Bidding document file detection method and device, electronic equipment and storage medium

Abstract

The invention discloses a bidding document file detection method and device, electronic equipment and a storage medium, and relates to the technical field of bidding document file detection. According to the technical scheme, the file specification library of the bidding document files is constructed, and the file specification library is a database formed by the bidding specifications of different files in the bid documents, so that the importance degree of each bidding specification in the file specification library can be quantified to obtain the specification weights of all the bidding specifications; according to the bidding specifications of all the files and the corresponding specification weights, training a preset file detection model to obtain a target detection model; and inputting a target file of the to-be-bid book into the target detection model, and determining a standard detection result of the target file according to an output result of the target detection model. According to the technical scheme, weight configuration is carried out on different bidding specifications, detection of key specifications can be highlighted when the target detection model carries out detection of the target file, and efficiency and accuracy of bidding document file detection are both considered.

Description

Technical Field

[0001] The present invention relates to the technical field of tender document detection, and in particular to a tender document detection method, device, electronic equipment and storage medium. Background Art

[0002] In the traditional bidding process, the detection of bid documents is a crucial link, which ensures the standardization and consistency of the bid documents, thereby ensuring the fairness and transparency of the bidding process. However, the existing bid document detection methods mainly rely on manual review. There are problems such as low efficiency, prone to errors, and difficulty in standardized review. The Chinese invention patent with publication number CN 117764039 A discloses a bid document generation method, system, terminal and storage medium based on a large model. Although its solution uses a large model to implement bid document detection, its detection strategy is implemented based on the same standard, and there is redundant processing for bid document detection and modification.

[0003] Therefore, tender document detection urgently needs an automated, efficient and accurate solution to improve the efficiency and accuracy of tender document review and ensure the standardization and consistency of the review process. Summary of the invention

[0004] In view of the above technical problems, the present invention provides a tender document detection method, device, electronic device and storage medium, which can take into account both the efficiency and accuracy of tender document detection.

[0005] The embodiment of the present invention provides the following solution:

[0006] In a first aspect, an embodiment of the present invention provides a bidding document detection method, the method comprising:

[0007] Constructing a document specification library for bidding documents, wherein the document specification library is a database consisting of bidding specifications of different documents in the bidding documents;

[0008] Quantify the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications;

[0009] The preset document detection model is trained according to the bidding specifications of all documents and the corresponding specification weights to obtain the target detection model;

[0010] The target file of the bid is input into the target detection model, and the standard detection result of the target file is determined according to the output result of the target detection model.

[0011] In an optional embodiment, the importance of each bidding specification in the document specification library is quantified to obtain the specification weight of all bidding specifications, including:

[0012] The information entropy of each bidding specification in the document specification library is calculated according to the bidding document scoring standard to obtain the information entropy of each bidding specification;

[0013] According to the information entropy of each bidding specification, the specification weight of each bidding specification is determined.

[0014] In an optional embodiment, building a file specification library of bid documents includes:

[0015] Extract format features of the tendered documents to obtain the format specifications of the tendered documents;

[0016] Boolean operation processing is performed according to the similarity between each format specification in all format specifications, and a database constructed by the processing results is determined as a file specification library.

[0017] In an optional embodiment, a Boolean operation is performed according to the similarity between the format specifications in all format specifications, and a database constructed by the processing results is determined as a file specification library, including:

[0018] Divide the specification categories according to the similarities between all format specifications to obtain multiple specification sets;

[0019] Performing intersection operation processing and union operation processing on the standard sets of the same category to obtain a first processing result of the intersection operation processing and a second processing result of the union operation processing;

[0020] Update the format specification in the same category specification set according to the first processing result and the second processing result, so as to increase the number of specification items and the specification coverage of the category format specification;

[0021] The updated specification set is determined as the target set of format specifications for this category, and the database constructed by all target sets is determined as the file specification library.

[0022] In an optional embodiment, the file detection model includes a CNN model and an RNN model; the preset file detection model is trained according to the bidding specifications of all files and the corresponding specification weights to obtain a target detection model, including:

[0023] According to the bidding specifications of all documents and the corresponding specification weights, an image training set, an image verification set, a text training set, and a text verification set are constructed;

[0024] The CNN model is trained and verified according to the image training set and the image verification set respectively until the CNN model reaches the first preset condition of training;

[0025] The RNN model is trained and verified according to the text training set and the text verification set respectively until the RNN model meets the second preset condition of training;

[0026] The detection model constructed by the CNN model that meets the first preset condition and the RNN model that meets the second preset condition is determined as the target detection model.

