CN118052627A - Intelligent filling method and system for bidding scheme - Google Patents
Intelligent filling method and system for bidding scheme Download PDFInfo
- Publication number
- CN118052627A CN118052627A CN202410448428.6A CN202410448428A CN118052627A CN 118052627 A CN118052627 A CN 118052627A CN 202410448428 A CN202410448428 A CN 202410448428A CN 118052627 A CN118052627 A CN 118052627A
- Authority
- CN
- China
- Prior art keywords
- bidding
- text
- scheme
- error correction
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 94
- 238000012937 correction Methods 0.000 claims abstract description 113
- 238000001514 detection method Methods 0.000 claims abstract description 24
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 8
- 238000003058 natural language processing Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 39
- 230000008569 process Effects 0.000 claims description 34
- 230000006870 function Effects 0.000 claims description 28
- 238000012549 training Methods 0.000 claims description 25
- 238000013145 classification model Methods 0.000 claims description 21
- 238000012360 testing method Methods 0.000 claims description 21
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000013480 data collection Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000013527 convolutional neural network Methods 0.000 abstract description 4
- 230000007246 mechanism Effects 0.000 description 11
- 238000004458 analytical method Methods 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005429 filling process Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an intelligent filling method and system for a bidding scheme. The invention belongs to the technical field of intelligent filling of bidding schemes, in particular to an intelligent filling method and an intelligent filling system of bidding schemes, wherein the scheme classifies the past bidding schemes by adopting a text classification optimization algorithm to obtain different types of bidding schemes, extracts templates for different types of bidding schemes by utilizing a natural language processing technology, reduces the workload of manually designing templates and improves the efficiency; and intelligent filling is carried out on the selected templates by using BERT and convolutional neural network, so that the correlation and accuracy of the filling content are ensured, and the error detection and correction are carried out on the filling content by combining with a text error correction network constructed by a bidirectional GRU model, thereby greatly improving the quality and the specialty of the scheme.
Description
Technical Field
The invention relates to the technical field of intelligent filling of bidding schemes, in particular to an intelligent filling method and system of bidding schemes.
Background
The bidding scheme preparation plays a crucial role in project purchasing and management, in the traditional bidding scheme preparation process, a professional or team usually designs and lays out a bidding scheme template according to the specific requirements and industry standards of the bidding project, and a great deal of time is required to manually create and adjust the template, so that the efficiency is low, errors and omission can be caused in the template due to human factors, in addition, the template is time-consuming and labor-consuming, errors are prone to occur, and the efficiency and accuracy are difficult to guarantee especially when a great deal of data and complex formats are processed.
Disclosure of Invention
Aiming at the situation, in order to overcome the defects of the prior art, the invention provides an intelligent filling method and system for a bidding scheme, and aims at designing and laying out a bidding scheme template according to the specific requirements and industry standards of bidding projects by professionals or teams in the traditional bidding scheme compiling process, and a great deal of time is required to manually create and adjust the template, so that the efficiency is low, the problem of errors and omission in the template possibly caused by human factors is solved, the traditional bidding scheme is classified by adopting a text classification optimization algorithm to obtain different types of bidding schemes, the template is extracted for different types of bidding schemes by utilizing a natural language processing technology, the workload of manually designing the template is reduced, and the efficiency is improved; aiming at the problems that manual template filling is time-consuming and labor-consuming and is easy to make mistakes, and particularly when a large amount of data and complex formats are processed, efficiency and accuracy are difficult to guarantee, the scheme utilizes the BERT and the convolutional neural network to intelligently fill the selected templates, ensures the relevance and accuracy of filling contents, combines a text error correction network constructed by a bidirectional GRU model to perform error detection and correction on the filling contents, and greatly improves the quality and the specialty of the scheme.
The technical scheme adopted by the invention is as follows: the invention provides an intelligent filling method of a bidding scheme, which comprises the following steps:
step S1: the method comprises the steps of sorting a bidding scheme, collecting a historical bidding scheme data set, wherein the historical bidding scheme data set comprises bidding scheme texts and labels, preprocessing the bidding scheme texts, obtaining an optimal feature subset through a feature selection algorithm, and sorting the historical bidding scheme by using a sorting model to obtain different types of bidding schemes;
Step S2: extracting a bidding scheme template, analyzing the structure of the bidding scheme by using a natural language processing algorithm according to different types of bidding schemes, extracting format features of each part, and generating different bidding templates;
Step S3: the method comprises the steps of intelligently filling a template, determining a bid-bidding scheme template according to existing data and requirements, using a BERT model to encode a bid-bidding scheme text, extracting global features, capturing local features in the bid-bidding scheme text through a local feature convolution network model, fusing the local features and the global features, generating final feature representation of the bid-bidding scheme, and intelligently filling each part in the template according to the final feature representation;
Step S4: filling content error correction, constructing a text error correction network, and correcting text errors in the filling content by using the text error correction network.
