Auxiliary report rapid generation method and system based on language large model

Abstract

The invention discloses an auxiliary report rapid generation method and system based on a language large model. The method comprises the steps of data collection, data preprocessing, knowledge extraction, information compression and report optimization. The invention belongs to the technical field of large language models, and particularly relates to an auxiliary report rapid generation method and system based on a large language model.According to the scheme, knowledge extraction is adopted, centroids of individuals in a target domain are calculated, then the Euclidean distance from each individual in a source domain to the centroid of the target domain is calculated, and the difference degree between the individuals and the centroids is measured; performing knowledge transfer according to the similarity; according to the method, information compression is adopted, convolution and down-sampling are used for reducing data dimensions, output serves as LSTM layer input for extracting a long-term dependency relationship and processing reconstructed data, in the decoding process, deconvolution and up-sampling are used for generating output approximate to original input, and an automatic encoder is trained to minimize the difference between the input data and the reconstructed data.

Description

Technical Field

[0001] The present invention belongs to the technical field of language big models, and in particular to a method and system for quickly generating auxiliary reports based on language big models. Background Art

[0002] The rapid report generation method is a method that uses a large language model to learn and train a large amount of text data, so that the large language model can understand the input information and quickly generate report content according to the report requirements and report framework.

[0003] However, existing methods for rapid report generation have technical problems such as difficulty in accurately locating knowledge and low targeted knowledge screening, which leads to broad and vague report content and lack of accurate focus on key knowledge points, thus reducing the credibility and practicality of the report; there are also technical problems such as excessive data volume in large language models, low data processing efficiency, and inaccurate grasp of data characteristics, which leads to long model training time and incomplete generated reports. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for quickly generating auxiliary reports based on a large language model. In view of the technical problems that it is difficult to accurately locate existing knowledge, the knowledge screening is less targeted, resulting in the report content being relatively broad and vague, lacking accurate focus on key knowledge points, and reducing the credibility and practicality of the report, knowledge extraction is adopted to calculate the centroid of individuals in the target domain, and then calculate the Euclidean distance from each individual in the source domain to the centroid of the target domain, measure the degree of difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid, and perform more targeted knowledge extraction; in view of the technical problems that the large language model has too much data, low data processing efficiency, inaccurate grasp of data features, resulting in too long model training time, and incomplete generated reports, information compression is adopted, and convolution and downsampling are used to reduce the data dimension, and the output of the convolution layer is used as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data. The decoding process uses deconvolution and upsampling to generate output data that approximates the original input, and the automatic encoder is trained to minimize the difference between the input data and the reconstructed data.

[0005] The technical solution adopted by the present invention is as follows: The method for quickly generating auxiliary reports based on a large language model provided by the present invention comprises the following steps:

[0006] Step S1: Data collection, collecting a large amount of text data related to the report topic;

[0007] Step S2: Data preprocessing, cleaning and word segmentation operations are performed on the collected text data to obtain preprocessed text data;

[0008] Step S3: Knowledge extraction: Calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the degree of difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid, and perform more targeted knowledge extraction;

[0009] Step S4: Information compression, taking the extracted knowledge as the original input data, using convolution and downsampling operations to reduce the data dimension, and using the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data. The decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and trains the autoencoder to minimize the difference between the input data and the reconstructed data.

[0010] Step S5: Report optimization, specifically reviewing the report content to ensure that no key information is missed during knowledge extraction and information compression, and adjusting the language model parameters according to the report quality.

[0011] Furthermore, in step S1, the data collection specifically includes clarifying the subject and field of the required report, determining the data source, setting the frequency and scope of data collection, and extracting relevant text data from the data source according to the settings;

[0012] The data sources include academic databases, industry databases, news websites and document libraries.

[0013] Furthermore, in step S2, the data preprocessing includes the following steps:

[0014] Step S21: Perform data cleaning to remove noise, duplicate data and special characters;

[0015] Step S22: Determine the part of speech, divide the text into words, and determine the part of speech of each word;

[0016] Step S23: Perform entity recognition to identify specific entities in the text, including names of people, places, and organizations;

[0017] Step S24: Obtain preprocessed text data by executing steps S21 to S23.

