Knowledge base construction method and system based on standard digital processing

Abstract

The invention discloses a knowledge base construction method and system based on standard digital processing, and belongs to the technical field of digital processing, and the method comprises the steps: based on an image recognition model, carrying out the picture selection and classification of text data, selecting or generating a standard template, and carrying out the numbering and marking; based on a semantic analysis model, converting the selected pictures into text data, and extracting key information in the pictures; performing master-slave association on the title space and the filling space, and expanding the keywords; on the basis of a semantic processing model, a keyword classification unit is used for screening keywords, grammar checking and semantic understanding are carried out on text data, and the text data are stored in a knowledge base; and continuously updating and optimizing the knowledge base. The efficiency and accuracy of extracting, sorting and classifying information from original text and image data are improved, and the richness and correlation of information in a knowledge base are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of digital processing technology, and in particular to a knowledge base construction method and system based on standard digital processing. Background Art

[0002] In today's digital age, the efficiency of knowledge management and information processing is crucial. Traditional knowledge base construction methods often rely on manual input and organization, which is not only time-consuming and labor-intensive, but also prone to errors. Therefore, it is particularly important to develop an automated knowledge base construction method based on advanced technology.

[0003] By combining image processing and natural language processing technologies, it is possible to extract and organize information from various text and image materials. First, image recognition technology is used to select and classify images of text materials, and then semantic analysis is used to convert images into text data to extract key information. Then, the information is effectively organized and associated through the tagging and sorting unit, and the extracted information is further screened and sorted using the keyword classification unit.

[0004] This method is very useful in various scenarios, especially in the fields of research, business, and education where information needs to be extracted and organized quickly from large amounts of data. Through automation and digital processing, it not only improves work efficiency, but also reduces human errors and improves the quality of data processing. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: how to improve the efficiency and accuracy of extracting, organizing and classifying information from original text and image materials, and how to enhance the richness and relevance of information in the knowledge base.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions, a method for constructing a knowledge base based on standard digital processing, comprising:

[0008] Based on the image recognition model, the frame selection unit 3 is used to select and classify pictures of text materials, select or generate standard templates, and mark them with numbers; based on the semantic analysis model, the data processing unit 4 is used to convert the selected pictures into text data, and extract key information from the pictures; using the tag sorting unit 5, the title space and the filling space are associated with each other, and the keywords are expanded; based on the semantic processing model, the keyword classification unit 6 is used to screen keywords, and the text data is grammatically checked and semantically understood, and stored in the knowledge base; the knowledge base is continuously updated, and the text data is optimized based on feedback.

[0009] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, the image selection and classification of text data includes setting a standard template by using the frame selection unit 3, and mapping the standard template on the shooting section, generating a shooting frame, reserving a title space and a filling space, using the title space and the filling space to compare and correct the text data, obtaining a plurality of groups of filling texts on the standard text, and after image selection of the filling texts, the selected images are numbered and marked in sequence;

[0010] Use the image recognition model to classify the selected images, and automatically select the appropriate standard template or generate a new template according to the category of the image. The steps are as follows:

[0011] The cropped image is input into the image recognition model, and after being processed by the convolution layer and the pooling layer, the feature vector in the image is extracted; the feature vector is input into the fully connected layer, and after being processed by the activation function and the softmax function, the category probability distribution of the image is output;

[0012] The image recognition model is expressed as:

[0013]

[0014] Among them, x (l) i, j, k represent the input feature map of layer l, w (l) m, n, p, k represent the convolution kernel parameters of the lth layer, represents the bias parameter of the lth layer, represents the convolution output of the lth layer, f represents the activation function, represents the activation output of layer l, represents the pooling output of the lth layer, M, N, P represent the size of the convolution kernel, and max represents the maximum value operation;

[0015] According to the probability distribution, the category with the highest probability is selected as the recognition result of the image, and a standard template matching the image category is selected from the preset standard template library;

[0016] If there is no standard template that matches the image category, or the image category is unclear, the image template generation model is used to migrate the image and generate a new template.

