CN118551760B - Purchasing file compliance checking system based on difference algorithm under AI large model - Google Patents

Abstract

The invention relates to a purchase file compliance checking system based on a difference algorithm under an AI large model, which is used for uploading and identifying a purchase file to be checked; identifying and extracting key words in a purchase file to be checked; inputting the key words into a preset violation word analysis unit, and outputting the key words judged to be the violation words; and outputting the key words judged to be illegal words, and generating a compliance check report of the purchase file. The method solves the problem that the part of speech judgment of the offensive word is not accurate enough because the vocabulary matched with the offensive word is not recognized in the prior art. In addition, the invention can also extract the related sentences or paragraphs of the keywords of different documents, judge whether the related paragraphs are consistent from front to back through the semantic recognition technology, report errors to the incompletely consistent content, automatically collect the use data and the use habit of the user, and automatically optimize the checking rule and the checking logic.

Description

Procurement document compliance checking system based on difference algorithm under AI big model

Technical Field

The present invention belongs to the technical field of document auditing and relates to a procurement document compliance checking system based on a difference algorithm under an AI large model.

Background Art

Compliance is a very important factor in the procurement process of an enterprise. Procurement documents need to comply with national laws and regulations, government policy requirements, and the internal procurement process and regulations of the enterprise. However, due to the large content of procurement documents, the wide range of fields involved, and the requirements of laws, regulations, finance, taxation, confidentiality and other aspects, manual inspection of the compliance of procurement documents is often time-consuming and laborious, and prone to loopholes and errors.

To this end, people have introduced a series of methods for compliance inspection of procurement documents. These methods infer whether the documents are compliant by identifying the illegal words in the documents, reducing the cost of manual inspection. However, these methods rely more on the identified illegal words and pay less attention to the words that are collocated with the illegal words, resulting in inaccurate judgment of the word class of the illegal words and reduced inspection efficiency. Illegal words often need to be in a special context to constitute violations. The same word may be illegal in one type of procurement document, but not in another type of procurement document. Therefore, in order to improve the accuracy of procurement compliance inspection, it is also necessary to identify the key words related to the illegal words to fully confirm whether the illegal words are actually illegal.

Summary of the invention

In order to solve the above problems existing in the prior art, the present invention provides a procurement document compliance checking system based on a difference algorithm under an AI big model.

The purpose of the present invention can be achieved through the following technical solutions:

The present application provides a procurement document compliance checking system based on a difference algorithm under an AI large model, including a file upload module, a word segmentation module, an illegal word analysis module and a report generation module, wherein the file upload module, the word segmentation module, the illegal word analysis module and the report generation module are communicatively connected, wherein:

The file upload module is used to upload and identify the procurement documents to be checked;

The word segmentation module is used to identify and extract key words in the procurement documents to be checked;

The illegal word analysis module is used to input key words into a preset illegal word analysis unit and output key words determined to be illegal words; the illegal word analysis unit includes the following construction steps:

S1. Collect a large-scale sample of procurement document data containing illegal words in historical periods;

S2, identifying the sentence or paragraph where the offending word is located, and extracting key words other than the offending word in the sentence or paragraph;

S3, using key words as explanatory variables and offending words as response variables, a random forest model was constructed to quantify the relationship between offending words and their corresponding key words;

The step of inputting the key words into a preset illegal word analysis unit and outputting the key words determined as illegal words comprises the following steps:

Determine the type of procurement document based on the procurement information of the procurement document to be checked;

Invoke the relevant offending words of the procurement document type in question;

Input the key words of the procurement document to be checked into a preset illegal word analysis unit, and output the predicted illegal words in each sentence or paragraph;

Compare the predicted illegal words with the related illegal words of the procurement document type. When the predicted illegal words are the same as the related illegal words, the sentence or paragraph containing the predicted illegal words is determined to be illegal.

The report generation module is used to generate a compliance check report for the procurement document based on the output of key words determined to be illegal words.

Furthermore, in the word segmentation module, the identification and extraction of key words in the procurement document to be checked includes the following steps:

T1. Data preprocessing: preprocess the procurement documents to be inspected, including removing redundant punctuation marks, stop words and noise characters;

T2. Entity recognition: Entity recognition technology is used to identify the procurement information of procurement documents, including supplier information, procurement items and procurement prices;

T3. Relationship extraction: Identify key words that have dependencies with procurement information.

