An artificial intelligence-based key data extraction method, system and memory

Abstract

A method, system, and storage for key data extraction based on artificial intelligence (AI) belong to the field of data processing technology. The method includes: training an AI language model using pre-training corpus, key text, and expert-provided questions to obtain a trained AI language model; representing multiple key questions extracted by the trained AI language model using vectors and acquiring semantic information of the key questions; clustering the multiple key questions; distributing the clustering results to each member of an expert group; each expert member revising the clustered key questions to obtain multiple revised key questions; re-inputting the revised key questions into the trained AI language model until the key questions converge; all experts reaching a consensus; and outputting the converged key questions as the final generated key questions. This invention can intelligently generate key questions, thereby improving the efficiency of expert aggregation.

Description

Technical Field

[0001] This invention relates to a key data extraction method, system, and memory based on artificial intelligence, belonging to the field of artificial intelligence technology. Background Technology

[0002] With the rapid development of deep learning and natural language processing technologies, language models trained with large amounts of labeled data can quickly and efficiently understand and analyze text data. They possess capabilities such as sentiment analysis, language translation, text classification, information extraction, and text generation. Through data-driven models, they can understand, analyze, and summarize document content and intelligently generate key questions.

[0003] However, there are domain differences between currently popular pre-trained language models and specific domains. Directly summarizing the key questions raised by various experts using existing language models results in low accuracy. Summary of the Invention

[0004] This invention provides a key data extraction method, system, and memory based on artificial intelligence. It uses key issues modified by various experts as training statements to train the initial training model, thereby improving the efficiency and accuracy of data collection.

[0005] To achieve the aforementioned objective, this invention provides a key data extraction method based on artificial intelligence, comprising the following steps:

[0006] S01: Establish an artificial intelligence language model, pre-train the artificial intelligence language model using pre-training corpus, and obtain the initial training artificial intelligence language model when the artificial intelligence language model's processing of the pre-training corpus reaches the first model accuracy; then train the initial training artificial intelligence language model using fine-tuning corpus, and obtain the first trained artificial intelligence language model when the initial training artificial intelligence language model's processing of the fine-tuning corpus reaches the second model accuracy.

[0007] S02: Input the key text into the first trained artificial intelligence language model, and extract the key questions from the first trained artificial intelligence language model;

[0008] S03: Distribute multiple first key questions extracted by the first trained artificial intelligence language model to each member of the expert group. Each member of the expert group corrects the multiple first key questions to obtain multiple second key questions. The multiple second key questions are re-inputted into the first trained artificial intelligence language model for training. When the processing accuracy of the first trained artificial intelligence language model for key text reaches that of the third model, the second trained artificial intelligence language model is obtained.

[0009] S04: Represent the multiple third key questions extracted from the second trained artificial intelligence language model using vectors;

[0010] S05: Based on the third key question vector extracted from the second trained artificial intelligence language model, obtain the semantic information of the key questions and cluster the third key questions;

[0011] S06: Distribute the clustering results to each member of the expert group. Each member of the expert group corrects the third key question after clustering to obtain multiple fourth key questions. Then, re-input the multiple fourth key questions into the second trained artificial intelligence language model.

[0012] Repeat steps S04, S05, and S06 until the critical problem converges, all experts reach a consensus, and the converged critical problem is output as the final generated critical problem.

[0013] The present invention also provides a key data extraction system based on artificial intelligence, which includes a memory and one or more processors, wherein the memory is used to store programs written in a computer language, and the processors call the program code to implement the above-described method.

[0014] The present invention also provides a memory, characterized in that the memory is used to store a computer program compiled using a computer language to implement the methods described in the first to third embodiments of the present invention.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] This invention provides a key data extraction method, system, and storage device based on artificial intelligence. It involves establishing an artificial intelligence language model, pre-training the model using pre-training corpus, and obtaining an initial-trained artificial intelligence language model when the model's processing of the pre-training corpus reaches a first model accuracy. Then, the initial-trained artificial intelligence language model is trained using fine-tuned corpus, and a first-trained artificial intelligence language model is obtained when the model's processing of the fine-tuned corpus reaches a second model accuracy. Key text is input into the first-trained artificial intelligence language model, and key questions are extracted from it. Multiple first key questions extracted by the first-trained artificial intelligence language model are distributed to each member of an expert group. Each member of the expert group corrects these first key questions to obtain multiple second key questions. These second key questions are then re-inputted into the first-trained artificial intelligence language model for training. When the first-trained artificial intelligence language model's processing of the key text reaches a third model accuracy, a second-trained artificial intelligence language model is obtained. This method effectively improves the accuracy of the artificial intelligence language model in extracting key questions.