[0027] In an optional embodiment, the target detection model includes a trained CNN model and an RNN model; the target file of the bid to be submitted is input into the target detection model, and a standard detection result of the target file is determined according to an output result of the target detection model, including:

[0028] Divide the target file into image features and text features to be detected;

[0029] The image features and text features are respectively input into the trained CNN model and RNN model to obtain the format detection results of the target file;

[0030] Perform confidence evaluation on the format detection result according to the Bayesian model to obtain an evaluation result;

[0031] When the evaluation result does not reach the target confidence level, the model parameters of the target detection model are updated and re-detected until the evaluation result reaches the target confidence level;

[0032] When the evaluation result reaches the target confidence level, the standard detection result of the target file is output according to the format detection result.

[0033] In an optional embodiment, outputting a standard detection result of the target file according to the format detection result includes:

[0034] The format detection result is input into the language processing model, and the standard detection result is obtained according to the detection report output by the language processing model, wherein the detection report includes the abnormal format location, abnormal format type, abnormal format severity and adjustment suggestions represented by the format detection result.

[0035] In a second aspect, an embodiment of the present invention further provides a tender document detection device, the device comprising:

[0036] A construction module, used to construct a document specification library of the bidding documents, wherein the document specification library is a database composed of bidding specifications of different documents in the bidding documents;

[0037] A processing module is used to quantify the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications;

[0038] A training acquisition module is used to train a preset file detection model according to the bidding specifications of all files and corresponding specification weights to obtain a target detection model;

[0039] The detection module is used to input the target file of the bid to be submitted into the target detection model, and determine the standard detection result of the target file according to the output result of the target detection model.

[0040] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of any one of the methods in the first aspect.

[0041] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods in the first aspect.

[0042] Compared with the prior art, the bidding document detection method, device, electronic device and storage medium of the present invention have the following advantages:

[0043] The technical solution of the present invention constructs a file specification library of the bidding document. Since the file specification library is a database composed of the bidding specifications of different files in the bidding documents, the importance of each bidding specification in the file specification library can be quantified to obtain the specification weights of all bidding specifications; the preset file detection model is trained according to the bidding specifications of all files and the corresponding specification weights to obtain the target detection model; the target file of the bidding document to be bid is input into the target detection model, and the specification detection result of the target file is determined according to the output result of the target detection model. The technical solution implements weight configuration for different bidding specifications, so that the target detection model can highlight the detection of key specifications when implementing the detection of target files, reduce the detection deviation and redundancy of all bidding specifications under the same detection standard, and thus take into account the efficiency and accuracy of the bidding document detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A flowchart of a bidding document detection method provided by an embodiment of the present invention;

[0046] Figure 2 A logical schematic diagram of a bidding document detection method provided by an embodiment of the present invention;

[0047] Figure 3 A schematic structural diagram of a bidding document detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention belong to the scope of protection of the embodiments of the present invention.

[0049] When implementing bidding, the tender documents involve diverse contents, and the bidding specifications of various documents present different requirements. Although the detection efficiency can be improved by implementing detection through traditional large models, the problems with lower importance will be further magnified based on rigid unified standards, which will cause redundant modifications and repeated detections. Automatic modification through large models may even cause the modification results to be inferior to the initial version, resulting in insufficient processing efficiency and accuracy of the tender documents. The following embodiments of the present invention will specifically explain how to solve the above problems.

[0050] See also Figure 1 , Figure 1 The present invention provides a flowchart of a method for detecting a bidding document. The method can be applied to a computer or a server device. By running the method, the document detection of the bidding document is implemented. The method includes:

[0051] S11. Construct a file specification library for the bidding documents, wherein the file specification library is a database composed of bidding specifications of different documents in the bidding documents.