[0017] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, the picture template generation model is expressed as:

[0018] G(z)=CNN(z)

[0019] D(x)=CNN(x)

[0020]

[0021] Where z represents a random noise vector, x represents a real image, G(z) represents an image generated by the generator, D(x) represents the probability of the discriminator distinguishing the real image, D(G(z)) represents the probability of the discriminator distinguishing the generated image, and CNN represents a convolutional neural network. represents the loss function of the generator and the discriminator, represents the expected value, p data (x) represents the distribution of real images, p z (z) represents the distribution of noise vector;

[0022] The steps of the image template generation model include inputting the image into the generator, processing it through the convolution layer and the deconvolution layer, and generating a new template;

[0023] The new template and the preset standard template are input into the discriminator together. After being processed by the convolutional layer and the fully connected layer, the true and false probability distribution of the image is output.

[0024] According to the probability distribution, the loss functions of the generator and the discriminator are calculated, and the parameters of the generator and the discriminator are updated using the back-propagation algorithm, so that the generator can generate more realistic pictures and the discriminator can better distinguish between real and fake pictures;

[0025] Until the generator and the discriminator reach a balance, that is, the pictures generated by the generator can deceive the discriminator, and the discriminator cannot distinguish between real and fake pictures; the pictures generated by the generator are used as new templates for frame selection screenshots.

[0026] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, wherein: the extracting key information from the image includes using a data processing unit (4) to convert the selected image into text data, and filling the text data into the corresponding filling space in sequence according to the number mark to obtain the text data;

[0027] Using the semantic analysis model, we can perform semantic analysis on the selected images and extract the key information in the images. This information is used as a supplement to the text data and expressed as:

[0028] h (l) =f(W (l) x (l-1) +b (l) )

[0029]

[0030]

[0031] Among them, x (l) represents the input feature map of the lth layer, h(l) represents the hidden feature map of layer l, represents the output feature map of the lth layer, W (l) represents the weight parameter of the lth layer, b (l) represents the bias parameter of the lth layer, f represents the activation function, x represents the pixel category distribution of the real picture, Represents the pixel category distribution of the predicted image, represents the loss function of semantic segmentation, log represents logarithmic operation, and ∑ represents summation operation;

[0032] The semantic analysis step includes inputting the cropped image into the SSN model, extracting the high-dimensional feature vector in the image after being processed by the encoder; inputting the feature vector into the decoder, restoring the original size of the image after being processed by the deconvolution layer and the upsampling layer, and outputting the pixel category distribution of the image; segmenting the image according to the pixel category distribution, dividing the image into several regions, each region corresponding to a semantic category, and performing semantic analysis on each region according to the segmentation results to extract the key information in the region as a supplement to the text data.

[0033] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, wherein: the master-slave association of the title space and the filling space includes using the tag sorting unit 5 to perform master-slave association on the title space and the filling space, establishing a master-slave association between the numbered tag of the selected picture corresponding to the filling space and the title space, and filling the tag name in the title space as a keyword according to the standard template;

[0034] Using the knowledge graph, keywords are expanded, including using keywords as query conditions, retrieving entities and relationships related to the keywords from GDB, and constructing subgraphs. Subgraphs are graphs composed of several nodes and edges, where nodes represent entities and edges represent relationships. Subgraphs are analyzed to extract important information from the subgraphs, which are used as supplements to keywords. Subgraphs are expanded, and more related entities and relationships are retrieved from GDB based on the entities and relationships in the subgraphs, generating more subgraphs, and using the entities and relationships in these subgraphs as related vocabulary as retrieval conditions for the knowledge base.