Further, in step S1, the bidding scheme classification specifically includes the following steps:
Step S11: the method comprises the steps of data collection and preprocessing, collecting a historical bidding scheme data set, wherein the historical bidding scheme data set comprises a bidding scheme text and a label, the label is a bidding scheme type, preprocessing is carried out on the bidding scheme text, including removal of useless characters and incomplete text data, a preprocessed bidding scheme data set is obtained, and the preprocessed bidding scheme data set is divided into a training set A and a testing set A according to the ratio of 7:3;
step S12: extracting features, namely extracting important named entities in the bidding scheme text by using a named entity recognition algorithm, and using a TF-IDF matrix of the identified important named entities as extracted features;
Step S13: initializing a feature selection algorithm, and determining an initial feature set by using a maximum correlation minimum redundancy algorithm;
Step S14: iterative optimization, defining a preset threshold, exploring a feature subset space by changing specific elements of an initial feature set in different iterations, evaluating the quality of each feature subset by using an fitness function, reserving a better feature subset in each iteration, terminating an algorithm when the performance improvement of the optimal feature subset in successive iterations is smaller than the preset threshold, and outputting a current optimal feature subset as a final result;
Step S15: training a classification model, training the classification model by using the current optimal feature subset, setting super parameters of the classification model, and comprising: learning rate and batch size, stopping training when the error of the test set A is no longer reduced in a plurality of continuous iteration times of the classification model, and adjusting the super parameters of the classification model according to the performance of the classification model on the test set A to obtain a classification model B;
step S16: and (3) model application, namely using the classification model B for the preprocessed bidding scheme data set to classify so as to obtain different types of bidding scheme sets.
Further, in step S2, the bidding scheme template is extracted, specifically including the following steps:
Step S21: template structure analysis, namely analyzing bidding schemes in different types of bidding scheme sets, identifying structural information of the bidding schemes, such as titles, introduction, technical requirements, contract terms and the like by using a regular expression or pattern matching technology, and identifying named entities in the bidding schemes, such as organization names, places, dates, currency amounts and the like by using an entity identification technology to obtain analysis results;
Step S22: generating a template, designing a template structure of each type according to an analysis result, including titles, chapters, sections and other format requirements, determining typesetting rules of each part, such as fonts, sizes, alignment modes and the like, and simultaneously reserving proper custom space to adapt to the requirements of specific projects;
Step S23: and (3) verifying and optimizing the template, testing the generated template by using a small part of bidding schemes, ensuring that the template can adapt to the type of bidding schemes, and adjusting and optimizing the template according to a test result and user feedback.
Further, in step S3, the intelligent filling template specifically includes the following steps:
Step S31: text encoding is carried out through a text encoding layer, global features are extracted, a BERT model is used for encoding a bidding scheme text, sentence-level feature vectors are extracted, and contents of different titles of the bidding scheme are defined as a single token sequence Wherein, the method comprises the steps of, wherein,A classification mark is indicated and is displayed,A segment paragraph marker is represented,AndRepresenting the first text of the corresponding textA number of marks;
Define the input as WhereinIs the firstThe individual tokens are embedded by corresponding tokens, segments and locations embedded in a summing structure,For the length of the maximum input sequence, the content embedding of different titles of the bidding scheme is calculated, and the steps are as follows:
step S311: the input of the multi-head self-attention mechanism is calculated using the following formula:
；
In the method, in the process of the invention, Representing the input of a multi-headed self-attention mechanism,Represent the firstContent embedding of different titles of the layer bidding scheme;
step S312: the output of the multi-headed self-attention mechanism is calculated using the following formula:
；
In the method, in the process of the invention, Representing the output of a multi-headed self-attention mechanism,Representing a matrix of parameters that can be learned,The dimensions of the dimensions are represented and,Representing the number of heads of the multi-head attention mechanism,The activation function is represented as a function of the activation,Representing a transpose of the matrix;
Step S313: the outputs of all heads in the multi-head self-attention mechanism are integrated, and the formula is as follows:
；
In the method, in the process of the invention, Representing the output of all heads in the integrated multi-head self-attention mechanism,Representing a matrix of parameters that can be learned,Representing merging all heads in a multi-head self-attention mechanism;
Step S314: the outputs of all heads in the multi-head self-attention mechanism are added to the inputs of the multi-head self-attention mechanism and layer normalized using the following formula:
；
；
In the method, in the process of the invention, Representing the output result of adding the output of the multi-headed self-attention mechanism to the original input and performing layer normalization,The representation layer is normalized and,Representing the input of the firstThe data of the individual samples are taken,AndThe scaling and translation parameters are represented respectively,AndRepresenting the mean and variance of the feature vector of each sample data separately,Representing a decimal fraction;
Step S315: the output of the local feature convolution network is calculated using the following formula:
；
In the method, in the process of the invention, Representing the output of the local feature convolution network,The activation function is represented as a function of the activation,，，，Representing a matrix of learnable parameters;
the global feature vector is calculated using the following formula:
；
In the method, in the process of the invention, Representing the global feature vector to obtain an output feature vector of the text coding layer;
Step S32: capturing local features in the bidding scheme text through the local feature convolution layer, and feeding output feature vectors of a first layer of the text coding layer to the local feature convolution layer, wherein the steps are as follows:
Step S321: the output feature vector of the first layer of the text encoding layer is expressed as a sequence vector, and the following formula is used:
；
In the method, in the process of the invention, The connection operator is represented by a connection symbol,Is the first in the input sequence of the corresponding text coding layerThe input of the individual tags is embedded,Representing the sequence length; Representing a sequence vector;
Step S322: features are generated using one-dimensional convolution operations using the following formulas:
；
In the method, in the process of the invention, As a function of the non-linearity,As a result of the bias term,The representation of the filter is made of,The representation is from the firstThe first token toThe number of tokens to be used in the process of the present invention,Representing the generated features;
step S323: the feature map is generated using a filter using the following formula:
；
In the method, in the process of the invention, Representing a feature map;
step S324: processing the feature map by using a maximum pooling operation, applying maximum pooling to the feature map to obtain a maximum value of the feature map, and taking the maximum value of the feature map as a corresponding feature of a specific filter, wherein each filter generates a significant feature, and connecting the significant features to generate an advanced feature vector, wherein the formula is as follows:
；
In the method, in the process of the invention, Representing a high-level feature vector of the model,Representing the number of filters;
step S33: feature fusion, namely fusing local features and global features, generating final feature representation of a bidding scheme, and adding and averaging global feature vectors to obtain new global feature vectors, wherein the formula is as follows:
；
In the method, in the process of the invention, A new global feature vector is represented and,Indicating the number of layers of the network;
and summing and averaging the advanced feature vector and the new global feature vector again to obtain a fusion feature vector, wherein the following formula is adopted:
；
In the method, in the process of the invention, Representing a fused feature vector;
Step S34: and filling the content of the bidding scheme, and intelligently filling each part in the template by utilizing the extracted fusion feature vector according to the existing data and the requirements.