[0052] Specifically, all types of documents in the bid documents comply with the bidding specifications, including the format requirements, file size, clarity requirements, etc. of various types of documents. The file specification library contains all bidding specifications in the bid documents. Bidding specifications can be extracted based on multiple bid documents of the user, and all bidding specifications can be marked. For example, the bidder's profile can be marked as "Company Profile", and the corresponding title font, format and other information can be stored as bidding specifications under this label. It should be noted that the format specifications represented by the bidding specifications include the font type, line spacing, first line indentation and other format requirements of the text in the file.

[0053] Exemplarily, step S11 includes sub-steps S11-1 to S11-2, which are specifically described as follows:

[0054] S11-1. Format feature extraction is performed on the bid documents to obtain the format specifications of the bid documents. When there are multiple bid documents, format feature extraction can be performed on each bid document, and the extraction results of all format specifications of each bid document are determined as the specification set of the bid document, and all elements in the specification set are the format specifications of the bid document. It can be understood that the format specification is only a kind of indication in the bidding specification, and the bidding specification can also include the picture resolution, picture size, picture layout method, etc. in the bid document.

[0055] S11-2, Boolean operation processing is performed according to the similarity between each format specification in all format specifications, and the database constructed by the processing results is determined as a file specification library. Different format specifications can be merged through Boolean operation to reduce repeated format specifications. The similarity between each format specification can be determined based on a calculation model, and the calculation model is constructed based on the corresponding algorithm of cosine similarity and Euclidean distance. By inputting two format specifications into the calculation model, the similarity between the two format specifications can be obtained. If the two format specifications are relatively similar, it means that the two are format specifications of the same category. The format requirements represented by the two are merged. After all similar format specifications are merged, they are constructed into a database to obtain a file specification library.

[0056] Of course, when constructing a file specification library, the similarity of format specifications can also be determined based on fuzzy mathematics theory. First, a fuzzy matching model of format specifications is constructed to detect the similarity between format specifications and standard specifications. Standard specifications are format specifications corresponding to the bidding scoring criteria. The membership function μ(x) of the fuzzy set is used to describe the similarity between format specifications and standard specifications. Fuzzy reasoning rules are used in combination with a weight matrix to comprehensively evaluate the format specifications to determine the weight of each format specification. After the weight is determined, a file specification library is constructed. Format specifications greater than the weight threshold can be included in the file specification library. This method can ensure that the weight and flexibility of the specifications are considered during the detection process.

[0057] Exemplarily, sub-step S11-2 includes the following steps:

[0058] In the first step, the specification categories are divided according to the similarities between all format specifications to obtain multiple specification sets. The similarities between all format specifications are calculated to obtain the similarity between two format specifications. When the similarity is greater than a preset threshold, it means that the two format specifications are format specifications of the same category and are divided into the same format specification set. For example, when there are multiple bids, the format specifications of each bid are divided into categories to obtain the categories to which the different format specifications of each bid belong.

[0059] In the second step, the standard sets of the same category are subjected to intersection operation processing and union operation processing to obtain a first processing result of the intersection operation processing and a second processing result of the union operation processing. The first processing result is the intersection operation processing result of the standard coverage, and the second processing result is the union operation processing result of the number of standard items.

[0060] The third step is to update the format specifications in the same category specification set according to the first processing result and the second processing result to increase the number of specification items and the specification coverage of the format specifications of this category. For example, set A contains the format specification {paragraph alignment a1, font size b1, line spacing c1}, and set B contains the format specification {font size b2, line spacing c2, first line indent d1}. The intersection of sets A and B is {font size b1∩b2, line spacing c1∩c2}, and the union of sets A and B is {paragraph alignment, font size, line spacing, first line indent}. The update is implemented after the intersection operation is processed, which can ensure that the number of specification items of the format specification is the largest and the specification coverage is the largest, so that the updated format specification contains all possible format requirements.

[0061] The fourth step is to determine the updated specification set as the target set of the format specification of this category, and determine the database constructed by all target sets as the file specification library. Taking set A and set B as examples, the target set is {paragraph alignment a1, font size b1∩b2, line spacing c1∩c2, first line indent d1}, and each target set represents the detailed specification items and the maximum coverage of the format specification of this category, making the constructed file specification library more accurate and ensuring that the format specifications in the file specification library are comprehensive and non-redundant.

[0062] At this point, the construction of the document specification library has been completed through the bidding specifications of the bid documents.