[0035] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, wherein: the keyword screening includes using the keyword classification unit 6 to screen the keywords, obtain the text data corresponding to the selected picture under the corresponding subordinate number mark, and sort the text data according to the time sequence to generate time-text data, and store it in the knowledge base 2 for storage;

[0036] The semantic processing model is used to perform grammar checking and semantic understanding on the text data, and to correct or prompt the errors or unclear expressions in the text data, which can be expressed as:

[0037]

[0038] in, represents the input feature vector of the lth layer, represents the forward hidden feature vector of the lth layer, represents the backward hidden feature vector of the lth layer, represents the bidirectional hidden feature vector of the lth layer, y t represents the output feature vector, W y represents the weight parameter, b y represents the bias parameter, f represents the activation function, y represents the actual grammatical and semantic labels, represents the predicted grammatical and semantic labels, represents the loss function of natural language processing, log represents logarithmic operation, ∑ represents summation operation, [;] represents vector concatenation operation;

[0039] The steps of semantic processing include segmenting text data, converting each word into a word vector, inputting the word vector into a BiRNN model, extracting context feature vectors from the text data through forward RNN and backward RNN processing; inputting the feature vector into a fully connected layer, processing the activation function and the softmax function, outputting grammatical and semantic labels of the text data, and performing grammatical checking and semantic understanding on the text data based on the grammatical and semantic labels, and correcting or prompting errors or unclear expressions in the text data.

[0040] As a preferred solution of the knowledge base construction method based on standard digital processing described in the present invention, the optimization of text data according to feedback includes taking the knowledge base as the environment, the optimizer as the intelligent agent, defining the state, action and reward of the optimizer, the state is the text data in the knowledge base, the action is to optimize the text data, and the reward is the user's feedback and evaluation; and the learning of the optimizer is realized by using a deep Q network.

[0041] Another object of the present invention is to provide a system for constructing a knowledge base based on standard digital processing, which can improve the efficiency and accuracy of extracting, organizing and classifying information from original text and image materials through knowledge graphs and deep learning, and how to enhance the richness and relevance of information in the knowledge base.

[0042] In order to solve the above technical problems, the present invention provides the following technical solutions: a knowledge base construction system based on standard digital processing, comprising a text conversion system 1 and a knowledge base 2;

[0043] The text conversion system 1 includes a frame selection unit 3, a data processing unit 4, a tag arrangement unit 5 and a keyword classification unit 6, wherein the frame selection unit 3 is connected to the data processing unit 4 and the tag arrangement unit 5 respectively, and the data processing unit 4 and the tag arrangement unit 5 are both connected to the keyword classification unit 6;

[0044] The frame selection unit 3 includes a template reference module 7, a comparison correction module 8 and a frame selection label module 9. The template reference module 7 is connected to the comparison correction module 8, and the comparison correction module 8 is connected to the frame selection label module 9.

[0045] The template reference module 7 is used to set a standard template and map the standard template on the shooting section to generate a shooting frame and reserve a title space and a filling space;

[0046] The comparison and correction module 8 is used to compare and correct the text data using the title space and the filling space, and obtain several groups of filling texts on the standard text;

[0047] The frame selection and numbering module 9 is used to sequentially number the selected pictures after selecting the pictures for the filling text;

[0048] The data processing unit 4 includes an optical character recognition module 10 and a comparison and filling module 11, and the optical character recognition module 10 is connected to the comparison and filling module 11;

[0049] The optical character recognition module 10 is used to convert the selected image into text data;

[0050] The comparison and filling module 11 is used to fill the text data into the corresponding filling space in sequence according to the number mark to obtain the text data;

[0051] The tag arrangement unit 5 includes a module number comparison module 12 and a tag definition module 13, and the module number comparison module 12 is connected to the tag definition module 13;

[0052] The keyword classification unit 6 includes a keyword screening module 14 and a time sorting module 15, and the keyword screening module 14 is connected to the time sorting module 15;

[0053] The keyword screening module 14 is used to screen keywords and obtain text data corresponding to the selected pictures marked with corresponding subordinate numbers;

[0054] The time sorting module 15 is used to sort the text data according to the time sequence, generate time-text data, and transmit the time-text data to the knowledge base 2.

[0055] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the processor implements the steps of a knowledge base construction method based on standard digital processing.

[0056] A computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the computer program implements the steps of a knowledge base construction method based on standard digital processing.

[0057] The beneficial effects of the present invention are as follows: through efficient image recognition, in-depth semantic analysis, precise data processing and intelligent feedback learning mechanism, a dynamic, accurate and user-friendly digital knowledge base is constructed, which not only speeds up the information processing speed, but also improves the quality and practicality of the knowledge base. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0059] Figure 1 A flowchart of a method for constructing a knowledge base based on standard digital processing is provided as an embodiment of the present invention.

[0060] Figure 2 A system principle block diagram of a knowledge base construction system based on standard digital processing provided by an embodiment of the present invention.

[0061] Figure 3 A system principle block diagram of a frame selection unit of a knowledge base construction system based on standard digital processing provided by an embodiment of the present invention.

[0062] Figure 4 A system principle block diagram of a data processing unit of a knowledge base construction system based on standard digital processing provided by an embodiment of the present invention.

[0063] Figure 5 A system principle block diagram of a marking and sorting unit of a knowledge base construction system based on standard digital processing provided by an embodiment of the present invention.

[0064] Figure 6A system principle block diagram of a keyword classification unit of a knowledge base construction system based on standard digital processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0066] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0067] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0068] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0069] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the 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 cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0070] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0071] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a knowledge base construction method based on standard digital processing, including:

[0072] S1: Based on the image recognition model, use the frame selection unit 3 to select and classify the text data, select or generate a standard template, and mark it with numbers;

[0073] Furthermore, the image selection and classification of the text data includes setting a standard template using the frame selection unit 3, and mapping the standard template on the shooting section, generating a shooting frame, reserving a title space and a filling space, using the title space and the filling space to compare and correct the text data, obtaining a plurality of groups of filling texts on the standard text, and after image selection of the filling texts, the selected images are numbered and marked in sequence;

[0074] Use image recognition models to classify the screenshots, automatically select appropriate standard templates based on the categories of the pictures, or generate new templates to improve the efficiency and accuracy of frame selection screenshots;

[0075] It should be noted that the image recognition model uses a convolutional neural network (CNN) to extract and classify images. CNN is a deep learning model consisting of multiple convolutional layers, pooling layers, and fully connected layers. It can automatically learn local and global features in images to achieve efficient recognition of images. The steps of image recognition technology are:

[0076] The cropped image is input into the image recognition model, and after being processed by the convolution layer and the pooling layer, the feature vector in the image is extracted;

[0077] The feature vector is input into the fully connected layer, and after being processed by the activation function and the softmax function, the category probability distribution of the image is output;

[0078] The image recognition model is expressed as:

[0079]

[0080] Among them, x (l) i, j, k represent the input feature map of layer l, w (l) m, n, p, k represent the convolution kernel parameters of the lth layer, represents the bias parameter of the lth layer, represents the convolution output of the lth layer, f represents the activation function, represents the activation output of layer l, represents the pooling output of the lth layer, M, N, P represent the size of the convolution kernel, and max represents the maximum value operation.

[0081] According to the probability distribution, select the category with the highest probability as the recognition result of the image, and select a standard template that matches the image category from the preset standard template library, such as "bar graph" or "table";

[0082] If there is no standard template that matches the image category, or the image category is unclear, the image template generation model is used to transfer the image style and generate a new template;

[0083] Furthermore, the image template generation model is expressed as:

[0084] G(z)=CNN(z)

[0085] D(x)=CNN(x)

[0086]

[0087] Where z represents a random noise vector, x represents a real image, G(z) represents an image generated by the generator, D(x) represents the probability of the discriminator distinguishing the real image, D(G(z)) represents the probability of the discriminator distinguishing the generated image, and CNN represents a convolutional neural network. represents the loss function of the generator and the discriminator, represents the expected value, p data (x) represents the distribution of real images, p z (z) represents the distribution of the noise vector.

[0088] It should be noted that the image template generation model consists of a generator and a discriminator, which can achieve high-quality generation of images;

[0089] The steps of the image template generation model include inputting the image into the generator, processing it through the convolution layer and the deconvolution layer, and generating a new image, i.e., a new template;

[0090] The new image and the preset standard template are input into the discriminator. After being processed by the convolutional layer and the fully connected layer, the true and false probability distribution of the image is output.