Further, in step S4, the padding content error correction specifically includes the steps of:
Step S41: the data acquisition, the Chinese text error correction data set is acquired, the Chinese text error correction data set is preprocessed, and the preprocessed Chinese text error correction data set is obtained and divided into a training set B and a testing set B;
Step S42: the text error correction network is constructed, the text error correction network comprises a detection network and an error correction network, the error correction network is composed of two full-connection layers, the detection network utilizes a bidirectional GRU model to carry out forward and reverse coding on a text sequence to obtain a forward coding hidden state and a reverse coding hidden state, the last hidden layer is utilized to combine the forward coding hidden state and the reverse coding hidden state to obtain a combined hidden state, the combined hidden state is sent into the error correction network, and the calculation process is as follows:
；
；
；
；
In the method, in the process of the invention, Representing the input of a bi-directional GRU model,Word embedding representing each character in a text sequence,The location of the representation is embedded,Representation segment embedding; Indicating the hidden state of the forward direction encoding, The hidden state after the combination is represented,Indicating the hidden state of the reverse coding,Representing the number of the gate-controlled circulation units,Indicating the previous hidden state of the forward encoding,Representing the latter hidden state of the reverse coding;
step S43: defining a loss function, wherein the loss function of the text error correction network consists of a loss function of the detection network and a loss function of the error correction network, and the formula is as follows:
；
；
；
In the method, in the process of the invention, Representing the cross entropy loss of the error correction network,Indicating that the network cross entropy loss is detected,AndRepresenting the linear combination coefficients of the loss function of the detection network and the loss function of the error correction network respectively,The input data is represented by a representation of the input data,Representing a probability distribution of the detected network output,Representing the probability distribution of the error correction network output,Representing the output of the error correction network,Representing an output of the detection network;
Step S44: model training and evaluation, namely inputting a training set B into a text error correction network to train, obtaining a text error correction network A after training is completed, evaluating the performance of the text error correction network A on a test set B, stopping training when the error of the test set B is not reduced in a plurality of continuous iteration times of the text error correction network A, and adjusting the hyper-parameters of the text error correction network A according to the performance of the text error correction network A on the test set B to obtain the text error correction network B;
step S45: error correction, text error correction network B is used for text error detection of the filled bidding scheme, and once an error is detected, text error correction network B will attempt to generate the correct text or provide candidate correction options.
The invention provides an intelligent filling system for a bid-in scheme, which comprises a bid-in scheme classification module, a bid-in scheme template extraction module, an intelligent filling template module and a filling content error correction module;
The bid-drawing scheme classification module collects a historical bid-drawing scheme data set, preprocesses the historical bid-drawing scheme, obtains an optimal feature subset through a feature selection algorithm, classifies the historical bid-drawing scheme by using a classification model to obtain different types of bid-drawing schemes, and sends the different types of bid-drawing schemes to the bid-drawing scheme template module;
The method comprises the steps that an extraction bidding scheme template module receives different types of bidding schemes sent by a bidding scheme classification module, analyzes the structure of the bidding scheme by utilizing a natural language processing algorithm according to the different types of bidding schemes, extracts the format characteristics of each part, generates different bidding templates, and sends the different bidding templates to an intelligent filling template module;
The intelligent filling template module receives different bidding templates sent by extracting bidding scheme templates, determines bidding scheme templates according to existing data and requirements, encodes bidding scheme texts by using a BERT model, extracts global features, captures local features in the bidding scheme texts by using a local feature convolution network model, fuses the local features and the global features, generates final feature representation of the bidding scheme, intelligently fills each part in the templates according to the final feature representation, obtains filled bidding templates, and sends the filled bidding templates to the filling content error correction module;
the filling content error correction module receives the filled bid template sent by the intelligent filling template module, a text error correction network is constructed, and text errors in the filling content are corrected by the text error correction network.
By adopting the scheme, the beneficial effects obtained by the invention are as follows:
(1) Aiming at the problems that in the traditional bidding scheme compiling process, a professional or team designs and lays out the bidding scheme template according to the specific requirements and industry standards of bidding projects, and a great deal of time is required to manually create and adjust the template, the efficiency is low, errors and omission in the template possibly exist due to human factors.