[0063] S12. Quantify the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications.

[0064] Specifically, different bidding specifications have corresponding importance relative to the bid document scoring criteria. The importance of each bidding specification can be evaluated by weighted average method and entropy weight method, and the importance of each bidding specification can be quantified. The bidding specifications and corresponding specification weights can be characterized by constructing a relationship table, so that each bidding specification and specification weight establish a one-to-one correspondence. Of course, the importance of each bidding specification can also be quantified based on the experience of technical personnel, and the importance of each bidding specification can be accurately quantified, and no specific restrictions are made here.

[0065] Exemplarily, step S12 includes sub-steps S12-1 to S12-2, specifically including:

[0066] S12-1. Calculate the information entropy of each bidding specification in the document specification library according to the bidding document scoring criteria to obtain the information entropy of each bidding specification. The information entropy can be calculated based on the formula:

[0067]

[0068] Among them, p(x i ) represents the probability of the occurrence of the i-th bidding specification, x i represents the i-th bidding specification, H(x i ) is a numerical value, which reflects the uncertainty of the entire bidding specification. The higher the probability of a bidding specification, the lower its information entropy, which means that this bidding specification may be more important in the bidding document detection. The importance of the bidding specification can be quantified through the information entropy of each bidding specification.

[0069] S12-2. Determine the specification weight of each bidding specification based on the information entropy of each bidding specification. By calculating the information entropy of each bidding specification, the randomness and uncertainty of the bidding specification can be identified, providing a basis for determining the weight of key specifications. The specification weight can also be determined based on a formula, for example, by the formula:

[0070]

[0071] Where m is the total number of bidding specifications, w i is the specification weight of the i-th bidding specification, which is mainly used to guide subsequent model training and data processing. The determination of the specification weight makes the key specifications occupy a higher proportion in format detection, thereby improving the accuracy of detection.

[0072] S13. Train a preset file detection model according to the bidding specifications of all files and the corresponding specification weights to obtain a target detection model.

[0073] Specifically, the file detection model can be configured based on actual needs. It can be a text detection model, or a combination of a text detection model and an image detection model, such as an AIGC model. Training samples can be constructed based on the bidding specifications and the corresponding specification weights, and the training samples can be input into the file detection model to implement model training. The file detection model can automatically identify bidding specifications with higher importance based on the bidding specifications and the corresponding specification weights. After the file detection model is trained, it is tested with a test set. After it is determined that the file detection model can accurately detect bidding specifications based on different specification weights, the trained file detection model can be determined as the target detection model.

[0074] Exemplarily, the configuration file detection model includes a CNN (Convolutional Neural Network) model and an RNN (Recurrent Neural Network) model; step S13 includes sub-steps S13-1 to S13-4, which are specifically described as follows:

[0075] S13-1. According to the bidding specifications of all documents and the corresponding specification weights, construct image training sets, image verification sets, text training sets and text verification sets. Data enhancement technology can be used to expand the training data in each training set, such as using strategies such as word replacement and sentence reorganization to generate new training samples and enhance the generalization ability of the model. PCA dimensionality reduction is performed on the feature vectors extracted in the data preprocessing stage, and the number of principal components k is selected so that the variance contribution rate reaches the preset threshold to reduce the computational complexity.

[0076] S13-2, train and verify the CNN model according to the image training set and the image verification set respectively, until the CNN model reaches the first preset condition of training. The CNN model is used to process local features in text data, such as phrases, sentence structures, etc.

[0077] S13-3, train and verify the RNN model according to the text training set and the text verification set respectively, until the RNN model meets the second preset condition of training. The RNN model is used to process the temporal features in the sequence data, such as the order of the specification.

[0078] To implement the training of CNN models and RNN models, the cross entropy loss function and back propagation algorithm can be used for model training. Regularization technology and Dropout strategy can be introduced to prevent model overfitting. The learning rate decay strategy can be used to optimize the model training process and improve the convergence speed and accuracy.

[0079] The cross entropy loss function is as follows:

[0080]

[0081] in, is the model prediction result, y i is the actual label probability, L is With y i The smaller the value, the more accurate the model's prediction is; conversely, the larger the prediction error is. For example, if the specification C (page margin) requires a page margin of 2 cm, but slightly more or less than that is also acceptable, then the actual label probability of C may be 0.8 or 0.9.