[0017] The process involves representing multiple third key questions extracted from a second trained AI language model using vectors; obtaining semantic information about the key questions based on these vectors; clustering the third key questions; distributing the clustering results to each member of the expert group, who then refines the clustered third key questions to obtain multiple fourth key questions, which are then re-inputted into the second trained AI language model; repeating these steps until the key questions converge, all experts reach a consensus, and the converged key questions are output. This approach aims to improve the scientific rigor of the key questions and increase the efficiency of data aggregation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for training a language model provided by the present invention;

[0019] Figure 2 This is a flowchart of a clustering method for addressing a key issue provided by the present invention. Detailed Implementation

[0020] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. "Several" means one or more, unless otherwise explicitly specified.

[0022] First Embodiment

[0023] Figure 1 This is a flowchart of the method for training a language model provided by the present invention; Figure 2 This is a flowchart of a clustering method for addressing a key issue provided by this invention, as shown below. Figure 1-2 As shown, the first embodiment of the present invention provides a key data extraction method based on artificial intelligence, which includes the following steps:

[0024] S01: Establish an artificial intelligence language model, pre-train the artificial intelligence language model using pre-training corpus, and obtain the initial trained artificial intelligence language model when the artificial intelligence language model's processing of the pre-training corpus reaches the first model accuracy; then train the initial trained artificial intelligence language model using fine-tuning corpus, and obtain the first trained artificial intelligence language model when the initial trained artificial intelligence language model's processing of the fine-tuning corpus reaches the second model accuracy.

[0025] S02: Input the special text into the first trained artificial intelligence language model, and extract the key questions from the first trained artificial intelligence language model;

[0026] S03: Distribute multiple first key questions extracted by the first trained artificial intelligence language model to each member of the expert group. Each member of the expert group corrects the multiple first key questions to obtain multiple second key questions. The multiple second key questions are re-inputted into the first trained artificial intelligence language model for training. When the processing accuracy of the first trained artificial intelligence language model for key text reaches that of the third model, the second trained artificial intelligence language model is obtained.

[0027] S04: Represent the multiple third key questions extracted from the second trained artificial intelligence language model using vectors;

[0028] S05: Based on the third key question vector extracted from the second trained artificial intelligence language model, obtain the semantic information of the key questions and cluster the third key questions;

[0029] S06: Distribute the clustering results to each member of the expert group. Each member of the expert group corrects the third key question after clustering to obtain multiple fourth key questions. Then, re-input the multiple fourth key questions into the second trained artificial intelligence language model.

[0030] Repeat steps S04, S05, and S06 until the critical problem converges, all experts reach a consensus, and the converged critical problem is output as the final generated critical problem.

[0031] In the first embodiment, the accuracy of the first model is evaluated using ambiguity Flcs:

[0032]

[0033] In the formula, LCS(X,Y) represents the length of the longest common order column (LCS) of the pre-training corpus sequence X and the processing result sequence Y of the AI ​​language model on the pre-training corpus, m represents the length of the processing result sequence Y of the AI ​​language model on the pre-training corpus sequence X, n represents the length of the pre-training corpus sequence X; β is a user-defined parameter, Rlcs For recall rate, P lcs For accuracy.

[0034] The accuracy of the second and third models is evaluated using a method similar to that of the first model. This will not be repeated here.

[0035] In the first embodiment, the construction of the pre-training corpus includes the following steps:

[0036] S02-1 divides the original text into sentences to obtain a set of n sentences, and the initial summary sentence set is empty;

[0037] S02-2 Takes a sentence from the original sentence set, merges it with the sentences in the summary sentence set, and calculates the common character sequence of the merged result and other sentences in the original sentence set;

[0038] For each sentence in the original sentence set, repeat the steps in S02-2, select the sentence with the longest common character sequence, add it to the summary sentence set, and remove it from the original sentence set;

[0039] S02-4 Repeat steps S02-2 and S02-3 until the number of sentences in the summary sentence set is greater than or equal to n / 4;

[0040] S02-5 merges all sentences in the original text sentence set and merges all sentences in the summary sentence set to form an "original text" - "summary" sentence pair, which serves as a pre-training data.

[0041] In the first embodiment, clustering the N third key issues specifically includes:

[0042] S3-1: Let there be N third key issues;

[0043] S3-2 will focus on one of the N third key questions related to equipment application through the hard attention mechanism model module;

[0044] S3-3: Learn the vector of a third key question into the two-dimensional intermediate layer of the self-competitive neural network;

[0045] S3-4 inputs the vectors of N-1 problems other than one of the N third key problems into the input layer of the self-competitive neural network in sequence. The self-competitive neural network clusters the N-1 problems other than one of the N third key problems. The k problems whose distance from the vector of the N-1 key problems other than one of the N third key problems to the vector of one problem in the intermediate layer is less than or equal to a set value are grouped into the same class as that one problem. The k key problems grouped into that one problem are removed from the N-1 problems to obtain Nk-1 problems that have not yet been clustered.

[0046] S3-5: Determine if Nk-1 is zero. If yes, end the classification; otherwise, proceed to step S3-6.

[0047] S3-6; N←Nk-1, and return to step S3-2.

[0048] Second Embodiment

[0049] The key data extraction method based on artificial intelligence provided in the second embodiment of the present invention will not be described again for the parts that are the same as those in the first embodiment; only the differences will be described. In the second embodiment, clustering of N third key issues specifically includes:

[0050] S4-1: Let there be N primary critical issues;

[0051] S4-2: Focus on k questions related to equipment application from multiple primary critical questions through a hard attention mechanism model module;

[0052] S4-3: Learn the vectors of k problems into the two-dimensional intermediate layer of the self-competitive neural network;

[0053] S4-4 inputs the vectors of Nk problems other than k problems out of N first key problems into the input layer of the self-competitive neural network in turn. The self-competitive neural network performs clustering. Problems whose distances from the vectors of the Nk first key problems to the vectors of the k problems in the intermediate layer are less than or equal to a set value are respectively classified into the same class of the k problems. Then, m problems that can be classified into the k problems are removed from the Nk problems to obtain Nkm problems that have not yet been clustered.

[0054] S4-5: Determine if Nkm is zero. If yes, end the classification; otherwise, proceed to step S4-6.

[0055] S4-6; N←Nkm, and return to step S4-2.

[0056] Third Embodiment

[0057] The key data extraction method based on artificial intelligence provided in the third embodiment of the present invention will not be described again for the parts that are the same as those in the first embodiment; only the differences will be described. In the third embodiment, clustering of N third key issues specifically includes:

[0058] S5-1 Randomly selects k third key questions as initial cluster centers;

[0059] S5-2 calculates the semantic similarity of each third key question to the k cluster centers and assigns it to the class corresponding to the cluster center with the highest semantic similarity.

[0060] S5-3 For each category, recalculate its cluster center by calculating the mean of the semantic vectors of all questions in each category, and use it as its new cluster center;

[0061] S5-4: Repeat steps S5-2 to S5-3 above until the problems in each category no longer change.

[0062] Fourth embodiment

[0063] The fourth embodiment of the present invention provides a key data extraction system based on artificial intelligence, which includes a processor, a ROM, a RAM and a storage medium. The processor performs various processes based on a program stored in the ROM or a program sent from the storage medium to the RAM. The program includes a computer program for implementing the methods of the first to third embodiments. The memory is used to store programs written in computer languages. The processor calls the program code to implement the method described in any of the first to third embodiments.

[0064] In the fourth embodiment, the processor has the following functions: encoding the data of a first unit to be written to the ROM to generate a first error correction code that together with the data of the first unit constitutes a first error correction code frame, and writing the data of the first unit to the ROM; when the first error correction code is written to the ROM or stored in RAM and the data of the first unit contained in n-1 error correction code frames is written to the ROM, encoding the data of the n-1 first units and a second data portion to be written to the ROM to generate a second error correction code that together with the data of the n-1 first units and the second unit data constitutes a second error correction code frame, and writing the second data and the second error correction code to the ROM, where n is an integer of 2 or more.

[0065] Fifth Embodiment: The fifth embodiment of the present invention provides a memory, characterized in that the memory is used to store a computer program compiled using a computer language to implement the methods described in the first to third embodiments of the present invention.

[0066] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A key data extraction method based on artificial intelligence, characterized in that, S01: Establish an artificial intelligence language model, pre-train the artificial intelligence language model using pre-training corpus, and obtain the initial training artificial intelligence language model when the artificial intelligence language model's processing of the pre-training corpus reaches the first model accuracy; then train the initial training artificial intelligence language model using fine-tuning corpus, and obtain the first trained artificial intelligence language model when the initial training artificial intelligence language model's processing of the fine-tuning corpus reaches the second model accuracy. S02: Input the special text into the first trained artificial intelligence language model, and extract multiple first key questions from the first trained artificial intelligence language model; S03: Distribute multiple first key questions extracted by the first trained artificial intelligence language model to each member of the expert group. Each member of the expert group corrects the multiple first key questions to obtain multiple second key questions. The multiple second key questions are re-inputted into the first trained artificial intelligence language model for training. When the first trained artificial intelligence language model's handling of key questions reaches the accuracy of the third model, the second trained artificial intelligence language model is obtained. S04: Represent the multiple third key questions extracted from the second trained artificial intelligence language model using vectors; S05: Based on the third key question vector extracted from the second trained artificial intelligence language model, obtain the semantic information of the key questions and cluster the third key questions; S06: Distribute the clustering results to each member of the expert group. Each member of the expert group corrects the third key question after clustering to obtain multiple fourth key questions. Then, re-input the multiple fourth key questions into the second trained artificial intelligence language model. Repeat steps S04, S05, and S06 until the critical problem converges, all experts reach a consensus, and the converged critical problem is output as the final generated critical problem.

2. The key data extraction method based on artificial intelligence according to claim 1, characterized in that, The construction of the pre-training corpus includes the following steps: S02-1: Segment the original text to obtain n sentences, and further obtain a set of original text sentences. Assume that the initial summary sentence set is empty. S02-2: Take a sentence from the original sentence set, merge it with the sentences in the summary sentence set, and calculate the common character sequence of the merged result and other sentences in the original sentence set; S02-3: For each sentence in the original sentence set, repeat the steps in S02-2, select the sentence with the longest common character sequence, add it to the summary sentence set, and remove it from the original sentence set; S02-4: Repeat steps S02-2 and S02-3 until the number of sentences in the summary sentence set is greater than or equal to n / 4; S02-5: Merge all sentences in the original text sentence set and merge all sentences in the summary sentence set to form an "original text" - "summary" sentence pair, which serves as a pre-training data.

3. The key data extraction method based on artificial intelligence according to claim 1, characterized in that, The accuracy of the first model is measured by ambiguity. Conduct an assessment: ， In the formula, , LCS(X,Y) represents the length of the longest common subsequence (LCS) between the pre-training corpus sequence X and the sequence Y resulting from the processing of the pre-training corpus by the AI ​​language model; m represents the length of the pre-training corpus sequence X, n represents the length of the sequence Y resulting from the processing of the pre-training corpus by the AI ​​language model; β is a user-defined parameter. For recall rate, For accuracy.

4. The key data extraction method based on artificial intelligence according to claim 1, characterized in that, Clustering of N third key questions specifically includes: S05-11: Let there be N third key issues; S05-12: Focus on one of the N third key questions related to equipment application through the hard attention mechanism model module; S05-13: Learn the vector of a third key question into a two-dimensional intermediate layer of a self-competitive neural network; S05-14: Input the vectors of N-1 problems other than one of the N third key problems into the input layer of the self-competitive neural network in sequence. The self-competitive neural network clusters the N-1 problems other than one of the N third key problems. The k problems whose distance from the vector of the N-1 key problems other than one of the N third key problems to the vector of one problem in the intermediate layer is less than or equal to a set value are grouped into the same class as that one problem. The k key problems grouped into that one problem are removed from the N-1 problems to obtain Nk-1 problems that have not yet been clustered. S05-15: Determine if Nk-1 is zero. If yes, end the clustering. If no, proceed to step S05-16. S05-16: N←Nk-1, and return to step S05-12.

5. The key data extraction method based on artificial intelligence according to claim 1, characterized in that, Clustering of N third key questions specifically includes: S05-21: Let there be N third key issues; S05-22: Focus on k questions related to equipment application among multiple third key questions through a hard attention mechanism model module; S05-23: Learn the vectors of k problems into the two-dimensional intermediate layer of a self-competitive neural network; S05-24: Input the vectors of Nk problems other than k problems out of N third key problems into the input layer of the self-competitive neural network in sequence. Cluster the Nk problems by performing clustering. Problems whose distance from the vectors of the Nk third key problems to the vectors of the k problems in the intermediate layer is less than or equal to a set value are respectively classified into the same class of the k problems. Remove the m problems that belong to the class of the above k problems from the remaining Nk problems to obtain Nkm problems that have not yet been clustered. S05-25: Determine if Nkm is zero. If yes, end the clustering; otherwise, proceed to step S05-26. S05-26: N←Nkm, and return to step S05-22.

6. The key data extraction method based on artificial intelligence according to claim 1, characterized in that, Clustering of N third key questions specifically includes: S05-31: Randomly select k third key questions as initial cluster centers; S05-32: For each third key question, calculate its semantic similarity to the k cluster centers and assign it to the class corresponding to the cluster center with the highest semantic similarity; S05-33: For each category, recalculate its cluster center by calculating the mean of the semantic vectors of all questions in each category, and use it as its new cluster center; S05-34: Repeat steps S05-32 to S05-33 above until the problems in each category no longer change.

7. A key data extraction system based on artificial intelligence, characterized in that, It includes a memory and one or more processors, the memory being used to store programs written in a computer language, the processors calling the program code to implement the key data extraction method based on artificial intelligence as described in any one of claims 1 to 6.

8. A memory, characterized in that, The memory is used to store a computer program that uses a computer language to compile the key data extraction method based on artificial intelligence as described in any one of claims 1 to 6.

Images