(2) Aiming at the problems that manual template filling is time-consuming and labor-consuming and is easy to make mistakes, and particularly when a large amount of data and complex formats are processed, efficiency and accuracy are difficult to guarantee, the scheme utilizes the BERT and the convolutional neural network model to intelligently fill the selected templates, ensures the relevance and accuracy of filling contents, combines a text error correction network constructed by a bidirectional GRU model to perform error detection and correction on the filling contents, and greatly improves the quality and the specialty of the scheme.
Drawings
FIG. 1 is a schematic flow chart of an intelligent filling method of a bidding scheme;
FIG. 2 is a schematic diagram of an intelligent filling system for a bidding scheme provided by the invention;
FIG. 3is a flow chart of step S1;
fig. 4 is a flow chart of step S4;
Fig. 5 is a schematic diagram of a user interface.
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
First embodiment, referring to fig. 1, the present invention provides an intelligent filling method for a bidding scheme, which includes the following steps:
step S1: the method comprises the steps of sorting a bidding scheme, collecting a historical bidding scheme data set, wherein the historical bidding scheme data set comprises bidding scheme texts and labels, preprocessing the bidding scheme texts, obtaining an optimal feature subset through a feature selection algorithm, and sorting the historical bidding scheme by using a sorting model to obtain different types of bidding schemes;
Step S2: extracting a bidding scheme template, analyzing the structure of the bidding scheme by using a natural language processing algorithm according to different types of bidding schemes, extracting format features of each part, and generating different bidding templates;
Step S3: the method comprises the steps of intelligently filling a template, determining a bid-bidding scheme template according to existing data and requirements, using a BERT model to encode a bid-bidding scheme text, extracting global features, capturing local features in the bid-bidding scheme text through a local feature convolution network model, fusing the local features and the global features, generating final feature representation of the bid-bidding scheme, and intelligently filling each part in the template according to the final feature representation;
Step S4: filling content error correction, constructing a text error correction network, and correcting text errors in the filling content by using the text error correction network.
Referring to fig. 1 and 3, in the second embodiment, based on the above embodiment, in step S11, data collection and preprocessing are performed to collect a historical bidding scheme data set, where the historical bidding scheme data set includes bidding scheme text and a label, the label is a bidding scheme type, and the bidding scheme type includes construction engineering, information technology, service, material purchase and medical health, preprocessing the bidding scheme text, including removing useless characters and incomplete text data, to obtain a preprocessed bidding scheme data set, and dividing the preprocessed bidding scheme data set into a training set a and a test set a according to a ratio of 7:3;
in step S13, the feature selection algorithm is initialized, and the initial feature set is determined using the maximum correlation minimum redundancy algorithm, specifically, referring to fig. 1 and 3, based on the above embodiment: each feature is composed of Binary vector representation of dimensions, where 1 represents feature selected and 0 represents feature unselected, to determine a good starting point, use maximum correlation minimum redundancy algorithm to select features, and based on the initial dimension of the TF-IDF matrix, by plottingThe cumulative sum of the selected features determines the optimal number of features,The values are 1000, 1500 and 2000, the first 1000 features are determined as starting points of the algorithms through analysis of the maximum correlation minimum redundancy algorithm, three experiments are performed for optimizing experimental results, and each time, different algorithms are used for initialization: maximum correlation minimum redundancy algorithm,Algorithm and methodAnd finally, determining an initial feature set by comparing the results of the three experiments and selecting a maximum correlation minimum redundancy algorithm.
Fourth embodiment, referring to fig. 1 and 3, the embodiment is based on the above embodiment, and in step S14, iterative optimization is specifically: defining a preset thresholdTo generate other possible feature subsets around the initial feature setEach feature subset hasEach element, algorithm uses two parametersAnd，Representing the maximum number of features to be changed in each iterationRepresenting the range of elements in the initial feature set that are to be considered for change, in a first iteration, the algorithm randomly changes the initial solution beforeOf individual elementsElements to generate a new solution; in the second iteration, after the changeOf individual elementsAn element; in the first placeIn the next iteration, change the firstOf individual elementsIn this manner, the algorithm explores different parts of the solution space in different iterations to find a more optimal solution, uses fitness functions to evaluate the quality of feature subsets, and retains the more optimal feature subsets in each iteration when the performance improvement of the optimal feature subsets in successive iterations is less than a preset thresholdWhen the algorithm is terminated, the current optimal feature subset is output as a final result, and the parameters are calculatedIs set to be 2000 a and is set to be,Is set to be 8 and is set to be a constant,Is set to be 12 with the number of the parts,Set to 0.0001.
In step S15, the classification model is trained, and the classification performance of the model is estimated by five-fold cross-validation method assuming that the classification model is an SVM, and the algorithm aims at finding a feature subset capable of maximizing the classification accuracy of the SVM model by optimization in an iterative process, and in each iteration, the algorithm selects a solution better than the initial solution to optimize the feature selection, referring to fig. 1 and 3.
Through executing the operation, in the process of compiling the traditional bidding scheme, professional staff or team usually designs and lays out the bidding scheme template according to the specific requirements and industry standards of the bidding project, and a great deal of time is required to manually create and adjust the template, so that the efficiency is low, the problem of errors and omission in the template possibly caused by human factors is solved.