[0082] The back propagation algorithm combines the gradient descent method to update the model parameters through the back propagation algorithm to minimize the cross entropy loss function. Apply L1 or L2 regularization technology to prevent the model from overfitting, where the regularization term is:

[0083]

[0084] Among them, λ is the regularization parameter, which controls the influence of the regularization term on the loss function; θ i is the i-th parameter in the model; n is the total number of model parameters.

[0085] The Dropout strategy randomly discards some neurons during the training process to reduce the model's dependence on specific training samples and enhance the model's generalization ability. The learning rate decay strategy is to gradually reduce the learning rate as the training progresses, such as using the learning rate decay formula

[0086]

[0087] Among them, η is the target learning rate, η0 is the initial learning rate, t is the current training step number, and T is the preset total training step number. The above method can accurately train the CNN model and the RNN model, and reduce the problems such as overfitting after training.

[0088] S13-4, the detection model constructed by the CNN model that meets the first preset condition and the RNN model that meets the second preset condition is determined as the target detection model. The target detection model architecture can be expressed as:

[0089] CNN → RNN → Attention

[0090] That is, the attention mechanism is introduced after the RNN layer: ij =softmax(RNN output × weight matrix), by calculating weights to enhance the model's ability to recognize key format features. The weight calculation formula is:

[0091]

[0092] Among them, e ij is the attention score calculated by the model, α ij According to the feature changes of the current sample, the model can focus more on the features most relevant to the current task.

[0093] It should be noted that the target detection model can detect non-text content such as charts and pictures in the bid based on the CNN model, and the RNN model can detect the text content. The attention mechanism can then evaluate the importance of different parts, thereby improving the detection efficiency of the bidding specifications in the bid.

[0094] S14. Input the target file of the bid to be submitted into the target detection model, and determine the standard detection result of the target file according to the output result of the target detection model.

[0095] Specifically, the target file can be part of the bid document or all of the documents. After inputting it into the target detection model, the bidding specifications such as the format specifications in the bid document can be detected, and the specification detection results can be output after the detection is completed. It can be understood that the specification detection result can be the output result of marking the location that does not meet the bidding specifications, or a detection report can be generated based on a template. It should be noted that when the detection of the target file is implemented based on the target detection model, the representation method of the specification detection result can be determined based on the specification weight. For example, when there is a format problem in the bidding specification whose specification weight is greater than the weight threshold, the format problem is highlighted; on the contrary, if there is a format problem in the bidding specification whose specification weight is less than or equal to the weight threshold, only a conventional reminder is given. Of course, different weight intervals can also be defined, and warnings are implemented for the corresponding specification weights of the bidding specifications to which different format problems belong, so that users can determine whether it is necessary to modify the files with format problems.

[0096] In practical applications, due to the variety of documents to be bid, which may involve the writing format of text or the layout of pictures, single-dimensional detection may result in insufficient detection accuracy. Based on this, the target detection model is configured to include a trained CNN model and RNN model; step S14 includes sub-steps S14-1 to S14-5, which are specifically described as follows:

[0097] S14-1. Divide the target file into image features and text features to be detected. Feature division can be performed on the target file based on computer vision and natural language processing, such as edge detection (Sobel, Canny, etc.) and corner detection (Harris, Shi-Tomasi, etc.) to detect image features in the target file; use scene text detection (STD) algorithms, such as CTPN, EAST, PSENet, etc., to locate the area containing text in the target file to extract text features.

[0098] S14-2, input the image features and text features into the trained CNN model and RNN model respectively to obtain the format detection result of the target file. The CNN model can detect the bidding specifications of the image features and output the first detection result of the image features; the RNN model can detect the bidding specifications of the text features and output the second detection result of the text features. The first detection result and the second detection result constitute the format detection result of the target file.

[0099] S14-3. Perform confidence assessment on the format detection results according to the Bayesian model to obtain the assessment results. The Bayesian model provides a decision-making method based on probability theory, which can be used to assess the confidence of the format detection results. Specifically, it includes the definition of prior probability, calculation of likelihood probability, evidence collection and Bayesian model inference. The confidence assessment through the Bayesian model can ensure the reliability of the detection results.

