Evaluation level prediction method based on evaluation rule and evaluation text content

Abstract

The invention relates to a review level prediction method based on a review rule and review text content, and the method comprises the following steps: collecting historical bid evaluation data, grading the historical bid evaluation data according to a scoring rule, and forming a data set D1 according to review terms and scoring levels; obtaining an evaluation rule, and analyzing a division rule of an evaluation criterion and a score condition; text content is obtained from the review terms, and key information and summary information of the overall content are extracted; the summary information and the keywords are converted into vector data, expert review grades are subjected to normalization processing, and then the expert review grades are input into a classification model to be trained; performing review grade classification prediction on the bidding key information in the bidding document by using the trained classification model; according to the method, the workload of review personnel is reduced, the review speed is increased, the efficiency of the whole review process is improved, multiple attributes and factors are comprehensively considered, the influence of subjective factors can be reduced, and the comprehensiveness and comprehensiveness of review are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bidding and tendering, and in particular relates to a method for predicting an evaluation grade based on evaluation rules and evaluation text content. Background Art

[0002] Open bidding can attract more suppliers with stronger strength and higher skills to participate in the bidding, so that purchasers have more choices, thereby optimizing resource allocation, improving procurement efficiency, avoiding the influence of political, interpersonal and other factors on procurement decisions, and ensuring the fairness, justice and transparency of procurement activities.

[0003] At present, illegal and irregular problems of bidding units in public bidding and procurement have persisted despite repeated prohibitions, and there are cases of exclusion and restriction of potential bidders, bid rigging and collusion, and unfair bid evaluation and review. With the development of enterprises and the digital transformation of procurement business, the volume of bidding business has continued to increase, and a large number of basic businesses have led to repetitive and mechanical work, which has seriously affected the quality, efficiency and benefits of bidding and procurement. Summary of the invention

[0004] The technical problem of the present invention is: to propose a method for predicting the evaluation level based on the evaluation rules and the evaluation text content, to extract the evaluation rules of different evaluation factors through the analysis and classification of historical data, to summarize the key information and the overall content of different evaluation factors, to input them into the classification model for training, and to classify and predict the evaluation level of the key bidding information in the bid document; The technical solution of the present invention is a review grade prediction method based on review rules and review text content, comprising the following steps: S1: Collect historical bid evaluation data and classify them according to the scoring rules, and then form data set D1 according to the evaluation terms and scoring levels; S2: Obtain the evaluation rules from the data set D1, and analyze the evaluation criteria and the scoring rules; S3: Use text generation and extraction technology to obtain text content from the review terms, and then extract key information and summary information of the overall content; S4: converting the summary information and keywords into vector data to obtain data set D2, and normalizing the expert review levels to obtain data set D3; S5: Input the data sets D2 and D3 into the classification model for training; S6: Use the trained classification model to classify and predict the evaluation level of the key bidding information in the bid document; S7: Optimize the review level classification result of step S6 in combination with the review terms and rules.

[0005] Optimized, in step S1, the historical bid evaluation data includes the manual judgment and scoring of the key bidding information in the bid documents of different bidders by the bid evaluation experts in the historical bid evaluation data.

[0006] Optimally, in step S2, the evaluation rules are obtained from the data set D1, including dividing the scoring conditions and extracting evaluation criteria of different levels from different evaluation rules. Furthermore, step S4 further includes the following sub-steps: 1) Establish an association between the review content in data set D2 and the review level in data set D3; 2) Reconstruct the review level by analyzing keywords, extracting core content, dividing the document, traversing the keyword table and removing non-keywords; 3) According to the keyword table, put the sentences containing the relevant review level keywords in the document in front; 4) Verify multi-label documents and remove redundant information; The optimization removes redundant information and also includes using the TF-IDF algorithm to retrieve high-frequency words in review-level documents, verifying the high-frequency words, and then deleting meaningless high-frequency words.

[0007] Furthermore, step S5 includes constructing a classification model of a support vector machine model, vectorizing the text, and performing feature selection and screening on the text vector before inputting it into the model for training.

[0008] Optimized, step S7, also includes the review experts optimizing the grade prediction result information of different review rules based on the review rules and the grade classification model prediction results.

[0009] Compared with the prior art, the beneficial effects of the present invention include: 1) The review grade prediction method of the present invention is based on review rules and review text content, which reduces the workload of reviewers, speeds up the review, improves the efficiency of the entire review process, comprehensively considers multiple attributes and factors, and improves the comprehensiveness and comprehensiveness of the review.

[0010] 2) The present invention can reduce the influence of subjective factors through summary generation technology and review rule identification, and can be applied to different fields and scenarios, making the review results more objective and scientific. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0012] Figure 1 This is a flow chart of the review grade prediction method based on review rules and review text content of the present invention. DETAILED DESCRIPTION

[0013] like Figure 1 As shown, the review level prediction method based on review rules and review text content includes the following steps: S1: Collect historical bid evaluation data, and classify the historical bid evaluation data according to the scoring rules, label the corresponding grades A, B, C and D, and then form a data set D1 according to the evaluation terms and scoring levels; The historical bid evaluation data in step S1 includes manual evaluation and scoring of key bidding information in bid documents of different bidders by bid evaluation experts in the historical bid evaluation data.