In a sixth embodiment, referring to fig. 1, the embodiment is based on the above embodiment, and in step S3, the intelligent filling template is implemented in the following experimental environment: using ADAMWEIGHTDECAY optimizers in the BERT model, the maximum input sequence length was set to 256, the batch size was set to 32, the training epoch number was set to 4, and the initial learning rate was set to。
Embodiment seven, referring to fig. 1 and 4, based on the above embodiment, in step S4, the padding content error correction specifically includes the following steps:
Step S41: the data acquisition, the Chinese text error correction data set is acquired, the Chinese text error correction data set is preprocessed, and the preprocessed Chinese text error correction data set is obtained and divided into a training set B and a testing set B;
Step S42: the text error correction network is constructed, the text error correction network comprises a detection network and an error correction network, the error correction network is composed of two full-connection layers, the detection network utilizes a bidirectional GRU model to carry out forward and reverse coding on a text sequence to obtain a forward coding hidden state and a reverse coding hidden state, the last hidden layer is utilized to combine the forward coding hidden state and the reverse coding hidden state to obtain a combined hidden state, the combined hidden state is sent into the error correction network, and the calculation process is as follows:
；
；
；
；
In the method, in the process of the invention, Representing the input of a bi-directional GRU model,Word embedding representing each character in a text sequence,The location of the representation is embedded,Representation segment embedding; Indicating the hidden state of the forward direction encoding, The hidden state after the combination is represented,Indicating the hidden state of the reverse coding,Representing the number of the gate-controlled circulation units,Indicating the previous hidden state of the forward encoding,Representing the latter hidden state of the reverse coding;
step S43: defining a loss function, wherein the loss function of the text error correction network consists of a loss function of the detection network and a loss function of the error correction network, and the formula is as follows:
；
；
；
In the method, in the process of the invention, Representing the cross entropy loss of the error correction network,Indicating that the network cross entropy loss is detected,AndRepresenting the linear combination coefficients of the loss function of the detection network and the loss function of the error correction network respectively,The input data is represented by a representation of the input data,Representing a probability distribution of the detected network output,Representing the probability distribution of the error correction network output,Representing the output of the error correction network,Representing an output of the detection network;
Step S44: model training and evaluation, namely inputting a training set B into a text error correction network to train, obtaining a text error correction network A after training is completed, evaluating the performance of the text error correction network A on a test set B, stopping training when the error of the test set B is not reduced in a plurality of continuous iteration times of the text error correction network A, and adjusting the hyper-parameters of the text error correction network A according to the performance of the text error correction network A on the test set B to obtain the text error correction network B;
step S45: error correction, text error correction network B is used for text error detection of the filled bidding scheme, and once an error is detected, text error correction network B will attempt to generate the correct text or provide candidate correction options.
By executing the above operation, aiming at the problems that manual template filling is time-consuming and labor-consuming and is easy to make mistakes, and particularly when a large amount of data and complex formats are processed, efficiency and accuracy are difficult to guarantee, the scheme utilizes the BERT and the convolutional neural network model to intelligently fill the selected templates, ensures the relevance and accuracy of filling contents, combines a text error correction network constructed by a bidirectional GRU model to perform error detection and correction on the filling contents, and greatly improves the quality and the specialty of the scheme.
An eighth embodiment, referring to fig. 1 and 5, is based on the above embodiment, and relates to a computer-implemented user interface for an intelligent filling bidding scheme, including file upload, template selection, intelligent filling and filling error correction processes, and the specific interface includes the following:
1) File upload area: providing an obvious button or drag-and-drop instruction, guiding a user to upload a file, displaying a file uploading progress bar, enabling the user to know the uploading state in real time, displaying a prompt message of successful uploading after successful uploading, and displaying a corresponding error message when an error occurs;
2) Template selection area: the available template previews are presented in a list or grid form, the template screening and searching functions are provided, a user can conveniently find a suitable template quickly, after selecting the template, the preview is displayed, and a confirmation button is arranged for the next operation;
3) Intelligent filling area: configuring necessary input fields including project names and dates so as to collect relevant information before intelligent filling, triggering an intelligent filling process by a 'start filling' button, displaying a progress bar or a state indication in the filling process, enabling a user to know the operation progress, and viewing the intelligently filled document in a document preview area;
4) Error correction region: by tabulating errors detected in the document, an editable document view or text editor is integrated, allowing the user to correct errors directly, providing functional buttons for saving and rechecking, ensuring the accuracy of the document.
An embodiment nine, referring to fig. 2, based on the foregoing embodiment, the present invention provides an intelligent filling system for a bidding scheme, including a bidding scheme classification module, an bidding scheme extraction template module, an intelligent filling template module, and a filling content error correction module;
The bid-drawing scheme classification module collects a historical bid-drawing scheme data set, preprocesses the historical bid-drawing scheme, obtains an optimal feature subset through a feature selection algorithm, classifies the historical bid-drawing scheme by using a classification model to obtain different types of bid-drawing schemes, and sends the different types of bid-drawing schemes to the bid-drawing scheme template module;
The method comprises the steps that an extraction bidding scheme template module receives different types of bidding schemes sent by a bidding scheme classification module, analyzes the structure of the bidding scheme by utilizing a natural language processing algorithm according to the different types of bidding schemes, extracts the format characteristics of each part, generates different bidding templates, and sends the different bidding templates to an intelligent filling template module;
The intelligent filling template module receives different bidding templates sent by extracting bidding scheme templates, determines bidding scheme templates according to existing data and requirements, encodes bidding scheme texts by using a BERT model, extracts global features, captures local features in the bidding scheme texts by using a local feature convolution network model, fuses the local features and the global features, generates final feature representation of the bidding scheme, intelligently fills each part in the templates according to the final feature representation, obtains filled bidding templates, and sends the filled bidding templates to the filling content error correction module;
the filling content error correction module receives the filled bid template sent by the intelligent filling template module, a text error correction network is constructed, and text errors in the filling content are corrected by the text error correction network.