[0100] S14-4. When the evaluation result does not reach the target confidence level, it indicates that the target detection model may have insufficient detection accuracy. The model parameters of the target detection model are updated and re-detected until the evaluation result reaches the target confidence level.

[0101] S14-5. When the evaluation result reaches the target confidence level, it indicates that the detection result output by the target detection model has a high reliability, and the standard detection result of the target file is output according to the format detection result. The standard detection result can list the location, type and severity of all abnormal formats detected.

[0102] The standard detection results can be represented in a way that is based on the detection requirements. For example, the format detection results can be input into a language processing model, and the standard detection results can be obtained based on the detection report output by the language processing model, wherein the detection report includes the abnormal format location, abnormal format type, severity of the abnormal format and adjustment suggestions represented by the format detection results.

[0103] The language processing model can generate a detection report using natural language, which describes the abnormal format location, abnormal format type, severity of abnormal format, and adjustment suggestions. The abnormal format location is used to specify the location of the abnormal format in the file, such as page number, paragraph, or specific area. The abnormal format type is used to describe the specific type of abnormal format, such as inconsistent fonts, abnormal line spacing, image compression format errors, etc. The severity of the abnormal format is used to evaluate the impact of the abnormality on the overall quality of the file. The severity can be divided into three levels: low, medium, and high. Adjustment suggestions are used to provide specific adjustment suggestions based on the detection results to help users correct abnormal formats.

[0104] The following embodiment of the present invention will take the example of a construction company needing to review multiple bidding documents to explain the overall process of checking the bid documents. The bid documents contain a large amount of text, tables and pictures, and need to be submitted strictly in a specific format. Figure 2 , Figure 2 This is a logical diagram of the tender document detection method. Specifically, it includes:

[0105] S201, data preprocessing. Obtain the bid documents to be submitted and perform data preprocessing on them to ensure data validity; collect a large number of bid documents that have been submitted and perform data preprocessing on them to ensure that the bid document data obtained is valid data of the bid documents that have been submitted.

[0106] S202: Format specification collection: Identify common format specifications, such as title size, page margins, table layout, etc.

[0107] S203, constructing a document specification library and a document standard library. A document specification library is established based on the bidding specifications such as format specifications and drawing specifications collected in step S202; and a document standard library is established based on the scoring criteria of the bid documents to be bid.

[0108] S204: File preprocessing: Preprocess the files of the bid documents, including image definition processing, etc.

[0109] S205: Text analysis: Extract text features from the bid documents.

[0110] S206: Non-text analysis: Extract non-text features from the bid, such as image features.

[0111] S207, model selection: Select a model construction method according to the detection requirements of the tender document.

[0112] S208, model construction: A model composed of a CNN model and an RNN model is configured as a document detection model for a bid document.

[0113] S209, model training. Based on the file standard library, the specification weights of the bidding specifications in the file specification library are determined, and the file detection model is trained based on the bidding specifications and the corresponding specification weights. When training the model, some techniques, such as data enhancement and dimensionality reduction, are used to improve its recognition accuracy and efficiency.

[0114] S210, model optimization. Configure the Bayesian model to perform confidence assessment on the format detection results, and configure the language processing model to fill in the detection report.

[0115] S211, format detection: Input the target file of the bid to be submitted into the trained file detection model.

[0116] S212, result analysis: Determine defects such as abnormal format location and abnormal format type of the target file.

[0117] S213, report generation. Output the test report using natural language generation, and describe the abnormal format location, abnormal format type, severity of the abnormal format, and adjustment suggestions in the test report.

[0118] Based on the same technical concept as the detection method, the embodiment of the present invention also provides a tender document detection device, see Figure 3 , Figure 3 It is a structural schematic diagram of a detection device, and the detection device comprises:

[0119] A construction module 301 is used to construct a file specification library of the bidding documents, wherein the file specification library is a database composed of bidding specifications of different documents in the bidding documents;

[0120] A processing module 302 is used to quantify the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications;

[0121] The training acquisition module 303 is used to train the preset file detection model according to the bidding specifications of all files and the corresponding specification weights to obtain the target detection model;

[0122] The detection module 304 is used to input the target file of the bid to be submitted into the target detection model, and determine the standard detection result of the target file according to the output result of the target detection model.