[0091] According to the probability distribution, the loss functions of the generator and the discriminator are calculated, and the parameters of the generator and the discriminator are updated using the back-propagation algorithm, so that the generator can generate more realistic pictures and the discriminator can better distinguish between real and fake pictures;

[0092] Until the generator and the discriminator reach a balance, that is, the pictures generated by the generator can deceive the discriminator, and the discriminator cannot distinguish between real and fake pictures; the pictures generated by the generator are used as new templates for frame selection screenshots.

[0093] S2: Based on the semantic analysis model, the data processing unit 4 is used to convert the selected image into text data and extract key information from the image;

[0094] Furthermore, the extracting key information from the image includes converting the selected image into text data by using the data processing unit 4, and filling the text data into the corresponding filling space in sequence according to the number mark to obtain the text data;

[0095] Use the semantic analysis model to perform semantic analysis on the selected images and extract key information in the images, such as objects, colors, shapes, locations, etc. This information is used as a supplement to the text data to increase the richness and readability of the text data.

[0096] The semantic analysis model is expressed as:

[0097] h (l) =f(W (l) x (l-1) +b (l) )

[0098]

[0099] Among them, x (l) represents the input feature map of the lth layer, h (l) represents the hidden feature map of layer l, represents the output feature map of the lth layer, W (l) represents the weight parameter of the lth layer, b (l) represents the bias parameter of the lth layer, f represents the activation function, x represents the pixel category distribution of the real picture, Represents the pixel category distribution of the predicted image, represents the loss function of semantic segmentation, log represents logarithmic operation, and ∑ represents summation operation;

[0100] It should be noted that the principle of the semantic analysis model is to use the semantic segmentation network (SSN) to classify images at the pixel level. SSN is a deep learning model consisting of an encoder and a decoder that can automatically learn the semantic information in the image and achieve high-precision segmentation of the image.

[0101] The steps of semantic analysis are: input the cropped image into the SSN model, extract the high-dimensional feature vector in the image after being processed by the encoder; input the feature vector into the decoder, restore the original size of the image after being processed by the deconvolution layer and the upsampling layer, and output the pixel category distribution of the image; segment the image according to the pixel category distribution, divide the image into several regions, each region corresponds to a semantic category, such as "person", "car", "tree", etc.; according to the segmentation results, perform semantic analysis on each region, extract key information in the region, such as the name, color, shape, location, etc. of the object, and use this information as a supplement to the text data to increase the richness and readability of the text data;

[0102] S3: using the tag arrangement unit 5, the title space and the filling space are associated with each other in a master-slave relationship, and the keywords are expanded;

[0103] Furthermore, the master-slave association of the title space and the filling space includes using the tag sorting unit 5 to perform master-slave association on the title space and the filling space, establishing a master-slave association between the number tag of the selected picture corresponding to the filling space and the title space, and filling the tag name in the title space according to the standard template as a keyword;

[0104] Use knowledge graphs to expand keywords;

[0105] It should be noted that according to the relationship between keywords and other concepts, more relevant words are generated as search conditions of the knowledge base to improve the coverage and search efficiency of the knowledge base;

[0106] The principle of knowledge graph technology is to use graph database GDB to store and manage knowledge. GDB is a non-relational database that represents data in a graph structure. It can effectively represent complex entities and relationships and realize efficient retrieval and reasoning of knowledge. The steps are as follows:

[0107] Use keywords as query conditions, retrieve entities and relationships related to keywords from GDB, and construct subgraphs. Subgraphs are graphs composed of several nodes and edges, where nodes represent entities and edges represent relationships, such as "chart-contains-bar graph" and "bar graph-represents-data". Analyze subgraphs and extract important information from them, such as entity attributes, relationship types, node degrees, etc. Use this information as a supplement to keywords to increase the richness and readability of keywords. Expand subgraphs and retrieve more related entities and relationships from GDB based on the entities and relationships in the subgraphs, generate more subgraphs, and use the entities and relationships in these subgraphs as related words and as search conditions for the knowledge base to improve the coverage and search efficiency of the knowledge base.

[0108] S4: Using the keyword classification unit 6 based on the semantic processing model, the keywords are screened, and the grammar check and semantic understanding of the text data are performed, and the text data is stored in the knowledge base;

[0109] Furthermore, the keyword screening includes screening the keywords using the keyword classification unit 6, obtaining text data corresponding to the selected pictures marked with the corresponding subordinate numbers, and sorting the text data according to the time sequence to generate time-text data, and storing it in the knowledge base 2 for storage;

[0110] The semantic processing model is used to perform grammar checking and semantic understanding on text data, and to correct or prompt errors or unclear expressions in text data, thereby improving the quality and credibility of text data. It should be noted that the principle of the semantic processing model is to use a bidirectional recurrent neural network BiRNN to analyze text data. BiRNN is a deep learning model composed of two recurrent neural networks RNN, which can simultaneously consider the context information of text data and realize efficient analysis of text data.

[0111] The semantic processing model is expressed as:

[0112]

[0113] in, represents the input feature vector of the lth layer, represents the forward hidden feature vector of the lth layer, represents the backward hidden feature vector of the lth layer, represents the bidirectional hidden feature vector of the lth layer, y t represents the output feature vector, W y represents the weight parameter, b y represents the bias parameter, f represents the activation function, such as ReLU, sigmoid, etc., y represents the real grammatical and semantic labels, represents the predicted grammatical and semantic labels, represents the loss function of natural language processing, log represents logarithmic operation, ∑ represents summation operation, and [;] represents vector concatenation operation.

[0114] The text data is segmented and each word is converted into a word vector. The word vector is a numerical representation that can express the semantic and grammatical information of the word. The word vector is input into the BiRNN model, and after being processed by the forward RNN and the backward RNN, the context feature vector in the text data is extracted. The feature vector is input into the fully connected layer, and after being processed by the activation function and the softmax function, the grammatical and semantic labels of the text data, such as "noun", "verb", "subject", "predicate", etc., are output. According to the grammatical and semantic labels, the text data is grammatically checked and semantically understood, and the errors or unclear expressions in the text data are corrected or prompted to improve the quality and credibility of the text data.

[0115] S5: Continuously update the knowledge base and optimize text data based on feedback.

[0116] Furthermore, the optimization of text data according to user feedback includes taking the knowledge base as the environment, the optimizer as the intelligent agent, defining the state, action and reward of the optimizer, where the state is the text data in the knowledge base, the action is to optimize the text data, and the reward is the user's feedback and evaluation; and using a deep Q network to realize the learning of the optimizer.

[0117] Example 2, reference Figure 2-Figure 6 , which is an embodiment of the present invention, provides a knowledge base construction system based on standard digital processing. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0118] The knowledge base construction system based on standard digital processing comprises a text conversion system (1) and a knowledge base (2); the text conversion system (1) comprises a frame selection unit (3), a data processing unit (4), a mark arrangement unit (5) and a keyword classification unit (6); the frame selection unit (3) is connected to the data processing unit (4) and the mark arrangement unit (5) respectively, and the data processing unit (4) and the mark arrangement unit (5) are both connected to the keyword classification unit (6); the frame selection unit (3) comprises a template reference module (7), a comparison correction module (8) and a template reference module (9). ) and a frame selection and labeling module (9), the template reference module (7) is connected to the comparison and correction module (8), and the comparison and correction module (8) is connected to the frame selection and labeling module (9); the template reference module (7) is used to set a standard template, and map the standard template on the shooting section to generate a shooting frame, and reserve a title space and a filling space; the comparison and correction module (8) is used to use the title space and the filling space to compare and correct the text data, and obtain a plurality of groups of filling texts on the standard text; the frame selection and labeling module (9) is used to select the filling text and then compare the filling text. The selected images are sequentially numbered and marked; the data processing unit (4) comprises an optical character recognition module (10) and a comparison and filling module (11), and the optical character recognition module (10) is connected to the comparison and filling module (11); the optical character recognition module (10) is used to convert the selected images into text data; the comparison and filling module (11) is used to fill the text data into the corresponding filling space in sequence according to the number mark to obtain the text data; the mark sorting unit (5) comprises a module number comparison module (12) and a mark definition module (13), and the module The block number comparison module (12) is connected to the tag definition module (13); the keyword classification unit (6) includes a keyword screening module (14) and a time sorting module (15), and the keyword screening module (14) is connected to the time sorting module (15); the keyword screening module (14) is used to screen keywords and obtain text data corresponding to the selected picture under the corresponding subordinate number tag; the time sorting module (15) is used to sort the text data according to the time sequence, generate time-text data, and transmit the time-text data to the knowledge base (2).