Furthermore, in step S3, the random forest model includes the following steps:

S31, encoding the key words according to the order of the key words in the sentence or paragraph, and converting the key words into a word sequence vector;

S32, using the word sequence vector as a feature and the illegal word as a label, and dividing the procurement document data sample into a training set and a test set;

S33, Random Sampling: Randomly extract a certain number of data samples from the training set, and construct multiple training sets containing different data samples with replacement;

S34, constructing a decision tree: for each randomly sampled training set, construct a decision tree model;

S35, Ensemble Decision Tree: Integrate multiple decision tree models into a random forest model and use the voting mechanism to make the final prediction;

S36. Model evaluation: Use evaluation metrics to evaluate the prediction performance of the random forest model on the test set.

Furthermore, in step S31, the key words are encoded according to their order in the sentence or paragraph, and the key words are converted into word sequence vectors, specifically using one-hot encoding, bag-of-words model or word embedding model.

Furthermore, in step S32, the procurement document data samples are divided into a training set and a test set, wherein 70% of the data samples are used as the training set and 30% of the data samples are used as the test set.

Furthermore, in step S34, the decision tree model is configured as a CART model.

Furthermore, in step S36, the evaluation index includes accuracy, precision, recall or F1 value.

Furthermore, in the report generation module, the compliance check report includes procurement information, a description of the problem in the sentence or paragraph generated with the illegal words, and modification suggestions.

Furthermore, the illegal word analysis module also uses semantic recognition technology to compare sentences or paragraphs containing the same keywords in the procurement documents to be checked of different templates. When the meanings of the sentences or paragraphs containing the same keywords are inconsistent, the system reports an error.

Furthermore, the report generation module is also trained based on the user's historical correction records of illegal words, and optimizes the illegal word checking results to mark key illegal sentences.

Furthermore, in the word segmentation module, the sentences or paragraphs containing the same key words in the procurement document to be checked and the stored procurement document are checked by a difference algorithm, and the differences between the sentences or paragraphs are determined and marked.

Beneficial effects of the present invention:

1) By identifying the offending words and other key words in the sentence or paragraph where the offending words are located, the connection between the key words and the offending words is established, which solves the problem that the existing technology does not identify the words related to the offending words, resulting in inaccurate part of speech judgment of the offending words;

2) For different types of procurement documents, by identifying the procurement information of the procurement documents and extracting the key words that have a dependency relationship with it, the key words of illegal words in different types of procurement documents are defined to improve the judgment of the part of speech of illegal words;

3) A random forest model is used in the process of building the connection between key words and illegal words to improve the accuracy of quantifying the relationship between the two;

4) Encode the key words according to their order in the sentence or paragraph, and convert them into word sequence vectors to highlight the key word features and improve the accuracy of model prediction.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

FIG1 is a flow chart of a procurement document compliance checking system based on a difference algorithm under an AI big model in the present invention.

DETAILED DESCRIPTION

In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

Please refer to Figure 1. This application provides a procurement document compliance inspection system based on a difference algorithm under an AI large model, including a file upload module, a word segmentation module, an illegal word analysis module, and a report generation module. The file upload module, the word segmentation module, the illegal word analysis module, and the report generation module are communicatively connected, wherein:

The file upload module is used to upload and identify the procurement documents to be checked;

In this embodiment, the system adopts an advanced B/S architecture and supports multiple users to upload procurement documents online at the same time. The interface design is simple and intuitive, and the operation process is clear and concise, which is convenient for users to get started quickly. Through the user-friendly graphical interface, users can easily complete operations such as uploading, viewing, reviewing and exporting files. Users can upload procurement documents through the software interface, and support multiple file formats (such as Word, Excel, etc.). The system has a powerful file recognition function and can automatically extract procurement information in the file, such as supplier information, purchased items, prices, etc., to provide data support for subsequent audit work.

The word segmentation module is used to identify and extract key words in the procurement documents to be checked;

Furthermore, in the word segmentation module, the identification and extraction of key words in the procurement document to be checked includes the following steps:

T1. Data preprocessing: preprocess the procurement documents to be inspected, including removing redundant punctuation marks, stop words and noise characters;

In this embodiment, the following steps may be followed to preprocess the procurement document to remove redundant punctuation marks, stop words, and noise characters:

1) Remove punctuation: Use regular expressions or string processing functions to remove punctuation from the text, such as commas, periods, exclamation marks, etc. This can retain keywords and content and exclude useless symbols.