[0123] In an optional embodiment, the building block includes:

[0124] The first acquisition submodule is used to extract format features of the tendered documents to obtain the format specifications of the tendered documents;

[0125] The first determination submodule is used to perform Boolean operation processing according to the similarity between each format specification in all format specifications, and determine the database constructed by the processing results as the file specification library.

[0126] In an optional embodiment, the first determining submodule includes:

[0127] A first obtaining unit is used to divide specification categories according to similarities between all format specifications to obtain multiple specification sets;

[0128] A second obtaining unit is used to perform intersection operation processing and union operation processing on the standard sets of the same category to obtain a first processing result of the intersection operation processing and a second processing result of the union operation processing;

[0129] An updating unit, configured to update the format specification in the same category specification set according to the first processing result and the second processing result, so as to increase the number of specification items and the specification coverage of the category format specification;

[0130] The determination unit is used to determine the updated specification set as the target set of format specifications of the category, and determine the database constructed by all target sets as the file specification library.

[0131] In an optional embodiment, the processing module includes:

[0132] The second acquisition submodule is used to calculate the information entropy of each bidding specification in the file specification library according to the bidding document scoring standard to obtain the information entropy of each bidding specification;

[0133] The second determination submodule is used to determine the specification weight of each bidding specification according to the information entropy of each bidding specification.

[0134] In an optional embodiment, the file detection model includes a CNN model and an RNN model; the training acquisition module includes:

[0135] A construction submodule, used for constructing an image training set, an image verification set, a text training set, and a text verification set according to the bidding specifications of all documents and the corresponding specification weights;

[0136] A first training and verification submodule is used to train and verify the CNN model according to the image training set and the image verification set, respectively, until the CNN model reaches a first preset condition for training;

[0137] A second training and verification submodule is used to train and verify the RNN model according to the text training set and the text verification set, respectively, until the RNN model meets the second preset condition of training;

[0138] The third determination submodule is used to determine the detection model constructed by the CNN model that meets the first preset condition and the RNN model that meets the second preset condition as the target detection model.

[0139] In an optional embodiment, the target detection model includes a trained CNN model and an RNN model; the detection module includes:

[0140] A division submodule is used to divide the target file into image features and text features to be detected;

[0141] The third acquisition submodule is used to input the image features and the text features into the trained CNN model and the RNN model respectively to obtain the format detection result of the target file;

[0142] The fourth acquisition submodule is used to perform confidence evaluation on the format detection result according to the Bayesian model to obtain an evaluation result;

[0143] A processing submodule, used for updating the model parameters of the target detection model and re-detecting when the evaluation result does not reach the target confidence level, until the evaluation result reaches the target confidence level;

[0144] The output submodule is used to output the standard detection result of the target file according to the format detection result when the evaluation result reaches the target confidence level.

[0145] In an optional embodiment, the output submodule includes:

[0146] The input acquisition unit is used to input the format detection result into the language processing model, and obtain the standard detection result according to the detection report output by the language processing model, wherein the detection report includes the abnormal format location, abnormal format type, severity of the abnormal format and adjustment suggestions represented by the format detection result.

[0147] Based on the same technical concept as the detection method, an embodiment of the present invention also provides an electronic device, including a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of any one of the detection methods.

[0148] Based on the same technical concept as the detection method, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the detection methods are implemented.

[0149] The technical solution provided in the embodiments of the present invention has at least the following technical effects or advantages:

[0150] By constructing a file specification library for bidding documents, since the file specification library is a database composed of bidding specifications of different documents in the bid documents, the importance of each bidding specification in the file specification library can be quantified to obtain the specification weights of all bidding specifications; the preset file detection model is trained according to the bidding specifications of all documents and the corresponding specification weights to obtain the target detection model; the target file of the bid document to be bid is input into the target detection model, and the specification detection result of the target file is determined according to the output result of the target detection model. This technical solution implements weight configuration for different bidding specifications, which can enable the target detection model to highlight the detection of key specifications when implementing the detection of target files, reduce the detection deviation and redundancy of all bidding specifications under the same detection standard, and thus take into account the efficiency and accuracy of bidding document detection.