Embodiment ten, referring to fig. 1, based on the above embodiment, in step S21, the template structure is analyzed to analyze bidding schemes in different types of bidding scheme sets, the structured information of the bidding schemes is identified using regular expressions, including title, introduction, technical requirement, and contract terms, and the named entities in the bidding schemes are identified using entity identification technology, including organization name, place, date, and monetary amount, to obtain the analysis result.
An eleventh embodiment, referring to fig. 1, is based on the above embodiment, and the present embodiment is different from the above embodiment only in terms of the technique for identifying the structured information of the bidding scheme, and uses a pattern matching technique to identify the structured information of the bidding scheme.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.
Claims (6)
1. An intelligent filling method for a bidding scheme is characterized by comprising the following steps of: the method comprises the following steps:
step S1: the method comprises the steps of sorting a bidding scheme, collecting a historical bidding scheme data set, wherein the historical bidding scheme data set comprises bidding scheme texts and labels, preprocessing the bidding scheme texts, obtaining an optimal feature subset through a feature selection algorithm, and sorting the historical bidding scheme by using a sorting model to obtain different types of bidding schemes;
Step S2: extracting a bidding scheme template, analyzing the structure of the bidding scheme by using a natural language processing algorithm according to different types of bidding schemes, extracting format features of each part, and generating different bidding templates;
Step S3: the method comprises the steps of intelligently filling a template, determining a bid-bidding scheme template according to existing data and requirements, using a BERT model to encode a bid-bidding scheme text, extracting global features, capturing local features in the bid-bidding scheme text through a local feature convolution network model, fusing the local features and the global features, generating final feature representation of the bid-bidding scheme, and intelligently filling each part in the template according to the final feature representation;
Step S4: filling content error correction, constructing a text error correction network, and correcting text errors in the filling content by using the text error correction network.
2. The bidding scheme intelligent filling method according to claim 1, wherein the method is characterized in that: in step S1, the bidding scheme classification includes the following steps:
Step S11: the method comprises the steps of data collection and preprocessing, collecting a historical bidding scheme data set, wherein the historical bidding scheme data set comprises a bidding scheme text and a label, the label is a bidding scheme type, preprocessing is carried out on the bidding scheme text to obtain a preprocessed bidding scheme data set, and the preprocessed bidding scheme data set is divided into a training set A and a testing set A;
step S12: extracting features, namely extracting important named entities in the bidding scheme text by using a named entity recognition algorithm, and using a TF-IDF matrix of the identified important named entities as extracted features;
Step S13: initializing a feature selection algorithm, and determining an initial feature set by using a maximum correlation minimum redundancy algorithm;
Step S14: iterative optimization, defining a preset threshold, exploring a feature subset space by changing specific elements of an initial feature set in different iterations, evaluating the quality of each feature subset by using an fitness function, reserving a better feature subset in each iteration, terminating an algorithm when the performance improvement of the optimal feature subset in successive iterations is smaller than the preset threshold, and outputting a current optimal feature subset as a final result;
Step S15: training a classification model, namely training the classification model by using the current optimal feature subset, setting the super parameters of the classification model, stopping training when the error of the test set A is not reduced in a plurality of continuous iteration times of the classification model, and adjusting the super parameters of the classification model according to the performance of the classification model on the test set A to obtain a classification model B;
step S16: and (3) model application, namely using the classification model B for the preprocessed bidding scheme data set to classify so as to obtain different types of bidding scheme sets.
3. The bidding scheme intelligent filling method according to claim 2, characterized by comprising the following steps: in step S3, the intelligent filling template includes the following steps:
Step S31: text encoding is carried out through a text encoding layer, global features are extracted, global feature vectors are obtained, a BERT model is used for encoding the text of the bidding scheme, content embedding of different titles of the bidding scheme is calculated, and sentence-level feature vectors are extracted;
step S32: capturing local features in the bidding scheme text through a local feature convolution layer, and feeding output feature vectors of a first layer of a text coding layer to the local feature convolution layer;
step S33: feature fusion, namely fusing local features and global features, generating final feature representation of a bidding scheme, and adding and averaging global feature vectors to obtain new global feature vectors, wherein the formula is as follows:
；
In the method, in the process of the invention, Representing a new global feature vector,/>Representing the number of layers of the BERT model,/>Representing a global feature vector;
and summing and averaging again by using the advanced feature vector and the new global feature vector to obtain a fusion feature vector, wherein the following formula is adopted:
；
In the method, in the process of the invention, Representing fusion feature vectors,/>Representing a high-level feature vector;
Step S34: and filling the content of the bidding scheme, and intelligently filling each part in the template by utilizing the extracted fusion feature vector according to the existing data and the requirements.