[0018] Furthermore, in step S3, the knowledge extraction includes the following steps:

[0019] Step S31: Calculate the centroid of individuals in the target domain. When calculating the centroid of individuals based on the preprocessed text data, first vectorize the text data, use the vectorized representation of the text data as the individual, perform weighted averaging on the values ​​of the individual in each dimension, and determine the position of the centroid in the vector space. The formula used is as follows:

[0020] ;

[0021] In the formula, C represents the centroid of the individual in the target domain, It represents the number of solutions that meet the Pareto optimal condition and is used to determine the weight when calculating the centroid of the target domain individuals. j represents the index of the individual, and J represents the total number of individuals. represents the jth individual in the target domain;

[0022] Step S32: Calculate the Euclidean distance from each individual in the source domain to the centroid of the target domain. The source domain is the source of knowledge, containing the knowledge to be extracted and transferred, and the target domain is the receiver and application of knowledge. It is used to measure the degree of difference between the individual and the centroid. The formula used is as follows:

[0023] ;

[0024] In the formula, Ed represents the calculated Euclidean distance, i represents the dimension index, and n represents the total number of dimensions. represents the value of the jth individual in the source domain on the i-th dimension, Represents the value of the i-th dimension of the centroid of the target domain;

[0025] Step S33: define the weight associated with each source domain, extract a specified number of individuals from each source domain for knowledge transfer, and define the weight associated with each source domain taking into account the similarities between different source domains and the target domain. The formula used is as follows:

[0026] ;

[0027] ;

[0028] In the formula, represents the weight associated with the t-th source domain, t represents the index of the source domain, T represents the total number of source domains, represents the similarity measure of the t-th source domain, represents the source domain parameters, represents the target domain parameters, f() represents the transformation function that converts the Euclidean distance into a similarity measure;

[0029] Step S34: Calculate the source domain closest to the centroid, which is used to focus on the source domain closest to the centroid during knowledge extraction, and help screen out more important source domains for targeted knowledge extraction and analysis. The formula used is as follows:

[0030] ;

[0031] In the formula, represents the source domain closest to the centroid, and N represents the source domain eigenvalue.

[0032] Furthermore, in step S4, the information compression includes the following steps:

[0033] Step S41: Calculate the reconstructed data, take the feature information of the target domain and the source domain as the original input data, reduce the data dimension through convolution and downsampling layers, which helps to reduce the amount of data and extract the main features, and take the output of the convolution layer as the input of the LSTM layer. The LSTM layer can capture the long-term dependencies of the data and further reduce the dimension. At the same time, the gating mechanism controls the addition and deletion of information to achieve effective processing and reconstruction of the data, reduce the amount of data while retaining key information as much as possible. The formula used is as follows:

[0034] ;

[0035] In the formula, Represents the reconstructed data after LSTM processing, MaxPool represents the maximum pooling operation, Z represents the output data of the convolution layer, and s represents the size parameter of the pooling operation;

[0036] Step S42: Capturing key features of the data. In the encoder neural network, convolution and downsampling operations generate potential feature maps and reduce dimensionality, which helps to extract key features and improve the quality and efficiency of compression.

[0037] Step S43: Decoding process, using the reconstructed data as the input of the decoding process, using deconvolution operation and upsampling to generate an output close to the original input, showing the restoration process of the compressed data, verifying the effectiveness and accuracy of the compression algorithm, the formula used is as follows:

[0038] ;

[0039] ;

[0040] In the formula, Indicates the final output that is close to the original input. UpSample is the upsampling function. represents the output of the deconvolution function, represents the size of upsampling, which usually corresponds to the pooling size of downsampling, g() represents the activation function of deconvolution, represents the deconvolution weight, Represents the bias term of the deconvolution layer;

[0041] Step S44: train the autoencoder to minimize the difference between the input data and the reconstructed data and reduce the information loss. The root mean square error is used as the loss function to evaluate the information loss. The formula used is as follows:

[0042] ;

[0043] Where MSE represents the root mean square error, K represents the total number of data points, and k represents the index of the data point. represents the original input data corresponding to the kth data point, Represents the approximate output value of the kth data point after reconstruction.

[0044] Furthermore, in step S5, the report is optimized, specifically by carefully reviewing the report content to ensure that no key information is missed in the process of knowledge extraction and information compression, further refining the language expression of the report to make the report more concise, checking the logical structure of the report to ensure the quality and readability of the report, and adjusting the language model parameters according to the report quality to continuously improve the effect of rapid auxiliary report generation.