[0151] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (modules, systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0156] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A bidding document detection method, characterized in that: The method comprises: Constructing a document specification library of the bidding documents, wherein the document specification library is a database composed of bidding specifications of different documents in the bidding documents; Quantifying the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications; The preset document detection model is trained according to the bidding specifications of all documents and the corresponding specification weights to obtain the target detection model; The target file of the bid to be submitted is input into the target detection model, and the standard detection result of the target file is determined according to the output result of the target detection model.

2. The tender document detection method according to claim 1, characterized in that: The importance of each bidding specification in the document specification library is quantified to obtain the specification weight of all bidding specifications, including: Calculating the information entropy of each bidding specification in the document specification library according to the bidding document scoring standard to obtain the information entropy of each bidding specification; According to the information entropy of each bidding specification, a specification weight of each bidding specification is determined.

3. The tender document detection method according to claim 1, characterized in that: The file specification library for constructing the bid document includes: Extracting format features of the tendered documents to obtain format specifications of the tendered documents; Boolean operation processing is performed according to the similarity between each format specification in all format specifications, and a database constructed by the processing results is determined as the file specification library.

4. The tender document detection method according to claim 3, characterized in that: The method of performing Boolean operation processing according to the similarity between the format specifications in all the format specifications and determining the database constructed by the processing results as the file specification library includes: Divide the specification categories according to the similarities between all format specifications to obtain multiple specification sets; Performing intersection operation processing and union operation processing on the standard sets of the same category to obtain a first processing result of the intersection operation processing and a second processing result of the union operation processing; updating the format specifications in the same category specification set according to the first processing result and the second processing result, so as to increase the number of specification items and the specification coverage of the format specifications of the category; The updated specification set is determined as the target set of the format specification of the category, and the database constructed by all the target sets is determined as the file specification library.

5. The tender document detection method according to claim 1, characterized in that: The file detection model includes a CNN model and an RNN model; the preset file detection model is trained according to the bidding specifications of all files and the corresponding specification weights to obtain a target detection model, including: According to the bidding specifications of all documents and the corresponding specification weights, an image training set, an image verification set, a text training set and a text verification set are constructed; The CNN model is trained and verified according to the image training set and the image verification set respectively until the CNN model reaches a first preset condition for training; The RNN model is trained and verified according to the text training set and the text verification set respectively until the RNN model meets a second preset condition for training; The detection model constructed by the CNN model that meets the first preset condition and the RNN model that meets the second preset condition is determined as the target detection model.

6. The tender document detection method according to claim 1, characterized in that: The target detection model includes a trained CNN model and an RNN model; the target file of the bid to be submitted is input into the target detection model, and the standard detection result of the target file is determined according to the output result of the target detection model, including: Dividing the target file into image features and text features to be detected; Inputting the image features and the text features into the trained CNN model and RNN model respectively to obtain the format detection result of the target file; Performing confidence evaluation on the format detection result according to the Bayesian model to obtain an evaluation result; When the evaluation result does not reach the target confidence level, updating the model parameters of the target detection model and re-detecting until the evaluation result reaches the target confidence level; When the evaluation result reaches the target confidence level, the standard detection result of the target file is output according to the format detection result.

7. The tender document detection method according to claim 6, characterized in that: Outputting the standard detection result of the target file according to the format detection result includes: The format detection result is input into a language processing model, and the standard detection result is obtained according to a detection report output by the language processing model, wherein the detection report includes the abnormal format location, abnormal format type, abnormal format severity and adjustment suggestions represented by the format detection result.

8. A tender document detection device, characterized in that: The device comprises: A construction module, used to construct a file specification library of the bidding documents, wherein the file specification library is a database composed of bidding specifications of different documents in the bidding documents; A processing module, used for quantifying the importance of each bidding specification in the document specification library to obtain the specification weight of all bidding specifications; A training acquisition module is used to train a preset file detection model according to the bidding specifications of all files and corresponding specification weights to obtain a target detection model; The detection module is used to input the target file of the bid to be submitted into the target detection model, and determine the standard detection result of the target file according to the output result of the target detection model.

9. An electronic device, characterized in that: The electronic device comprises a processor and a memory, wherein the memory is coupled to the processor and stores instructions, and when the instructions are executed by the processor, the electronic device executes the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images