4. A bidding scheme intelligent filling method according to claim 3, characterized in that: in step S32, the capturing of the local feature in the text by the local feature convolution layer, and feeding the output feature vector of the first layer of the text encoding layer to the local feature convolution layer, includes the steps of:
Step S321: the output feature vector of the first layer of the text encoding layer is expressed as a sequence vector, and the following formula is used:
；
In the method, in the process of the invention, Representing a join operator,/>Is the input sequence of the corresponding text coding layerInput embedding of individual tags,/>Representing the sequence length; /(I)Representing a sequence vector;
Step S322: features are generated using one-dimensional convolution operations using the following formulas:
；
In the method, in the process of the invention, As a nonlinear function,/>Is an offset term,/>Representing a filter,/>The representation is from the/>The first token toPersonal token,/>Representing the generated features;
step S323: the feature map is generated using a filter using the following formula:
；
In the method, in the process of the invention, Representing a feature map;
step S324: processing the feature map by using a maximum pooling operation, applying maximum pooling to the feature map to obtain a maximum value of the feature map, and taking the maximum value of the feature map as a corresponding feature of a specific filter, wherein each filter generates a significant feature, and connecting the significant features to generate an advanced feature vector, wherein the formula is as follows:
；
In the method, in the process of the invention, Representing advanced feature vectors,/>Representing the number of filters.
5. The intelligent filling method for the bidding scheme according to claim 4, wherein the intelligent filling method is characterized by comprising the following steps: in step S4, the text correction network is constructed, and the text correction network is used to correct text errors in the filling content, including the following steps:
Step S41: the data acquisition, the Chinese text error correction data set is acquired, the Chinese text error correction data set is preprocessed, and the preprocessed Chinese text error correction data set is obtained and divided into a training set B and a testing set B;
Step S42: the text error correction network is constructed, the text error correction network comprises a detection network and an error correction network, the error correction network is composed of two full-connection layers, the detection network utilizes a bidirectional GRU model to carry out forward and reverse coding on a text sequence to obtain a forward coding hidden state and a reverse coding hidden state, the last hidden layer is utilized to combine the forward coding hidden state and the reverse coding hidden state to obtain a combined hidden state, the combined hidden state is sent into the error correction network, and the calculation process is as follows:
；
；
；
；
In the method, in the process of the invention, Input representing a bidirectional GRU model,/>Word embedding for representing each character in a text sequence,/>Representation location embedding,/>Representation segment embedding; /(I)Representing the hidden state of forward coding,/>Representing the hidden state after merging,/>Representing hidden state of reverse coding,/>Representing a gated loop unit,/>Representing the previous hidden state of forward coding,/>Representing the latter hidden state of the reverse coding;
step S43: defining a loss function, wherein the loss function of the text error correction network consists of a loss function of the detection network and a loss function of the error correction network, and the formula is as follows:
；
；
；
In the method, in the process of the invention, Representing cross entropy loss of error correction network,/>Representing detection of network cross entropy loss,/>And/>Linear combination coefficients representing the loss function of the detection network and the loss function of the error correction network, respectively,/>Representing input data,/>Representing a probability distribution of the output of the detection network,/>Representing the probability distribution of the output of an error correction network,/>Representing the output of an error correction network,/>Representing an output of the detection network;
Step S44: model training and evaluation, namely inputting a training set B into a text error correction network to train, obtaining a text error correction network A after training is completed, evaluating the performance of the text error correction network A on a test set B, stopping training when the error of the test set B is not reduced in a plurality of continuous iteration times of the text error correction network A, and adjusting the hyper-parameters of the text error correction network A according to the performance of the text error correction network A on the test set B to obtain the text error correction network B;
step S45: error correction, text error correction network B is used for text error detection of the filled bidding scheme, and once an error is detected, text error correction network B will attempt to generate the correct text or provide candidate correction options.
6. An intelligent filling system for a bidding scheme, for implementing the intelligent filling method for the bidding scheme as claimed in any one of claims 1-5, characterized in that: the system comprises a bid-drawing scheme classification module, a bid-drawing scheme template module, an intelligent filling template module and a filling content error correction module;
The bid-drawing scheme classification module collects a historical bid-drawing scheme data set, preprocesses the historical bid-drawing scheme, obtains an optimal feature subset through a feature selection algorithm, classifies the historical bid-drawing scheme by using a classification model to obtain different types of bid-drawing schemes, and sends the different types of bid-drawing schemes to the bid-drawing scheme template module;
The method comprises the steps that an extraction bidding scheme template module receives different types of bidding schemes sent by a bidding scheme classification module, analyzes the structure of the bidding scheme by utilizing a natural language processing algorithm according to the different types of bidding schemes, extracts the format characteristics of each part, generates different bidding templates, and sends the different bidding templates to an intelligent filling template module;