[0045] The auxiliary report rapid generation system based on the language large model provided by the present invention includes a data acquisition module, a data preprocessing module, a knowledge extraction module, an information compression module and a report optimization module;

[0046] The data collection specifically refers to collecting a large amount of text data related to the report topic;

[0047] The data preprocessing specifically includes cleaning and word segmentation operations on the collected text data to obtain preprocessed text data;

[0048] The knowledge extraction is specifically to calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid distance, and perform more targeted knowledge extraction;

[0049] The information compression specifically includes taking the extracted knowledge as the original input data, reducing the data dimension by using convolution and downsampling operations, taking the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data, the decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and training the autoencoder to minimize the difference between the input data and the reconstructed data;

[0050] The report optimization specifically includes reviewing the report content, ensuring that no key information is omitted during knowledge extraction and information compression, and adjusting the language model parameters according to the report quality.

[0051] The beneficial results achieved by the present invention using the above scheme are as follows:

[0052] (1) In order to solve the technical problems that existing knowledge is difficult to accurately locate and knowledge screening is less targeted, resulting in broad and vague report content, lack of accurate focus on key knowledge points, and reduced credibility and practicality of the report, knowledge extraction is used to calculate the centroid of individuals in the target domain, and then calculate the Euclidean distance from each individual in the source domain to the centroid of the target domain, measure the degree of difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid, and perform more targeted knowledge extraction;

[0053] (2) To address the technical issues of excessive data volume in large language models, low data processing efficiency, inaccurate grasp of data features, long model training time, and incomplete generated reports, information compression is used to reduce data dimensions using convolution and downsampling. The output of the convolution layer is used as the input of the LSTM layer to extract long-term dependencies and process reconstructed data. The decoding process uses deconvolution and upsampling to generate output data that approximates the original input, and the autoencoder is trained to minimize the difference between the input data and the reconstructed data. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of the process of the method for quickly generating auxiliary reports based on a large language model provided by the present invention;

[0055] Figure 2 A schematic diagram of the auxiliary report rapid generation system based on the language large model provided by the present invention;

[0056] Figure 3 is a schematic diagram of the process of step S3;

[0057] Figure 4 It is a schematic diagram of the process of step S4.

[0058] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0061] Example 1, see Figure 1 The invention provides a method for quickly generating an auxiliary report based on a large language model, the method comprising the following steps:

[0062] Step S1: Data collection, collecting a large amount of text data related to the report topic;

[0063] Step S2: Data preprocessing, cleaning and word segmentation operations are performed on the collected text data to obtain preprocessed text data;

[0064] Step S3: Knowledge extraction: Calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the degree of difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid, and perform more targeted knowledge extraction;

[0065] Step S4: Information compression, taking the extracted knowledge as the original input data, using convolution and downsampling operations to reduce the data dimension, and using the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data. The decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and trains the autoencoder to minimize the difference between the input data and the reconstructed data.

[0066] Step S5: Report optimization, review the report content to ensure that no key information is missed during knowledge extraction and information compression, and adjust the language model parameters according to the report quality.

[0067] Example 2, see Figure 1 ,This embodiment is based on the above embodiment. In step S1, the data collection specifically includes clarifying the subject and field of the required report, determining the data source, setting the frequency and scope of data collection, and extracting relevant text data from the data source according to the settings;

[0068] The data sources include academic databases, industry databases, news websites and document libraries.

[0069] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the data preprocessing includes the following steps:

[0070] Step S21: Perform data cleaning to remove noise, duplicate data and special characters;

[0071] Step S22: Determine the part of speech, divide the text into words, and determine the part of speech of each word;

[0072] Step S23: Perform entity recognition to identify specific entities in the text, including names of people, places, and organizations;

[0073] Step S24: Obtain preprocessed text data by executing steps S21 to S23.