2) Remove stop words: Stop words are words that appear frequently in the text but usually do not carry important meanings, such as "的", "和", "是", etc. You can use a stop word list (such as the default stop word list provided by the NLTK library) or a custom stop word list to remove these words to reduce noise and extract purchase information.

3) Remove noise characters: Use regular expressions or string processing functions to remove noise characters in the text, such as HTML tags, special characters such as #, @, etc. This can clean the text and improve the effect of subsequent processing.

It should be noted that during preprocessing, the punctuation, stop words and noise characters that need to be removed should be determined according to specific needs and scenarios to avoid accidentally deleting important information. At the same time, the preprocessed text should be properly verified to ensure that it does not affect the integrity and readability of the procurement documents.

T2. Entity recognition: Entity recognition technology is used to identify the procurement information of procurement documents, including supplier information, procurement items and procurement prices;

T3. Relationship extraction: Identify key words that have dependencies with procurement information.

In this embodiment, when entity recognition technology is used to identify the procurement information of the procurement document, the focus can be placed on important contents such as supplier information, purchased items and purchase price. Through entity recognition technology, these procurement information in the document can be automatically marked, so that it is easy to identify and extract. At the same time, it is also very important to understand the dependent key words between the procurement information. These dependent key words can help us understand the connection and function between the procurement information. For example, supplier information may be associated with the delivery capability, quality assurance, etc. of the purchased items; the purchase price may be associated with the quantity, specifications, etc. of the purchased items. In order to realize the identification of procurement information and dependent key words, entity relationship extraction technology can be used. This technology can model and mark the procurement information and the dependency relationship between them by marking the procurement documents. Then, using the trained entity relationship extraction model, these relationships can be automatically identified and extracted. The whole process includes steps such as data preparation, entity recognition, entity relationship labeling, entity relationship extraction model training and entity relationship extraction. For specific fields and task requirements, the model can be optimized and customized to improve accuracy and adaptability.

It should be noted that the performance of entity relationship extraction is affected by many factors, including data quality, annotation accuracy, and diversity of training data. Therefore, in practical applications, the model needs to be carefully adjusted and optimized to ensure the accuracy and reliability of entity relationship extraction.

Furthermore, in the word segmentation module, the sentences or paragraphs containing the same key words in the procurement document to be checked and the stored procurement document are checked by a difference algorithm, and the differences between the sentences or paragraphs are determined and marked.

In this embodiment, a difference algorithm is a computer algorithm for comparing the differences between two or more texts. It determines the similarities and differences between them by comparing the differences between characters, words, sentences or other text units in the text. Difference algorithms can be used in various applications, including version control, document comparison, text merging, etc. In text comparison, the difference algorithm can identify the parts that are added, deleted or modified between two texts. One of the most common difference algorithms is the longest common subsequence (LCS) algorithm, which can determine the longest common sequence between two texts. By finding the longest common subsequence, the difference algorithm can determine which parts are the same and which parts are different. Another commonly used difference algorithm is a line-based difference algorithm, such as line diff or revision difference algorithm (diff). These algorithms compare the differences between text lines and give indications of added, deleted and modified lines. Difference algorithms can also provide more complex text comparison functions, such as comparison based on text structure and comparison at the semantic level. These algorithms can identify deeper differences between texts, such as sentence rearrangement, semantic rewriting, etc. In general, difference algorithms can help us identify and understand the differences between texts, thereby providing support for tasks such as text processing, version control and document comparison. It can be used in procurement document audits to detect information errors, inconsistencies and plagiarism, and improve the quality and accuracy of procurement documents:

First, the difference algorithm can help find information errors and inconsistencies in documents. In multiple documents, sentences or paragraphs describing the same key words may have differences, such as inconsistent numbers, inappropriate words, or logical contradictions. Through the difference algorithm, these differences can be accurately detected, allowing procurement personnel to quickly identify and correct errors and ensure the accuracy and consistency of procurement documents.

Secondly, the difference algorithm can be used to detect issues with plagiarism or unoriginal content in documents. Although plagiarism may not be common in procurement documents, the difference algorithm can still find duplication or modification of text in different documents, helping to identify possible intellectual property issues. By comparing the differences in sentences or paragraphs where key words are located, procurement personnel can avoid using unauthorized or insufficiently cited content, ensuring the legitimacy and credibility of procurement documents.

Thirdly, the difference algorithm can also help find document loopholes and omissions. In multiple documents, information about the same key word may be missing or missing. Through difference checking, the differences between documents can be revealed, the missing information can be identified, and the procurement staff can be provided with directions for supplementation and improvement. This helps ensure that the procurement documents contain comprehensive and comprehensive information and reduce the risks and problems caused by omissions.

Then, the difference algorithm can also provide accurate summary and comprehensive information. By comparing the differences in sentences or paragraphs where key words are located, a comprehensive perception and summary report can be generated. This summary and comprehensive information is provided to procurement personnel to help them better understand the similarities and differences between multiple documents, thereby providing a more comprehensive and accurate basis for the decision-making and evaluation process.

Finally, the difference algorithm can assist the procurement decision and evaluation process. By comparing the differences, procurement personnel can evaluate the differences between different solutions or providers and understand the unique advantages and disadvantages in each document. This helps procurement personnel make more informed decisions and evaluate whether the provider meets procurement needs and standards.

The illegal word analysis module is used to input key words into a preset illegal word analysis unit and output key words determined to be illegal words; the illegal word analysis unit includes the following construction steps:

S1. Collect a large-scale sample of procurement document data containing illegal words in historical periods;

S2, identifying the sentence or paragraph where the offending word is located, and extracting key words other than the offending word in the sentence or paragraph;

S3, using key words as explanatory variables and offending words as response variables, a random forest model was constructed to quantify the relationship between offending words and their corresponding key words;

In this embodiment, Random Forest is an ensemble learning method, which is composed of multiple decision trees. Each decision tree is trained independently, and ensemble prediction is performed by voting or averaging.

Following are the main features and steps of Random Forest:

1) Feature randomness: When constructing each decision tree, random forest introduces feature randomness. This means that when each node is divided, only a portion of the randomly selected features is considered for division. This can help reduce the correlation between decision trees, improve the diversity of the model, and reduce the risk of overfitting.

2) Sample randomness: For each decision tree training sample, random forest uses bootstrap sampling to randomly select. This means that each decision tree is trained based on a different sample set, increasing the diversity of the model.

3) Building a decision tree: Each decision tree performs feature selection and node division based on the purity index of the node division, such as Gini Index or information gain. The depth of the decision tree can be controlled by adjusting parameters.

4) Ensemble prediction: For classification problems, random forests determine the final prediction results by voting. For regression problems, random forests get the final prediction results by averaging the output of each decision tree.

The random forest model has the following advantages:

1) Robustness: Random forests can handle high-dimensional data and a large number of features, and are robust to outliers and noise.

2) Interpretability: Random forests can provide feature importance indicators to help understand the contribution of features to model predictions.

3) Anti-overfitting: By introducing feature randomness and sample randomness, random forest can reduce the variance of the model, improve generalization ability, and reduce the risk of overfitting.

The step of inputting the key words into a preset illegal word analysis unit and outputting the key words determined as illegal words comprises the following steps:

Determine the type of procurement document based on the procurement information of the procurement document to be checked;

Invoke the relevant offending words of the procurement document type in question;

Input the key words of the procurement document to be checked into a preset illegal word analysis unit, and output the predicted illegal words in each sentence or paragraph;

Compare the predicted illegal words with the related illegal words of the procurement document type. When the predicted illegal words are the same as the related illegal words, the sentence or paragraph containing the predicted illegal words is determined to be illegal.

At present, the technical procurement document verification system basically requires that the procurement documents be structured in the system, and then the specified keywords are identified in the specified position for verification. This technology is not easy to promote because the procurement document templates of different industries and different customers are different. However, this embodiment can also identify the sentences or paragraphs where the keywords appear, and use semantic recognition technology to compare whether the relevant paragraphs express the same content. If the content expresses the same content but the meaning is not completely consistent, an error will be reported. Through this process, users do not need to structure the procurement documents in the system to realize the audit of procurement documents with different templates.