The intelligent filling template module receives different bidding templates sent by extracting bidding scheme templates, determines bidding scheme templates according to existing data and requirements, encodes bidding scheme texts by using a BERT model, extracts global features, captures local features in the bidding scheme texts by using a local feature convolution network model, fuses the local features and the global features, generates final feature representation of the bidding scheme, intelligently fills each part in the templates according to the final feature representation, obtains filled bidding templates, and sends the filled bidding templates to the filling content error correction module;
the filling content error correction module receives the filled bid template sent by the intelligent filling template module, a text error correction network is constructed, and text errors in the filling content are corrected by the text error correction network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410448428.6A CN118052627A (en) | 2024-04-15 | 2024-04-15 | Intelligent filling method and system for bidding scheme |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410448428.6A CN118052627A (en) | 2024-04-15 | 2024-04-15 | Intelligent filling method and system for bidding scheme |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118052627A true CN118052627A (en) | 2024-05-17 |
Family
ID=91052199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410448428.6A Pending CN118052627A (en) | 2024-04-15 | 2024-04-15 | Intelligent filling method and system for bidding scheme |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118052627A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114218926A (en) * | 2021-12-17 | 2022-03-22 | 中山大学 | Chinese spelling error correction method and system based on word segmentation and knowledge graph |
| CN115293150A (en) * | 2022-08-10 | 2022-11-04 | 辽宁省肿瘤医院 | Automatic encoding method for operation record by fusing convolutional neural network and self-attention mechanism |
| CN116167920A (en) * | 2023-03-24 | 2023-05-26 | 浙江师范大学 | Image compression and reconstruction method based on super-resolution and priori knowledge |
| CN116305119A (en) * | 2023-02-24 | 2023-06-23 | 中国科学院信息工程研究所 | APT malicious software classification method and device based on predictive guidance prototype |
| WO2023173533A1 (en) * | 2022-03-17 | 2023-09-21 | 平安科技（深圳）有限公司 | Text error correction method and apparatus, device, and storage medium |
| CN117112782A (en) * | 2023-06-06 | 2023-11-24 | 四川文渊阁科技有限公司 | Method for extracting bid announcement information |
| CN117151084A (en) * | 2023-10-31 | 2023-12-01 | 山东齐鲁壹点传媒有限公司 | Chinese spelling and grammar error correction method, storage medium and equipment |
| CN117408650A (en) * | 2023-12-15 | 2024-01-16 | 辽宁省网联数字科技产业有限公司 | Digital bidding document making and evaluating system based on artificial intelligence |
-
2024
- 2024-04-15 CN CN202410448428.6A patent/CN118052627A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114218926A (en) * | 2021-12-17 | 2022-03-22 | 中山大学 | Chinese spelling error correction method and system based on word segmentation and knowledge graph |
| WO2023173533A1 (en) * | 2022-03-17 | 2023-09-21 | 平安科技（深圳）有限公司 | Text error correction method and apparatus, device, and storage medium |
| CN115293150A (en) * | 2022-08-10 | 2022-11-04 | 辽宁省肿瘤医院 | Automatic encoding method for operation record by fusing convolutional neural network and self-attention mechanism |
| CN116305119A (en) * | 2023-02-24 | 2023-06-23 | 中国科学院信息工程研究所 | APT malicious software classification method and device based on predictive guidance prototype |
| CN116167920A (en) * | 2023-03-24 | 2023-05-26 | 浙江师范大学 | Image compression and reconstruction method based on super-resolution and priori knowledge |
| CN117112782A (en) * | 2023-06-06 | 2023-11-24 | 四川文渊阁科技有限公司 | Method for extracting bid announcement information |
| CN117151084A (en) * | 2023-10-31 | 2023-12-01 | 山东齐鲁壹点传媒有限公司 | Chinese spelling and grammar error correction method, storage medium and equipment |
| CN117408650A (en) * | 2023-12-15 | 2024-01-16 | 辽宁省网联数字科技产业有限公司 | Digital bidding document making and evaluating system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TW389865B (en) | System and method for automated interpretation of input expressions using novel a posteriori probability measures and optimally trained information processing network | |
| CN110490081B (en) | Remote sensing object interpretation method based on focusing weight matrix and variable-scale semantic segmentation neural network | |
| RU2760471C1 (en) | Methods and systems for identifying fields in a document | |
| CN112434691A (en) | HS code matching and displaying method and system based on intelligent analysis and identification and storage medium | |
| CN108875769A (en) | Data mask method, device and system and storage medium | |
| CN109740515A (en) | One kind reading and appraising method and device | |
| Misra et al. | Web-SpikeSegNet: deep learning framework for recognition and counting of spikes from visual images of wheat plants | |
| Shan et al. | Robust encoder-decoder learning framework towards offline handwritten mathematical expression recognition based on multi-scale deep neural network | |
| CN113836891A (en) | Method and device for extracting structured information based on multi-element labeling strategy | |
| CN114647715A (en) | Entity recognition method based on pre-training language model | |
| Ma et al. | UniTranSeR: A unified transformer semantic representation framework for multimodal task-oriented dialog system | |
| Strait et al. | Automatic detection and uncertainty quantification of landmarks on elastic curves | |
| CN118052627A (en) | Intelligent filling method and system for bidding scheme | |
| Xie et al. | ICDAR 2023 CROHME: Competition on Recognition of Handwritten Mathematical Expressions | |
| He et al. | Text2analysis: A benchmark of table question answering with advanced data analysis and unclear queries | |
| CN109522892A (en) | The character image information labeling method of neural network aiding | |
| CN115454423A (en) | Static webpage generation method and device, electronic equipment and storage medium | |
| CN115512340A (en) | Intention detection method and device based on picture | |
| Niels et al. | Introducing Trigraph-trimodal writer identification | |
| Zhou et al. | SRRNet: A Transformer Structure with Adaptive 2D Spatial Attention Mechanism for Cell Phone-Captured Shopping Receipt Recognition | |
| CN115952259B (en) | Intelligent generation method of enterprise portrait tag | |
| Wei | An artificial intelligence method for text placement evaluation in maps | |
| Lods et al. | Graph Edit Distance for the analysis of children's on-line handwritten arithmetical operations | |
| US20230306769A1 (en) | Model Generation System and Model Generation Method | |
| Wick | Optical Medieval Music Recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |