Water affairs expert question-answering assistance method and system based on water affairs big model

Abstract

The present invention relates to the field of water affairs technology, and provides a water affairs expert question-answering assistance method and system based on a water affairs big model, including: classifying water affairs problems to obtain water affairs problem texts and water affairs problem images, respectively performing problem screening on water affairs problem texts and water affairs problem images to obtain target problem texts and target problem images; extracting features from problems in a water affairs problem set to obtain problem features; calculating the feature correlation strength between problem features, setting the problem priority corresponding to the target problem text, mining the problem information corresponding to the target problem text, and generating a problem identifier corresponding to the target problem text; retrieving the problem analysis corresponding to the water affairs problem set from a preset water affairs big model, calibrating the problem analysis to obtain the calibrated analysis, and generating a question-answering assistance report corresponding to the water affairs expert. The main purpose of the present invention is to improve the processing efficiency of water affairs expert question-answering assistance.

Description

Technical Field

[0001] The present invention relates to the field of water technology, and in particular to a water expert question-answering assistance method and system based on a water large model. Background Art

[0002] With the acceleration of urbanization and population growth, water issues are receiving increasing attention. In the field of water affairs, the knowledge and experience of experts are crucial to solving problems and providing decision support. However, due to time and resource constraints, water experts cannot answer every question in person. Therefore, an effective water expert question-and-answer assistance method is needed to help users quickly obtain accurate answers.

[0003] The existing water expert question-and-answer assistance currently adopts the method of building an online knowledge base. This method is to establish a knowledge base containing knowledge in the water field, and retrieve relevant answers from the knowledge base by analyzing and matching user questions. However, the knowledge and information of the online knowledge base in this method may not be comprehensive and accurate enough to meet the needs of users, and the corresponding importance of different questions cannot be identified, resulting in some urgent questions not being answered in time. Therefore, a more effective water expert question-and-answer assistance method is needed to help users quickly obtain accurate answers. Summary of the invention

[0004] The present invention provides a water affairs expert question and answer assistance method and system based on a water affairs big model, the main purpose of which is to improve the processing efficiency of water affairs expert question and answer assistance.

[0005] To achieve the above purpose, the present invention provides a water affairs expert question-answering assistance method based on a water affairs big model, comprising:

[0006] Collect water affairs problems raised by water affairs experts, classify the water affairs problems to obtain water affairs problem texts and water affairs problem images, and perform problem screening on the water affairs problem texts and the water affairs problem images to obtain target problem texts and target problem images;

[0007] Filtering a visual feature map from the target problem image, extracting a water affairs descriptor in the visual feature map, matching a descriptor text corresponding to the water affairs descriptor, and determining an image problem text corresponding to the target problem image according to the descriptor text;

[0008] Combining the image problem text and the target problem text, generating a water problem set corresponding to the water problem, performing feature extraction on the problems in the water problem set, and obtaining problem features;

[0009] Calculating the feature correlation strength between the question features, setting the question priority corresponding to the target question text according to the feature correlation strength, mining the question information corresponding to the target question text, and generating the question identifier corresponding to the target question text according to the question information;

[0010] In combination with the problem priority and the problem identifier, the problem analysis corresponding to the water problem set is retrieved from the preset water problem big model, the problem analysis is calibrated to obtain a calibrated analysis, and based on the calibrated analysis, a question-and-answer auxiliary report corresponding to the water expert is generated.

[0011] Optionally, the performing problem screening processing on the water problem text and the water problem image respectively to obtain a target problem text and a target problem image includes:

[0012] Performing semantic analysis on the water affairs problem text to obtain the problem semantics;

[0013] According to the semantics of the problem, identifying the text stop words in the water problem text, and calculating the image overlap between the water problem images;

[0014] In combination with the stop words in the text and the overlap degree of the image, the water problem text and the water problem image are respectively subjected to problem screening processing to obtain a target problem text and a target problem image.

[0015] Optionally, calculating the image overlap between the water problem images includes:

[0016] Performing noise reduction processing on the water affairs problem image to obtain a noise-reduced problem image;

[0017] Performing grayscale processing on the noise reduction problem image to obtain a grayscale problem image;

[0018] Calculate the image hash value corresponding to the grayscale problem image;

[0019] The image overlap between the water problem images is calculated according to the image hash values.

[0020] Optionally, calculating the image overlap between the water problem images according to the image hash values ​​includes:

[0021] The number of pixel rows and pixel columns corresponding to the water problem image is counted, and the image overlap between the water problem images is calculated by combining the image hash value, the number of pixel rows and the number of pixel columns using the following formula:

[0022] ;

[0023] Among them, A represents the image overlap between water problem images, The coordinates of one pixel in the water problem image are ( ) corresponding to the image hash value, The pixel coordinates of another image between the water problem images are The corresponding image hash value, B and D represent the number of pixel rows and pixel columns respectively. The equality operator, ( )and represents the image pixels at the same position in the two water problem images, Indicates the number of times the same image hash value was calculated.

[0024] Optionally, the step of filtering out a visual feature map from the target problem image includes:

[0025] Dispatching historical water affairs data from a database storing the water affairs issues, and analyzing water affairs elements corresponding to the historical water affairs data;

[0026] Detecting water conservancy elements in the target problem image, and performing vectorization processing on the water conservancy elements and the water conservancy elements respectively to obtain element vectors and element vectors;

[0027] Calculating the vector similarity between the feature vector and the element vector;

[0028] Combined with the vector similarity, a visual feature map is screened out from the target problem image.

[0029] Optionally, the calculating the vector similarity between the feature vector and the element vector includes:

[0030] The element vector and the element vector are respectively subjected to dimensionality reduction processing to obtain a reduced-dimensionality element element and a reduced-dimensionality element vector, and the reduced-dimensionality element element and the reduced-dimensionality element vector are respectively subjected to normalization processing to obtain a target element vector and a target element vector, and the vector similarity between the target element vector and the target element vector is calculated by the following formula:

[0031] ;

[0032] Among them, E represents the vector similarity between the target feature vector and the target element vector, represents the i-th vector in the target feature vector, represents the jth vector in the target element vector, i and j represent the sequence numbers corresponding to the target feature vector and the target element vector respectively, and They represent the norms corresponding to the i-th vector in the target feature vector and the j-th vector in the target element vector, respectively.

[0033] Optionally, extracting the water affairs descriptor from the visual feature map includes:

[0034] Performing image cropping processing on the visual feature map to obtain a cropped feature map;

[0035] Extracting water service features from the cropped feature image, extracting water service labels corresponding to the water service features, and calculating label gain values ​​corresponding to the water service labels;

[0036] According to the tag gain value, feature selection processing is performed on the water service feature to obtain a target water service feature;

[0037] A water service descriptor in the visual feature map is generated according to the target water service feature.

[0038] Optionally, extracting features of the problems in the water affairs problem set to obtain problem features includes:

[0039] Performing word segmentation processing on each question in the water affairs question set to obtain question word segmentations;

[0040] Calculating the word segmentation weight corresponding to the problem word segmentation, and extracting the problem keywords of each problem in the water affairs problem set from the problem word segmentation according to the word segmentation weight;

[0041] Analyze the keyword attributes corresponding to the question keywords, and extract the attribute features corresponding to the keyword attributes;

[0042] According to the attribute characteristics, the problem characteristics of each problem in the water problem set are obtained.

[0043] Optionally, the calculating the feature correlation strength between the problem features includes:

[0044] Constructing a feature vector corresponding to the problem feature, calculating a feature value corresponding to the problem feature based on the feature vector, and measuring the feature quantity of the problem feature;

[0045] Calculate the feature mean corresponding to the problem feature according to the feature value and the feature quantity;

[0046] Combining the feature value, the feature mean and the feature quantity, the feature correlation strength between the problem features is calculated using the following formula:

[0047] ;

[0048] Among them, T represents the feature correlation strength between problem features, and They represent the ath feature and a+1th feature in the problem feature, a and a+1 represent the serial numbers corresponding to the problem feature, and p represents the number of features. represents the feature mean.

[0049] A water affairs expert question-answering assistance system based on a water affairs big model, characterized in that the system comprises:

[0050] A problem screening module is used to collect water problems raised by water experts, classify the water problems, obtain water problem texts and water problem images, and perform problem screening on the water problem texts and the water problem images to obtain target problem texts and target problem images;

[0051] An image processing module, used to filter out a visual feature map from the target problem image, extract a water affairs descriptor in the visual feature map, match a descriptor text corresponding to the water affairs descriptor, and determine an image problem text corresponding to the target problem image based on the descriptor text;

[0052] A feature extraction module, used to combine the image problem text and the target problem text to generate a water problem set corresponding to the water problem, and extract features of the problems in the water problem set to obtain problem features;

[0053] A priority setting module, used to calculate the feature correlation strength between the question features, set the question priority corresponding to the target question text according to the feature correlation strength, mine the question information corresponding to the target question text, and generate the question identifier corresponding to the target question text according to the question information;

[0054] A report generation module is used to retrieve the problem analysis corresponding to the water problem set from a preset water problem model in combination with the problem priority and the problem identifier, calibrate the problem analysis to obtain a calibrated analysis, and generate a question-and-answer auxiliary report corresponding to the water expert based on the calibrated analysis.

[0055] The present invention can remove redundant information in the water problem text and the water problem image by performing problem screening processing on the water problem text and the water problem image respectively, so as to improve the quality of water problems and reduce subsequent problem retrieval processing. The present invention can obtain a representative image in the target problem image by screening a visual feature map from the target problem image, which provides a basis for determining the image problem text corresponding to the target problem image in the subsequent process. The present invention can obtain representative attributes of each problem in the water problem set by performing feature extraction on the problems in the water problem set, which provides a basis for setting the problem priority corresponding to the target problem text in the subsequent process. The present invention can understand the feature correlation degree of the problem features by calculating the feature correlation strength between the problem features, and then judge the corresponding importance of the problem features, so as to improve the accuracy of setting the subsequent problem priority. The present invention retrieves the problem analysis corresponding to the water problem set from a preset water problem big model by combining the problem priority and the problem identifier, so as to improve the retrieval efficiency of the water problem set, and can improve the accuracy of the problem analysis by calibrating the problem analysis. Therefore, the water affairs expert question and answer assistance method and system based on the water affairs big model provided in the embodiment of the present invention can improve the processing efficiency of water affairs expert question and answer assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flowchart of a water affairs expert question-and-answer assistance method based on a water affairs big model provided by an embodiment of the present invention;

[0057] Figure 2 A functional module diagram of a water affairs expert question and answer assistance system based on a water affairs big model provided in one embodiment of the present invention.

[0058] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0059] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0060] The embodiment of the present application provides a water affairs expert question and answer assistance method based on a water affairs big model. In the embodiment of the present application, the execution subject of the water affairs expert question and answer assistance method based on the water affairs big model includes but is not limited to at least one of the electronic devices such as the server and the terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the water affairs expert question and answer assistance method based on the water affairs big model can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0061] Reference Figure 1 FIG. 1 is a flow chart of a water affairs expert question-answering assistance method based on a water affairs big model provided by an embodiment of the present invention. In this embodiment, the water affairs expert question-answering assistance method based on a water affairs big model includes steps S1-S5.

[0062] S1. Collect water problems raised by water experts, classify the water problems to obtain water problem texts and water problem images, and perform problem screening on the water problem texts and the water problem images to obtain target problem texts and target problem images.

[0063] The present invention can remove redundant information in the water problem text and the water problem image by performing problem screening processing on the water problem text and the water problem image respectively, so as to improve the quality of water problems and reduce subsequent problem retrieval processing, wherein the water problem is a series of problems related to water resource management, supply, distribution, use and protection, such as water pollution problems, the water problem text is a text-type problem in the water problem, and the water problem image is an image-type problem in the water problem, such as a user taking an image of a water area of ​​a part of the area, and raising questions based on the water area image, or an image corresponding to a corresponding problem is marked in the water area image. Furthermore, the classification processing of the water problem can be achieved through a decision tree algorithm.

[0064] As an embodiment of the present invention, the step of performing problem screening processing on the water problem text and the water problem image to obtain a target problem text and a target problem image includes:

[0065] Performing semantic analysis on the water affairs problem text to obtain the problem semantics;

[0066] According to the semantics of the problem, identifying the text stop words in the water problem text, and calculating the image overlap between the water problem images;

[0067] In combination with the stop words in the text and the overlap degree of the image, the water problem text and the water problem image are respectively subjected to problem screening processing to obtain a target problem text and a target problem image.

[0068] Among them, the problem semantics is the corresponding meaning of the water problem text, the image overlap represents the degree of overlap between the water problem images, and further, the semantic parsing processing of the water problem text can be achieved through semantic analysis; the text stop words in the water problem text can be achieved through OCR recognition technology; the problem screening processing of the water problem text and the water problem image can be achieved by deleting the text stop words and comparing the image overlap with the corresponding preset threshold value. When the image overlap is greater than the preset threshold value, the problem image is retained, and the one that does not meet the condition is deleted, so as to obtain the target problem text and target problem image.

[0069] Optionally, as an optional embodiment of the present invention, the calculating the image overlap between the water problem images includes: performing noise reduction processing on the water problem images to obtain noise-reduced problem images;

[0070] Performing grayscale processing on the noise reduction problem image to obtain a grayscale problem image;

[0071] Calculate the image hash value corresponding to the grayscale problem image;

[0072] The image overlap between the water problem images is calculated according to the image hash values.

[0073] Among them, the denoised problem image is an image obtained after the noise in the water problem image is removed, and the image hash value is a fixed-length digital sequence or character string obtained by converting the grayscale problem image through a specific algorithm. Furthermore, the denoising processing of the water problem image can be achieved through a low-pass filter; the grayscale processing of the denoised problem image can be achieved through the average method; the RGB value of each pixel of the denoised problem image is averaged as the grayscale value; the calculation of the image hash value corresponding to the grayscale problem image can be achieved through the average hash algorithm, the average grayscale value of the image is calculated, the grayscale problem image is binarized, and the image hash value is obtained.

[0074] Optionally, as an optional embodiment of the present invention, calculating the image overlap between the water problem images according to the image hash values ​​includes:

[0075] The number of pixel rows and pixel columns corresponding to the water problem image is counted, and the image overlap between the water problem images is calculated by combining the image hash value, the number of pixel rows and the number of pixel columns using the following formula:

[0076] ;

[0077] Among them, A represents the image overlap between water problem images, The coordinates of one pixel in the water problem image are ( ) corresponding to the image hash value, The pixel coordinates of another image between the water problem images are The corresponding image hash value, B and D represent the number of pixel rows and pixel columns respectively. The equality operator, ( )and represents the image pixels at the same position in the two water problem images, Indicates the number of times the same image hash value was calculated.

[0078] in, It is used to compare whether two hash values ​​hash(i, j) and hash(i', j') are equal.

[0079] S2. Filter out a visual feature map from the target problem image, extract a water affairs descriptor from the visual feature map, match a descriptor text corresponding to the water affairs descriptor, and determine an image problem text corresponding to the target problem image based on the descriptor text.

[0080] The present invention can obtain a representative image in the target problem image by screening out a visual feature map from the target problem image, thereby providing a basis for subsequently determining the image problem text corresponding to the target problem image, wherein the visual feature map is a representative image in the target problem image.

[0081] As an embodiment of the present invention, the step of selecting a visual feature map from the target problem image includes:

[0082] Dispatching historical water affairs data from a database storing the water affairs issues, and analyzing water affairs elements corresponding to the historical water affairs data;

[0083] Detecting water conservancy elements in the target problem image, and performing vectorization processing on the water conservancy elements and the water conservancy elements respectively to obtain element vectors and element vectors;

[0084] Calculating the vector similarity between the feature vector and the element vector;

[0085] Combined with the vector similarity, a visual feature map is screened out from the target problem image.

[0086] Among them, the historical water affairs data are historical water affairs related data recorded in the database, such as water consumption data, the water affairs elements are factors corresponding to the historical water affairs data, such as water quality, the water affairs elements are elements or objects related to water affairs in the target problem image, the element vector and the element vector are respectively the expression vectors corresponding to the water affairs elements and the water affairs elements, and the vector similarity represents the degree of similarity between the element vector and the element vector.

[0087] Optionally, scheduling historical water data from a database storing the water problem can be achieved through a scheduling function, and the scheduling function is compiled by a programming language; the water elements corresponding to the historical water data can be implemented through principal component analysis; the detection of water elements in the target problem image can be implemented through a target detection algorithm, such as an R-CNN algorithm; the water elements and the vectorization processing of the water elements can be implemented through a One-hot encoding method; the vector similarity is compared with a similarity threshold, and if the vector similarity is greater than the similarity threshold, a visual feature map is screened out from the target problem image.

[0088] Optionally, as an optional embodiment of the present invention, the calculating the vector similarity between the feature vector and the element vector includes:

[0089] The element vector and the element vector are respectively subjected to dimensionality reduction processing to obtain a reduced-dimensionality element element and a reduced-dimensionality element vector, and the reduced-dimensionality element element and the reduced-dimensionality element vector are respectively subjected to normalization processing to obtain a target element vector and a target element vector, and the vector similarity between the target element vector and the target element vector is calculated by the following formula:

[0090] ;

[0091] Among them, E represents the vector similarity between the target feature vector and the target element vector, represents the i-th vector in the target feature vector, represents the jth vector in the target element vector, i and j represent the sequence numbers corresponding to the target feature vector and the target element vector respectively, and They represent the norms corresponding to the i-th vector in the target feature vector and the j-th vector in the target element vector, respectively.

[0092] The present invention can obtain the water affairs features in the visual feature map by extracting the water affairs descriptor in the visual feature map, and can further convert the water affairs descriptor into the corresponding text by matching the descriptor text corresponding to the water affairs descriptor, so as to facilitate the subsequent conversion of the target problem image into the corresponding image problem text, wherein the water affairs descriptor is a water affairs-related feature in the visual feature map, and the descriptor text is an expression character corresponding to the water affairs descriptor. Furthermore, the descriptor text matching corresponding to the water affairs descriptor can be achieved through a text generation model, such as an ELMO model.

[0093] As an embodiment of the present invention, extracting the water descriptor from the visual feature map includes:

[0094] Performing image cropping processing on the visual feature map to obtain a cropped feature map;

[0095] Extracting water service features from the cropped feature image, extracting water service labels corresponding to the water service features, and calculating label gain values ​​corresponding to the water service labels;

[0096] According to the tag gain value, feature selection processing is performed on the water service feature to obtain a target water service feature;

[0097] A water service descriptor in the visual feature map is generated according to the target water service feature.

[0098] Among them, the cropped feature map is an image obtained by cropping the image unrelated to water affairs in the visual feature map, the water affairs feature is a feature related to water affairs in the cropped feature map, the water affairs label is the annotation information corresponding to the water affairs feature, the label gain value indicates the importance corresponding to the water affairs label, and the target water affairs feature is an important feature among the water affairs features.

[0099] Furthermore, the image cropping processing of the visual feature map can be implemented through OpenCV functions; the extraction of water service features in the cropped feature map can be implemented through corresponding extraction methods, for example, using edge detection, texture analysis and other technologies to extract corresponding water service features; the water service labels corresponding to the water service features can be implemented through convolutional neural networks; the calculation of the label gain values ​​corresponding to the water service labels can be implemented through the Shannon entropy algorithm; the feature selection processing of the water service features can be implemented by combining the numerical values ​​of the label gain values, and the features with the highest numerical values ​​in the label gain values ​​are screened out from the water service features of the same type, so as to obtain the target water service features; the target water service features are grouped together to generate the water service descriptors in the visual feature map.

[0100] S3. Combining the image problem text and the target problem text, generating a water problem set corresponding to the water problem, performing feature extraction on the problems in the water problem set, and obtaining problem features.

[0101] The present invention can obtain representative attributes of each problem in the water problem set by extracting features from the problems in the water problem set, thereby providing a basis for setting the problem priority corresponding to the target problem text in the subsequent process; wherein the water problem set is a collection of problems generated by combining the image problem text and the target problem text.

[0102] As an embodiment of the present invention, the problem feature extraction of the problem in the water problem set to obtain the problem feature includes:

[0103] Performing word segmentation processing on each question in the water affairs question set to obtain question word segmentations;

[0104] Calculating the word segmentation weight corresponding to the problem word segmentation, and extracting the problem keywords of each problem in the water affairs problem set from the problem word segmentation according to the word segmentation weight;

[0105] Analyze the keyword attributes corresponding to the question keywords, and extract the attribute features corresponding to the keyword attributes;

[0106] According to the attribute characteristics, the problem characteristics of each problem in the water problem set are obtained.

[0107] Among them, the problem segmentation is the vocabulary corresponding to the text characters of each problem in the water problem set, the segmentation weight represents the importance corresponding to the problem segmentation, the problem keyword is an important segmentation in the problem segmentation, the keyword attribute is the keyword property corresponding to the problem keyword, such as type, part of speech, semantics, etc., the attribute feature is a specific feature or description corresponding to the keyword attribute, which can be a certain aspect, property, feature or other information related to the keyword. Optionally, the segmentation processing of each problem in the water problem set can be implemented by an N-gram algorithm; the segmentation weight corresponding to the text segmentation can be implemented by a weight calculator, and the weight calculator is compiled by a programming language; the problem keyword of each problem in the water problem set can be compared by combining the segmentation weight and a preset weight value, and the problem keyword is extracted from the problem segmentation according to the comparison result; the analysis of the keyword attribute corresponding to the problem keyword can be implemented by a context analysis method, and its attribute is inferred according to the location of the problem keyword. The extraction of attribute features corresponding to the keyword attribute can be implemented by a neural network model, such as a generative adversarial network.

[0108] S4. Calculate the feature correlation strength between the question features, set the question priority corresponding to the target question text according to the feature correlation strength, mine the question information corresponding to the target question text, and generate the question identifier corresponding to the target question text according to the question information.

[0109] By calculating the feature association strength between the problem features, the present invention can understand the feature association degree of the problem features, and then judge the importance corresponding to the problem features, so as to improve the accuracy of setting subsequent problem priorities, wherein the problem priority represents the processing priority corresponding to the target problem text, and further, the feature association strength is standardized, the feature association strength is mapped to the same range, and then the problem priority corresponding to the target problem text is set according to the mapped association strength.

[0110] As an embodiment of the present invention, the calculation of the feature association strength between the problem features includes: constructing a feature vector corresponding to the problem feature, calculating the feature value corresponding to the problem feature based on the feature vector, and measuring the feature quantity of the problem feature, calculating the feature mean corresponding to the problem feature based on the feature value and the feature quantity, and calculating the feature association strength between the problem features by combining the feature value, the feature mean and the feature quantity.

[0111] Among them, the feature vector is the expression vector corresponding to the problem feature, and the feature value is the expression value corresponding to the problem feature. Furthermore, the construction of the feature vector corresponding to the problem feature can be achieved through the above-mentioned One-hot encoding method; the calculation of the feature value corresponding to the problem feature can be achieved through the QR algorithm; the calculation of the feature mean corresponding to the problem feature can be achieved by the average function.

[0112] Optionally, as an optional embodiment of the present invention, the combining the feature value, the feature mean and the feature quantity to calculate the feature correlation strength between the problem features includes:

[0113] Constructing a feature vector corresponding to the problem feature, calculating a feature value corresponding to the problem feature based on the feature vector, and measuring the feature quantity of the problem feature;

[0114] Calculate the feature mean corresponding to the problem feature according to the feature value and the feature quantity;

[0115] Combining the feature value, the feature mean and the feature quantity, the feature correlation strength between the problem features is calculated using the following formula:

[0116] ;

[0117] Among them, T represents the feature correlation strength between problem features, and They represent the ath feature and a+1th feature in the problem feature, a and a+1 represent the serial numbers corresponding to the problem feature, and p represents the number of features. represents the feature mean.

[0118] The present invention can obtain the core content of the target question text by mining the question information corresponding to the target question text, and generate a question identifier corresponding to the target question text based on the question information, so as to obtain a unique identifier corresponding to the target question text, thereby improving the retrieval accuracy of subsequent question analysis. Furthermore, the mining of the question information corresponding to the target question text can be achieved through an information mining algorithm, such as a decision tree algorithm; the question identifier corresponding to the target question text can be generated based on the feature information in the question information.

[0119] S5. In combination with the problem priority and the problem identifier, the problem analysis corresponding to the water problem set is retrieved from the preset water problem big model, the problem analysis is calibrated to obtain a calibrated analysis, and based on the calibrated analysis, a question-and-answer auxiliary report corresponding to the water expert is generated.

[0120] The present invention combines the problem priority and the problem identifier to retrieve the problem analysis corresponding to the water problem set from a preset water problem big model to improve the retrieval efficiency of the water problem set, and calibrates the problem analysis to further improve the accuracy of the problem analysis, wherein the water problem big model is a model constructed by collecting a large number of water-related problems and corresponding answers, and the problem analysis is the solution or answer corresponding to the water problem set. Furthermore, retrieving the problem analysis corresponding to the water problem set from the preset water problem big model can be achieved through a retrieval algorithm, and the retrieval algorithm includes a parallel retrieval algorithm; the calibration processing of the problem analysis can be performed by associating the corresponding problem text, and the relevant problem answers in the problem analysis are replaced and correspondingly calculated by the calibration model with the factors in the problem text to obtain a calibrated analysis, wherein the calibration model includes a support vector machine model.

[0121] The present invention can remove redundant information in the water problem text and the water problem image by performing problem screening processing on the water problem text and the water problem image respectively, so as to improve the quality of water problems and reduce subsequent problem retrieval processing. The present invention can obtain a representative image in the target problem image by screening a visual feature map from the target problem image, which provides a basis for determining the image problem text corresponding to the target problem image in the subsequent process. The present invention can obtain representative attributes of each problem in the water problem set by performing feature extraction on the problems in the water problem set, which provides a basis for setting the problem priority corresponding to the target problem text in the subsequent process. The present invention can understand the feature correlation degree of the problem features by calculating the feature correlation strength between the problem features, and then judge the corresponding importance of the problem features, so as to improve the accuracy of setting the subsequent problem priority. The present invention retrieves the problem analysis corresponding to the water problem set from a preset water problem big model by combining the problem priority and the problem identifier, so as to improve the retrieval efficiency of the water problem set, and can improve the accuracy of the problem analysis by calibrating the problem analysis. Therefore, the water affairs expert question and answer assistance method based on the water affairs big model provided in the embodiment of the present invention can improve the processing efficiency of water affairs expert question and answer assistance.