[0074] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the knowledge extraction includes the following steps:

[0075] Step S31: Calculate the centroid of individuals in the target domain. When calculating the centroid of individuals based on the preprocessed text data, first vectorize the text data, use the vectorized representation of the text data as the individual, perform weighted averaging on the values ​​of the individual in each dimension, and determine the position of the centroid in the vector space. The formula used is as follows:

[0076] ;

[0077] In the formula, C represents the centroid of the individual in the target domain, It represents the number of solutions that meet the Pareto optimal condition and is used to determine the weight when calculating the centroid of the target domain individuals. j represents the index of the individual, and J represents the total number of individuals. represents the jth individual in the target domain;

[0078] Step S32: Calculate the Euclidean distance from each individual in the source domain to the centroid of the target domain. The source domain is the source of knowledge, containing the knowledge to be extracted and transferred, and the target domain is the receiver and application of knowledge. It is used to measure the degree of difference between the individual and the centroid. The formula used is as follows:

[0079] ;

[0080] In the formula, Ed represents the calculated Euclidean distance, i represents the dimension index, and n represents the total number of dimensions. represents the value of the jth individual in the source domain on the i-th dimension, Represents the value of the i-th dimension of the centroid of the target domain;

[0081] Step S33: define the weight associated with each source domain, extract a specified number of individuals from each source domain for knowledge transfer, and define the weight associated with each source domain taking into account the similarities between different source domains and the target domain. The formula used is as follows:

[0082] ;

[0083] ;

[0084] In the formula, represents the weight associated with the t-th source domain, t represents the index of the source domain, T represents the total number of source domains, represents the similarity measure of the t-th source domain, represents the source domain parameters, represents the target domain parameters, f() represents the transformation function that converts the Euclidean distance into a similarity measure;

[0085] Step S34: Calculate the source domain closest to the centroid, which is used to focus on the source domain closest to the centroid during knowledge extraction, and help screen out more important source domains for targeted knowledge extraction and analysis. The formula used is as follows:

[0086] ;

[0087] In the formula, represents the source domain closest to the centroid, and N represents the source domain eigenvalue.

[0088] By performing the above operations, knowledge extraction is used to calculate the centroid of individuals in the target domain, and then the Euclidean distance from each individual in the source domain to the centroid of the target domain is calculated to measure the degree of difference between the individual and the centroid. Individuals are extracted from the source domain for knowledge transfer, and the source domain closest to the centroid is calculated to perform more targeted knowledge extraction. This solves the technical problems of difficulty in accurately locating knowledge, low targeted knowledge screening, resulting in broad and vague report content, lack of accurate focus on key knowledge points, and reduced credibility and practicality of the report.

[0089] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the information compression includes the following steps:

[0090] Step S41: Calculate the reconstructed data, take the feature information of the target domain and the source domain as the original input data, reduce the data dimension through convolution and downsampling layers, which helps to reduce the amount of data and extract the main features, and take the output of the convolution layer as the input of the LSTM layer. The LSTM layer can capture the long-term dependencies of the data and further reduce the dimension. At the same time, the gating mechanism controls the addition and deletion of information to achieve effective processing and reconstruction of the data, reduce the amount of data while retaining key information as much as possible. The formula used is as follows:

[0091] ;

[0092] In the formula, Represents the reconstructed data after LSTM processing, MaxPool represents the maximum pooling operation, Z represents the output data of the convolution layer, and s represents the size parameter of the pooling operation;

[0093] Step S42: Capturing key features of the data. In the encoder neural network, convolution and downsampling operations generate potential feature maps and reduce dimensionality, which helps to extract key features and improve the quality and efficiency of compression.

[0094] Step S43: Decoding process, using the reconstructed data as the input of the decoding process, using deconvolution operation and upsampling to generate an output close to the original input, showing the restoration process of the compressed data, verifying the effectiveness and accuracy of the compression algorithm, the formula used is as follows:

[0095] ;

[0096] ;

[0097] In the formula, Indicates the final output that is close to the original input. UpSample is the upsampling function. represents the output of the deconvolution function, represents the size of upsampling, which usually corresponds to the pooling size of downsampling, g() represents the activation function of deconvolution, represents the deconvolution weight, Represents the bias term of the deconvolution layer;

[0098] Step S44: train the autoencoder to minimize the difference between the input data and the reconstructed data and reduce the information loss. The root mean square error is used as the loss function to evaluate the information loss. The formula used is as follows:

[0099] ;

[0100] Where MSE represents the root mean square error, K represents the total number of data points, and k represents the index of the data point. represents the original input data corresponding to the kth data point, Represents the approximate output value of the kth data point after reconstruction.