Furthermore, the illegal word analysis module also uses semantic recognition technology to compare sentences or paragraphs containing the same keywords in the procurement documents to be checked of different templates. When the meanings of the sentences or paragraphs containing the same keywords are inconsistent, the system reports an error.

The report generation module is used to generate a compliance check report for the procurement document based on the output of key words determined to be illegal words.

Furthermore, in step S3, the random forest model includes the following steps:

S31, encoding the key words according to the order of the key words in the sentence or paragraph, and converting the key words into a word sequence vector;

In this embodiment, the sentence or paragraph containing the offending word in each data sample contains multiple keywords, which are often matched with the offending words in a certain order. The key words are encoded according to the order of the key words in the sentence or paragraph, and the key words are converted into word sequence vectors. The text form of the key words can be converted into data form. At the same time, the sequential encoding can form a data sequence with the characteristics of the key words, making the characteristics more obvious and improving the accuracy of the model. The process of converting key words into word sequence vectors can use different encoding methods, such as one-hot encoding, bag of words model or word embedding model (such as Word2Vec, GloVe, etc.), and the specific selection is related to the characteristics of the data and the task requirements.

By converting key words into sequence vectors, the order information of key words in sentences or paragraphs can be extracted and converted into numerical features that can be quantified and compared. This can help the model better understand the contextual relationship and semantic information of key words and learn specific patterns and rules from them. By encoding the order of key words, more obvious features can be provided to the model, improving the model's recognition accuracy of key words. The sequential encoding of key words can help the model identify specific collocations and phrase structures, thereby better understanding the semantics of the sentence or paragraph where the offending word is located and making more accurate predictions or classifications.

Furthermore, in step S31, the key words are encoded according to their order in the sentence or paragraph, and the key words are converted into word sequence vectors, specifically using one-hot encoding, bag-of-words model or word embedding model.

S32. Use word sequence vectors as features and illegal words as labels, and divide the procurement document data samples into training sets and test sets, where 70% of the data samples are used as training sets and 30% of the data samples are used as test sets;

S33, Random Sampling: Randomly extract a certain number of data samples from the training set, and construct multiple training sets containing different data samples with replacement;

S34, constructing a decision tree: for each randomly sampled training set, construct a decision tree model;

Furthermore, in step S34, the decision tree model is configured as a CART model.

Classification and Regression Trees (CART) is a decision tree algorithm used for classification and regression analysis. CART recursively divides the dataset into smaller subsets, selects the best features and split points at each node to minimize the Gini impurity (for classification) or the mean square error (for regression), and finally generates a tree for prediction. The CART model is easy to understand and interpretable, and can handle nonlinear relationships and high-dimensional data. By repeatedly dividing the dataset, CART can effectively capture complex patterns in the data and is also robust to missing values and outliers. In a random forest, each decision tree is built based on the Classification and Regression Tree (CART) algorithm. Specifically, each tree in a random forest is an independent CART decision tree that randomly selects features for modeling. Random forests make the final decision by integrating the prediction results of multiple CART decision trees, thereby improving the accuracy and generalization ability of the overall model. Each tree is built for a different random subset of the training data, and only one random subset of features is considered when selecting the best feature at each node, which makes the random forest have a certain degree of randomness and reduces the risk of overfitting. Therefore, the random forest takes advantage of the CART decision tree, combines the prediction results of multiple trees, improves the stability and prediction ability of the model, and has achieved widespread success in practical applications.

S35, Ensemble Decision Tree: Integrate multiple decision tree models into a random forest model and use the voting mechanism to make the final prediction;

It should be noted that in a random forest, each decision tree is built based on a different training data set and feature subset, and has a certain degree of randomness. The final prediction result can be determined by a voting mechanism or by taking the average. In classification problems, a majority vote can be used to select the category with the most votes as the final prediction result; in regression problems, the prediction results of multiple trees can be averaged as the final output. By integrating multiple decision trees into a random forest model, the risk of overfitting can be effectively reduced and the robustness and accuracy of the model can be improved.

S36. Model evaluation: Use evaluation metrics to evaluate the prediction performance of the random forest model on the test set.

Furthermore, in step S36, the evaluation index includes accuracy, precision, recall or F1 value.