[0122] like Figure 2 The figure shows a functional module diagram of a water affairs expert question-and-answer assistance system based on a water affairs big model provided by an embodiment of the present invention.

[0123] The water affairs expert question and answer assistance system 100 based on the water affairs big model described in the present invention can be installed in an electronic device. According to the functions implemented, the water affairs expert question and answer assistance system 100 based on the water affairs big model can include a question screening module 101, an image processing module 102, a feature extraction module 103, a priority setting module 104 and a report generation module 105. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0124] In this embodiment, the functions of each module / unit are as follows:

[0125] The problem screening module 101 is used to collect water problems raised by water experts, classify the water problems to obtain water problem texts and water problem images, and perform problem screening on the water problem texts and water problem images to obtain target problem texts and target problem images;

[0126] The image processing module 102 is used to filter out a visual feature map from the target problem image, extract a water affairs descriptor in the visual feature map, match a descriptor text corresponding to the water affairs descriptor, and determine an image problem text corresponding to the target problem image according to the descriptor text;

[0127] The feature extraction module 103 is used to combine the image problem text and the target problem text to generate a water problem set corresponding to the water problem, and extract features of the problems in the water problem set to obtain problem features;

[0128] The priority setting module 104 is used to calculate the feature correlation strength between the question features, set the question priority corresponding to the target question text according to the feature correlation strength, mine the question information corresponding to the target question text, and generate the question identifier corresponding to the target question text according to the question information;

[0129] The report generation module 105 is used to combine the problem priority and the problem identifier, retrieve the problem analysis corresponding to the water problem set from the preset water problem big model, calibrate the problem analysis to obtain a calibrated analysis, and generate a question-and-answer auxiliary report corresponding to the water expert based on the calibrated analysis.

[0130] In detail, each module described in the water affairs expert question-answering auxiliary system 100 based on the water affairs big model described in the embodiment of the present application is used in the same manner as described above. Figure 1 The technical means described in the text is the same as the water expert question and answer auxiliary method based on the water affairs big model, and can produce the same technical effects, so I will not go into details here.

[0131] In the several embodiments provided by the present invention, it should be understood that the provided methods and systems can be implemented in other ways. For example, the method embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A water affairs expert question-answering assistance method based on a water affairs big model, characterized in that: The method comprises: Collect water affairs problems raised by water affairs experts, classify the water affairs problems to obtain water affairs problem texts and water affairs problem images, and perform problem screening on the water affairs problem texts and the water affairs problem images to obtain target problem texts and target problem images; Filtering a visual feature map from the target problem image, extracting a water affairs descriptor in the visual feature map, matching a descriptor text corresponding to the water affairs descriptor, and determining an image problem text corresponding to the target problem image according to the descriptor text; Combining the image problem text and the target problem text, generating a water problem set corresponding to the water problem, performing feature extraction on the problems in the water problem set, and obtaining problem features; Calculating the feature correlation strength between the question features, setting the question priority corresponding to the target question text according to the feature correlation strength, mining the question information corresponding to the target question text, and generating the question identifier corresponding to the target question text according to the question information; In combination with the problem priority and the problem identifier, the problem analysis corresponding to the water problem set is retrieved from the preset water affairs big model, the problem analysis is calibrated to obtain a calibrated analysis, and based on the calibrated analysis, a question-and-answer auxiliary report corresponding to the water affairs expert is generated.

2. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 1 is characterized in that: The step of respectively performing problem screening processing on the water affairs problem text and the water affairs problem image to obtain a target problem text and a target problem image includes: Performing semantic analysis on the water affairs problem text to obtain the problem semantics; According to the semantics of the problem, identifying the text stop words in the water problem text, and calculating the image overlap between the water problem images; In combination with the stop words in the text and the overlap degree of the image, the water problem text and the water problem image are respectively subjected to problem screening processing to obtain a target problem text and a target problem image.

3. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 2 is characterized in that: The calculating the image overlap between the water problem images includes: Performing noise reduction processing on the water affairs problem image to obtain a noise-reduced problem image; Performing grayscale processing on the noise reduction problem image to obtain a grayscale problem image; Calculate the image hash value corresponding to the grayscale problem image; The image overlap between the water problem images is calculated according to the image hash values.

4. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 3 is characterized in that: The calculating the image overlap between the water problem images according to the image hash values ​​includes: The number of pixel rows and pixel columns corresponding to the water problem image is counted, and the image overlap between the water problem images is calculated by combining the image hash value, the number of pixel rows and the number of pixel columns using the following formula: ； Among them, A represents the image overlap between water problem images, The coordinates of one pixel in the water problem image are ( ) corresponding to the image hash value, The pixel coordinates of another image between the water problem images are The corresponding image hash value, B and D represent the number of pixel rows and pixel columns respectively. The equality operator, ( )and represents the image pixels at the same position in the two water problem images, Indicates the number of identical image hash values ​​calculated.

5. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 4 is characterized in that: The step of selecting a visual feature map from the target problem image includes: Dispatching historical water affairs data from a database storing the water affairs issues, and analyzing water affairs elements corresponding to the historical water affairs data; Detecting water conservancy elements in the target problem image, and performing vectorization processing on the water conservancy elements and the water conservancy elements respectively to obtain element vectors and element vectors; Calculating the vector similarity between the feature vector and the element vector; Combined with the vector similarity, a visual feature map is screened out from the target problem image.

6. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 5 is characterized in that: The calculating the vector similarity between the feature vector and the element vector comprises: The element vector and the element vector are respectively subjected to dimensionality reduction processing to obtain a reduced-dimensionality element element and a reduced-dimensionality element vector, and the reduced-dimensionality element element and the reduced-dimensionality element vector are respectively subjected to normalization processing to obtain a target element vector and a target element vector, and the vector similarity between the target element vector and the target element vector is calculated by the following formula: ； Among them, E represents the vector similarity between the target feature vector and the target element vector, represents the i-th vector in the target feature vector, represents the jth vector in the target element vector, i and j represent the sequence numbers corresponding to the target feature vector and the target element vector respectively, and They represent the norms corresponding to the i-th vector in the target feature vector and the j-th vector in the target element vector, respectively.

7. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 6 is characterized in that: The extracting the water affairs descriptor from the visual feature map comprises: Performing image cropping processing on the visual feature map to obtain a cropped feature map; Extracting water service features from the cropped feature image, extracting water service labels corresponding to the water service features, and calculating label gain values ​​corresponding to the water service labels; According to the tag gain value, feature selection processing is performed on the water service feature to obtain a target water service feature; A water service descriptor in the visual feature map is generated according to the target water service feature.

8. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 7 is characterized in that: The problem features in the water problem set are extracted to obtain problem features, including: Performing word segmentation processing on each question in the water affairs question set to obtain question word segmentations; Calculating the word segmentation weight corresponding to the problem word segmentation, and extracting the problem keywords of each problem in the water affairs problem set from the problem word segmentation according to the word segmentation weight; Analyze the keyword attributes corresponding to the question keywords, and extract the attribute features corresponding to the keyword attributes; According to the attribute characteristics, the problem characteristics of each problem in the water problem set are obtained.

9. The water affairs expert question-answering assistance method based on the water affairs big model as claimed in claim 8, characterized in that: The calculating the feature correlation strength between the problem features includes: Constructing a feature vector corresponding to the problem feature, calculating a feature value corresponding to the problem feature based on the feature vector, and measuring the feature quantity of the problem feature; Calculate the feature mean corresponding to the problem feature according to the feature value and the feature quantity; Combining the feature value, the feature mean and the feature quantity, the feature correlation strength between the problem features is calculated using the following formula: ； Among them, T represents the feature correlation strength between problem features, and They represent the ath feature and a+1th feature in the problem feature, a and a+1 represent the serial numbers corresponding to the problem feature, and p represents the number of features. represents the feature mean.

10. A water affairs expert question-answering assistance system based on a water affairs big model, characterized in that: The system comprises: A problem screening module is used to collect water problems raised by water experts, classify the water problems, obtain water problem texts and water problem images, and perform problem screening on the water problem texts and the water problem images to obtain target problem texts and target problem images; An image processing module, used to filter out a visual feature map from the target problem image, extract a water affairs descriptor in the visual feature map, match a descriptor text corresponding to the water affairs descriptor, and determine an image problem text corresponding to the target problem image based on the descriptor text; A feature extraction module, used to combine the image problem text and the target problem text to generate a water problem set corresponding to the water problem, and extract features of the problems in the water problem set to obtain problem features; A priority setting module, used to calculate the feature correlation strength between the question features, set the question priority corresponding to the target question text according to the feature correlation strength, mine the question information corresponding to the target question text, and generate the question identifier corresponding to the target question text according to the question information; A report generation module is used to retrieve the problem analysis corresponding to the water problem set from a preset water problem model in combination with the problem priority and the problem identifier, calibrate the problem analysis to obtain a calibrated analysis, and generate a question-and-answer auxiliary report corresponding to the water expert based on the calibrated analysis.

Images