[0101] By performing the above operations, information compression is adopted, convolution and downsampling are used to reduce the data dimension, and the output of the convolution layer is used as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data. The decoding process uses deconvolution and upsampling to generate output data that is close to the original input. The autoencoder is trained to minimize the difference between the input data and the reconstructed data. This solves the technical problems of excessive data volume in large language models, low data processing efficiency, inaccurate grasp of data features, long model training time, and incomplete generated reports.

[0102] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the report optimization is specifically to conduct a detailed review of the report content to ensure that no key information is omitted in the process of knowledge extraction and information compression, further refine the language expression of the report to make the report more concise, check the logical structure of the report to ensure the quality and readability of the report, adjust the language model parameters according to the report quality, and continuously improve the effect of rapid generation of auxiliary reports.

[0103] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, the language model-based auxiliary report rapid generation system provided by the present invention includes a data acquisition module, a data preprocessing module, a knowledge extraction module, an information compression module and a report optimization module;

[0104] The data collection specifically refers to collecting a large amount of text data related to the report topic;

[0105] The data preprocessing specifically includes cleaning and word segmentation operations on the collected text data to obtain preprocessed text data;

[0106] The knowledge extraction is specifically to calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid distance, and perform more targeted knowledge extraction;

[0107] The information compression specifically includes taking the extracted knowledge as the original input data, reducing the data dimension by using convolution and downsampling operations, taking the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data, the decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and training the autoencoder to minimize the difference between the input data and the reconstructed data;

[0108] The report optimization specifically includes reviewing the report content, ensuring that no key information is omitted during knowledge extraction and information compression, and adjusting the language model parameters according to the report quality.

[0109] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0110] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0111] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for quickly generating auxiliary reports based on a large language model, characterized by: The method comprises the following steps: Step S1: Data collection, collecting a large amount of text data related to the report topic; Step S2: Data preprocessing, cleaning and word segmentation operations are performed on the collected text data to obtain preprocessed text data; Step S3: Knowledge extraction: Calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the degree of difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid, and perform more targeted knowledge extraction; Step S4: Information compression, using the extracted knowledge as the original input data, using convolution and downsampling operations to reduce the data dimension, and using the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data. The decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and trains the autoencoder to minimize the difference between the input and reconstructed data; Step S5: Report optimization, specifically reviewing the report content to ensure that no key information is missed during knowledge extraction and information compression, and adjusting the language model parameters according to the report quality.

2. The method for quickly generating auxiliary reports based on a large language model according to claim 1 is characterized in that: In step S3, the knowledge extraction includes the following steps: Step S31: Calculate the centroid of individuals in the target domain. When calculating the centroid of individuals based on the preprocessed text data, first vectorize the text data, use the vectorized representation of the text data as the individual, perform weighted averaging on the values ​​of the individual in each dimension, and determine the position of the centroid in the vector space. The formula used is as follows: ； In the formula, C represents the centroid of the individual in the target domain, It represents the number of solutions that meet the Pareto optimal condition and is used to determine the weight when calculating the centroid of the target domain individuals. j represents the index of the individual, and J represents the total number of individuals. represents the jth individual in the target domain; Step S32: Calculate the Euclidean distance from each individual in the source domain to the centroid of the target domain. The formula used is as follows: ； In the formula, Ed represents the calculated Euclidean distance, i represents the dimension index, and n represents the total number of dimensions. represents the value of the jth individual in the source domain on the i-th dimension, Represents the value of the i-th dimension of the centroid of the target domain; Step S33: define the weight associated with each source domain, extract individuals from each source domain for knowledge transfer, and define the weight associated with each source domain taking into account the similarities between different source domains and the target domain. The formula used is as follows: ； ； In the formula, represents the weight associated with the t-th source domain, t represents the index of the source domain, T represents the total number of source domains, represents the similarity measure of the t-th source domain, represents the source domain parameters, represents the target domain parameters, f() represents the transformation function that converts the Euclidean distance into a similarity measure; Step S34: Calculate the source domain closest to the centroid, which is used to focus on the source domain closest to the centroid during knowledge extraction, and select more important source domains for targeted knowledge extraction and analysis. The formula used is as follows: ； In the formula, represents the source domain closest to the centroid, and N represents the source domain eigenvalue.