[0014] S2: Obtain the evaluation rules from the data set D1, and analyze the evaluation criteria and the scoring rules; In step S2, the evaluation rules are obtained from the data set D1, including dividing the scores and extracting evaluation criteria of different levels from different evaluation rules.

[0015] 60 points will be awarded for performance that meets the procurement document qualification requirements, and 10 points will be added for each additional performance requirement. The scores will be divided into grades, with the grades of 100-85 points, 85-70 points, 70-60 points, and below 60 points being A, B, C, and D respectively; S3: Use text generation and extraction technology to obtain text content from the review terms, and then extract key information and summary information of the overall content; S4: converting the summary information and keywords into vector data to obtain data set D2, and normalizing the expert review levels to obtain data set D3; Step S4 also includes the following sub-steps: 1) Establish an association between the review content in data set D2 and the review level in data set D3; Extract core words from the summary information in the review content, and determine whether the core words are key words. If they are key words, add the core words to the keyword table, and associate the summary information of the bid document with one or more review levels.

[0016] 2) For multi-label summary information, the review level is reconstructed by analyzing keywords, extracting core content, dividing the document, traversing the keyword table and removing non-keywords; 3) According to the keyword table, put the sentences containing the relevant review level keywords in the document in front; Taking review level A as an example, read the keyword list of A, classify it and get a list, then traverse the keyword table for matching. If it exists, put the list at the first place, if not, process it as redundant information.

[0017] 4) Verify multi-label documents and remove redundant information; The multi-label summary information of the above steps is judged whether it belongs to multiple categories, and the high-frequency words of the review level documents are retrieved using the TF-IDF algorithm, and the high-frequency words are verified, and then the meaningless high-frequency words are deleted.

[0018] S5: Input the data sets D2 and D3 into the classification model for training; Step S5 includes constructing a classification model of a support vector machine model, vectorizing the text, performing feature selection and screening on the text vector, and then inputting the text vector into the model for training.

[0019] S6: Use the trained classification model to classify and predict the evaluation level of the key bidding information in the bid document; S7: Optimize the review level classification result of step S6 in combination with the review terms and rules.

[0020] Step S7 also includes the review experts optimizing the grade prediction result information of different review rules based on the review rules and the grade classification model prediction results.

[0021] In order to make the technical personnel in the technical field better understand the scheme of the present invention, the above-mentioned embodiments are only the preferred technical schemes of the present invention and should not be regarded as limitations of the present invention. The protection scope of the present invention should be the technical schemes recorded in the claims, including the equivalent replacement schemes of the technical features in the technical schemes recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A review grade prediction method based on review rules and review text content, characterized in that: The following steps are involved: S1: Collect historical bid evaluation data and classify them according to the scoring rules, and then form data set D1 according to the evaluation terms and scoring levels; S2: Obtain the evaluation rules from the data set D1, and analyze the evaluation criteria and the scoring rules; S3: Use text generation and extraction technology to obtain text content from the review terms, and then extract key information and summary information of the overall content; S4: converting the summary information and keywords into vector data to obtain data set D2, and normalizing the expert review levels to obtain data set D3; S5: Input the data sets D2 and D3 into the classification model for training; S6: Use the trained classification model to classify and predict the evaluation level of the key bidding information in the bid document; S7: Optimize the review level classification result of step S6 in combination with the review terms and rules.

2. The method for predicting review grades based on review rules and review text content according to claim 1, characterized in that: In step S1, the historical bid evaluation data includes manual evaluation and scoring of key bidding information in bid documents of different bidders by bid evaluation experts in the historical bid evaluation data.

3. The method for predicting review grades based on review rules and review text content according to claim 1, characterized in that: In step S2, the evaluation rules are obtained from the data set D1, including dividing the scores and extracting evaluation criteria of different levels from different evaluation rules.

4. According to the review grade prediction method based on review rules and review text content according to claim 1, the step S4 further comprises the following sub-steps: 1) Establish a correlation between the review content in data set D2 and the review level in data set D3, and create a keyword table; 2) For multi-label summary information, the review level is reconstructed by analyzing keywords, extracting core content, dividing the document, traversing the keyword table and removing non-keywords; 3) According to the keyword table, put the sentences containing the relevant review level keywords in the document in front; 4) Verify multi-label documents and remove redundant information.

5. According to the review grade prediction method based on review rules and review text content in claim 4, the deletion of redundant information also includes using the TF-IDF algorithm to retrieve high-frequency words in the review grade document, and verifying the high-frequency words, and then deleting meaningless high-frequency words.

6. According to the review level prediction method based on review rules and review text content in claim 1, step S5 includes constructing a classification model of a support vector machine model, vectorizing the text, and performing feature selection and screening on the text vector before inputting it into the model for training.

7. According to the review grade prediction method based on review rules and review text content as described in claim 1, step S7 also includes review experts optimizing the grade prediction result information of different review rules based on the review rules and grade classification model prediction results.

Images