Commonly used indicators for evaluating the predictive performance of random forest models include accuracy, precision, recall, and F1 value. The following is a description of these indicators:

1) Accuracy: Accuracy indicates the ratio of the number of samples correctly predicted by the model to the total number of samples. The calculation formula is: Accuracy = Number of samples correctly predicted / Total number of samples. Accuracy provides an overall evaluation indicator, but it may cause bias when the samples are unbalanced.

2) Precision: Precision indicates the proportion of samples predicted as positive by the model that are actually positive. The calculation formula is: Precision = True Positive (predicted as positive and actually positive) / (True Positive + False Positive). Precision measures the model's ability to make correct predictions among samples predicted as positive.

3) Recall: Recall represents the proportion of samples that are actually positive that are predicted as positive by the model. The calculation formula is: Recall = True Positive / (True Positive + False Negative). Recall measures the ability of the model to identify positive samples.

4) F1 value: F1 value is the harmonic mean of precision and recall, which can comprehensively consider the precision and recall of the model. The calculation formula is: F1 = 2 * (precision * recall) / (precision + recall). The higher the F1 value, the better the balance between precision and recall of the model.

Furthermore, in the report generation module, the compliance check report includes procurement information, a description of the problem in the sentence or paragraph generated with the illegal words, and modification suggestions.

In this embodiment, the content and structure of the compliance check report are roughly as follows:

1) Introduction:

In the introduction, briefly introduce the purpose and background of the compliance review, explain the content and scope of this report, and provide basic information about the compliance review, such as time, scope, and participants.

2) Purchasing information:

In this section, list the purchase information, including the products or services purchased, supplier information, purchase time, purchase amount, etc. Make sure to provide enough details so that users understand the context and background of the inspection.

3) Description of the sentence or paragraph where the offending word is located:

In this section, describe the problems with the sentences or paragraphs where the offending words are located one by one. For each problem, briefly describe which offending words are used, where they are located in the sentence or paragraph, and the impact on compliance. You can give specific examples and provide screenshots or quotes to the original content when necessary.

4) Suggestions for modification:

In this section, provide suggestions for each problem. For each use of the offending word, provide a compliant alternative word or expression. Also, provide guidance on modifying sentence or paragraph structure, reorganizing content, or changing expression. Suggestions should be specific and clear, and explain why such changes are made as much as possible.

5) Summary:

In the summary section, summarize the results of the compliance check, briefly summarize the problems found and the recommended solutions. Emphasize the importance of compliance and encourage taking measures to solve problems and comply with compliance regulations.

When generating compliance inspection reports, ensure that the issues and recommendations are clearly communicated, use concise language, and provide sufficient supporting material to support the conclusions in the report. Also, follow any organization- or industry-specific requirements or guidelines to ensure consistency and readability of the report.

Furthermore, the report generation module is also trained based on the user's historical correction records of illegal words, and optimizes the illegal word checking results to mark key illegal sentences.

During the daily audit process, normal words, sentences or paragraphs in the procurement documents may be mistakenly judged as violations, and users need to screen the correct audit rules from the generated illegal entries based on experience. To this end, this application is trained through the user's historical correction records of illegal words to record the user's correction habits, improve the audit rules of procurement documents, optimize the inspection results of illegal words, and finally delete the sentences and paragraphs that are mistakenly judged as violations and mark the key illegal sentences and paragraphs. Machine learning algorithms such as support vector machines and random forests can be used to train the correction records of illegal words.

Beneficial effects of the present invention:

1) By identifying the offending words and other key words in the sentence or paragraph where the offending words are located, the connection between the key words and the offending words is established, which solves the problem that the existing technology does not identify the words related to the offending words, resulting in inaccurate part of speech judgment of the offending words;

2) For different types of procurement documents, by identifying the procurement information of the procurement documents and extracting the key words that have a dependency relationship with it, the key words of illegal words in different types of procurement documents are defined to improve the judgment of the part of speech of illegal words;

3) A random forest model is used in the process of building the connection between key words and illegal words to improve the accuracy of quantifying the relationship between the two;

4) Encode the key words according to their order in the sentence or paragraph, and convert them into word sequence vectors to highlight the key word features and improve the accuracy of model prediction.