3. The method for quickly generating auxiliary reports based on a large language model according to claim 1 is characterized in that: In step S4, the information is compressed, including the following steps: Step S41: Calculate and reconstruct the data. The extracted knowledge is used as the original input data. The data dimension is reduced through convolution and downsampling layers. The output of the convolution layer is used as the input of the LSTM layer. The LSTM layer can capture the long-term dependencies of the data. At the same time, the gating mechanism controls the addition and deletion of information to achieve effective processing and reconstruction of the data. The formula used is as follows: ； In the formula, Represents the reconstructed data after LSTM processing, MaxPool represents the maximum pooling operation, Z represents the output data of the convolution layer, and s represents the size parameter of the pooling operation; Step S42: capturing key features of data; Step S43: Decoding process, using the reconstructed data as the input of the decoding process, using deconvolution operation and upsampling to generate an output that is close to the original input, the formula used is as follows: ； ； In the formula, Indicates the final output that is close to the original input. UpSample is the upsampling function. represents the output of the deconvolution function, represents the upsampling size, g() represents the activation function of deconvolution, represents the deconvolution weight, Represents the bias term of the deconvolution layer; Step S44: train the autoencoder to minimize the difference between the input data and the reconstructed data, and use the root mean square error as the loss function to evaluate the information loss. The formula used is as follows: ； Where MSE represents the root mean square error, K represents the total number of data points, and k represents the index of the data point. represents the original input data corresponding to the kth data point, Represents the approximate output value of the kth data point after reconstruction.

4. The method for quickly generating auxiliary reports based on a large language model according to claim 1 is characterized in that: In step S1, the data collection specifically includes clarifying the subject and field of the required report, determining the data source, setting the frequency and scope of data collection, and extracting text data from the data source according to the settings; The data sources include academic databases, industry databases, news websites and document libraries.

5. The method for quickly generating auxiliary reports based on a large language model according to claim 1 is characterized in that: In step S2, the data preprocessing includes the following steps: Step S21: Perform data cleaning to remove noise, duplicate data and special characters; Step S22: Determine the part of speech, divide the text into words, and determine the part of speech of each word; Step S23: Perform entity recognition to identify specific entities in the text, including names of people, places, and organizations; Step S24: Obtain preprocessed text data by executing steps S21 to S23.

6. The method for quickly generating auxiliary reports based on a large language model according to claim 1 is characterized in that: In step S5, the report optimization is specifically to conduct a detailed review of the report content to ensure that no key information is missed in the process of knowledge extraction and information compression, further refine the language expression of the report to make the report more concise, check the logical structure of the report, adjust the language model parameters according to the report quality, and continuously improve the effect of rapid generation of auxiliary reports.

7. A system for quickly generating auxiliary reports based on a large language model, used to implement a method for quickly generating auxiliary reports based on a large language model as claimed in any one of claims 1 to 6, characterized in that: It includes data acquisition module, data preprocessing module, knowledge extraction module, information compression module and report optimization module.

8. The system for quickly generating auxiliary reports based on a large language model according to claim 7 is characterized in that: The data collection specifically refers to collecting a large amount of text data related to the report topic; The data preprocessing specifically includes cleaning and word segmentation operations on the collected text data to obtain preprocessed text data; The knowledge extraction is specifically to calculate the centroid of individuals in the target domain based on the preprocessed text data, then calculate the Euclidean distance between each individual in the source domain and the centroid of the target domain, measure the difference between the individual and the centroid, extract individuals from the source domain for knowledge transfer, calculate the source domain closest to the centroid distance, and perform more targeted knowledge extraction; The information compression specifically includes taking the extracted knowledge as the original input data, reducing the data dimension by using convolution and downsampling operations, taking the output of the convolution layer as the input of the LSTM layer to extract long-term dependencies and process the reconstructed data, the decoding process uses deconvolution and upsampling to generate an output that approximates the original input, and training the autoencoder to minimize the difference between the input data and the reconstructed data; The report optimization specifically includes reviewing the report content, ensuring that no key information is omitted during knowledge extraction and information compression, and adjusting the language model parameters according to the report quality.

Images