table data is structurally transformed to obtain the third performance content fragment corresponding to the multi-dimensional performance document.

3. The method according to claim 2, characterized in that, The step of performing image content recognition on the multi-dimensional performance document to obtain at least one color-marked region, and determining the second performance content segment corresponding to the multi-dimensional performance document based on the at least one color-marked region, includes: Identify the image content in the multi-dimensional performance document, perform pixel analysis on the image content, and segment the image content into at least one color-marked region; Extract the visual features corresponding to the at least one color-marked region; Determine the target text content corresponding to the visual features from the multi-dimensional performance documents, and extract text semantic features from the target text content; The text semantic features and the visual features are concatenated, and the execution status information corresponding to the image content is identified based on the concatenated features. The second performance content fragment is then determined from the image content based on the execution status information.

4. The method according to claim 2, characterized in that, The step of identifying the table structure of the multi-dimensional performance document to obtain at least one table data, and then performing a structure transformation on the table data to obtain a third performance content fragment corresponding to the multi-dimensional performance document, including: Identify the table content in the multi-dimensional performance document, and determine at least one cell in the table content and the cell position corresponding to the at least one cell; Based on the cell position, at least one cell is used as a graph node, and connecting edges are established between the at least one cell to generate a table structure graph corresponding to the at least one cell; Based on the table structure diagram, the row and column relationships between the at least one cell are analyzed, and the structure of the at least one cell is transformed based on the row and column relationships to obtain the third performance content fragment.

5. The method according to claim 1, characterized in that, The step of selecting target performance content segments that meet the verification conditions from at least one performance content segment based on indicator information in the education contract text to be verified, and associating the indicator information with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified, includes: The target language model is used to perform deep semantic analysis on the educational contract text to be verified to obtain the indicator information in the educational contract text. The target language model is obtained by adaptive training on educational corpus. The indicator information and the at least one performance content fragment are encoded in the same semantic vector space to obtain the indicator vector corresponding to the indicator information and the at least one content vector corresponding to the at least one performance content fragment. Determine the semantic similarity data between the indicator vector and the at least one content vector; Based on the semantic similarity data, segments with a similarity greater than a similarity threshold are selected from the at least one performance content segment, and the selected segments are used as candidate performance content segments. Information alignment and segment extraction operations are performed on the candidate performance content segments to select target performance content segments that meet the verification conditions from the candidate performance content segments; By associating the indicator information with the target performance content fragment, a performance content framework corresponding to the education contract text to be verified is obtained.

6. The method according to claim 5, characterized in that, The step of performing information alignment and segment extraction operations on the candidate performance content segments to select target performance content segments that meet the verification conditions includes: The verification information corresponding to the indicator vector is extracted from the candidate performance content fragment, and the verification information is associated with the indicator vector to obtain at least one set of verification association information; Verification element information is extracted from the at least one set of verification association information, and verification parameter information is extracted from the multi-dimensional performance documents. Based on the verification element information and the verification parameter information, a target performance content segment that meets the verification conditions is selected from the candidate performance content segments.

7. The method according to claim 6, characterized in that, The step of associating the indicator information with the target performance content fragment to verify the execution status of the education contract text to be verified, and obtaining the performance status information corresponding to the education contract text to be verified, includes: Based on the target performance content fragment, generate at least one verification unit group corresponding to the indicator vector, and generate at least one verification feature vector corresponding to the at least one verification unit group; The indicator vector and the target performance content fragment are used as the verification premises for the at least one verification feature vector. The at least one verification feature vector is verified to obtain the probability distribution information corresponding to the at least one verification feature vector. Based on the probability distribution information, the performance status information corresponding to the education contract text to be verified is generated.

8. A device for verifying the performance of an education contract, characterized in that, include: The acquisition module is configured to acquire the text of the education contract to be verified and the multi-dimensional performance documents corresponding to the text of the education contract to be verified in the education service project. The representation module is configured to uniformly represent at least one multimedia content in the multi-dimensional performance document; the uniform representation is to convert the content of different multimedia formats in the multi-dimensional performance document into a standardized information format that can be processed and associated by a machine, to obtain at least one performance content fragment corresponding to the multi-dimensional performance document, wherein the at least one multimedia content includes at least one of text content, image content and chart content. The association module is configured to select target performance content segments that meet the verification conditions from the at least one performance content segment based on the indicator information in the education contract text to be verified, and associate the indicator information with the target performance content segments to obtain the performance content framework corresponding to the education contract text to be verified; the indicator information is structured information extracted from the education contract text to be verified that can be used to measure the performance of the clauses. The verification module is configured to verify the execution status of the education contract text to be verified based on the performance content framework, and obtain the performance status information corresponding to the education contract text to be verified.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images