The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims (9)

1. A procurement document compliance check system based on a difference algorithm under an AI large model, characterized in that it includes a file upload module, a word segmentation module, a violation word analysis module and a report generation module, wherein the file upload module, the word segmentation module, the violation word analysis module and the report generation module are communicatively connected, wherein: The file upload module is used to upload and identify the procurement documents to be checked; The word segmentation module is used to identify and extract key words in the procurement documents to be checked; The illegal word analysis module is used to input key words into a preset illegal word analysis unit and output key words determined to be illegal words; the illegal word analysis unit includes the following construction steps: S1. Collect a large-scale sample of procurement document data containing illegal words in historical periods; S2, identifying the sentence or paragraph where the offending word is located, and extracting key words other than the offending word in the sentence or paragraph; S3, using key words as explanatory variables and offending words as response variables, a random forest model was constructed to quantify the relationship between offending words and their corresponding key words; The random forest model comprises the following steps: S31, encoding the key words according to the order of the key words in the sentence or paragraph, and converting the key words into a word sequence vector; S32, using the word sequence vector as a feature and the illegal word as a label, and dividing the procurement document data sample into a training set and a test set; S33, Random Sampling: Randomly extract a certain number of data samples from the training set, and construct multiple training sets containing different data samples with replacement; S34, constructing a decision tree: for each randomly sampled training set, construct a decision tree model; S35, Ensemble Decision Tree: Integrate multiple decision tree models into a random forest model and use the voting mechanism to make the final prediction; S36. Model evaluation: Use evaluation metrics to evaluate the prediction performance of the random forest model on the test set; The step of inputting the key words into a preset illegal word analysis unit and outputting the key words determined as illegal words comprises the following steps: Determine the type of procurement document based on the procurement information of the procurement document to be checked; Invoke the relevant offending words of the procurement document type in question; Input the key words of the procurement document to be checked into a preset illegal word analysis unit, and output the predicted illegal words in each sentence or paragraph; Compare the predicted illegal words with the related illegal words of the procurement document type. When the predicted illegal words are the same as the related illegal words, the sentence or paragraph containing the predicted illegal words is determined to be illegal. The report generation module is used to generate a compliance check report for the procurement document based on the output of key words determined to be illegal words. 2. According to claim 1, a procurement document compliance inspection system based on a difference algorithm under an AI large model is characterized in that: in the word segmentation module, the identification and extraction of key words in the procurement document to be inspected includes the following steps: T1. Data preprocessing: preprocess the procurement documents to be inspected, including removing redundant punctuation marks, stop words and noise characters; T2. Entity recognition: Entity recognition technology is used to identify the procurement information of procurement documents, including supplier information, procurement items and procurement prices; T3. Relationship extraction: Identify key words that have dependencies with procurement information. 3. According to claim 1, a procurement document compliance check system based on the difference algorithm under the AI large model is characterized in that: in step S31, the key words are encoded according to the order of the key words in the sentence or paragraph, and the key words are converted into word sequence vectors, specifically using one-hot encoding, bag of words model or word embedding model. 4. According to claim 1, a procurement document compliance checking system based on the difference algorithm under the AI large model is characterized in that: in step S32, the procurement document data samples are divided into a training set and a test set, wherein 70% of the data samples are used as the training set and 30% of the data samples are used as the test set. 5. According to claim 1, a procurement document compliance checking system based on a difference algorithm under an AI large model is characterized in that: in step S34, the decision tree model is configured as a CART model. 6. According to claim 1, a procurement document compliance check system based on a difference algorithm under an AI large model is characterized in that: the illegal word analysis module also compares sentences or paragraphs containing the same keywords in procurement documents to be checked of different templates through semantic recognition technology. When the content and meaning of the sentences or paragraphs containing the same keywords are inconsistent, the system reports an error. 7. According to claim 1, a procurement document compliance check system based on a difference algorithm under an AI large model is characterized in that: in the report generation module, the compliance check report includes procurement information, a description of the problem of the sentence or paragraph generated by the illegal words, and modification suggestions. 8. According to claim 1, a procurement document compliance check system based on a difference algorithm under an AI large model is characterized in that: the report generation module is also trained according to the error correction records of illegal words in the user's historical period, optimizes the illegal word check results, and marks key illegal sentences. 9. According to claim 1, a procurement document compliance check system based on a difference algorithm under an AI large model is characterized in that: in the word segmentation module, the difference algorithm is also used to check the sentences or paragraphs containing the same key words in the procurement document to be checked and the stored procurement document, and the differences in the sentences or paragraphs are determined and marked.