[0119] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0120] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0121] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0122] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0123] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0124] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0125] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A knowledge base construction method based on standard digital processing, characterized in that: include, Based on the image recognition model, the frame selection unit (3) is used to select and classify the text data, select or generate a standard template, and mark it with numbers; Based on the semantic analysis model, the data processing unit (4) is used to convert the selected image into text data and extract key information from the image; Using the tag arrangement unit (5), the title space and the filling space are associated with each other in a master-slave relationship, and the keywords are expanded; Based on the semantic processing model, the keyword classification unit (6) is used to screen keywords, perform grammar checking and semantic understanding on the text data, and store them in the knowledge base; The knowledge base is continuously updated and text data is optimized based on feedback.

2. A method for constructing a knowledge base based on standard digital processing as claimed in claim 1, characterized in that: The image selection and classification of text data includes setting a standard template using a frame selection unit (3), mapping the standard template on the shooting section, generating a shooting frame, reserving a title space and a filling space, using the title space and the filling space to compare and correct the text data, obtaining the filling text on the standard text, performing image selection on the filling text, and marking the selected image with a number. At the same time, using an image recognition model to classify the selected image, automatically selecting a standard template according to the category of the image, or generating a new template, the steps are as follows: The cropped image is input into the image recognition model, and after being processed by the convolution layer and the pooling layer, the feature vector in the image is extracted; the feature vector is input into the fully connected layer, and after being processed by the activation function and the softmax function, the category probability distribution of the image is output; The image recognition model is expressed as: Among them, x(l)i,j,k represents the input feature map of the first layer, w(l)m,n,p,k represents the convolution kernel parameters of the first layer, represents the bias parameter of layer 1, represents the convolution output of the lth layer, f represents the activation function, represents the activation output of layer 1, represents the pooled output of the first layer, M, N, P represent the size of the convolution kernel, and max represents the maximum value operation; According to the probability distribution, the category with the highest probability is selected as the recognition result of the image, and a standard template matching the image category is selected from the preset standard template library; if there is no standard template matching the image category, or the image category is unclear, the image template generation model is used to migrate the image and generate a new template.

3. A method for constructing a knowledge base based on standard digital processing as claimed in claim 2, characterized in that: The picture template generation model is expressed as: G(z)=CNN(z) D(x)=CNN(x) Where z represents a random noise vector, x represents a real image, G(z) represents an image generated by the generator, D(x) represents the probability of the discriminator distinguishing the real image, D(G(z)) represents the probability of the discriminator distinguishing the generated image, and CNN represents a convolutional neural network. represents the loss function of the generator and the discriminator, represents the expected value, pdata(x) represents the distribution of the real image, and pz(z) represents the distribution of the noise vector; The steps of the image template generation model include inputting the image into the generator, generating a new template through processing by the convolution layer and the deconvolution layer, inputting the new template and the preset standard template into the discriminator together, outputting the true and false probability distribution of the image through processing by the convolution layer and the fully connected layer, calculating the loss function of the generator and the discriminator according to the probability distribution, and updating the parameters of the generator and the discriminator using the back propagation algorithm until the generator and the discriminator reach a balance, that is, the discriminator cannot distinguish the true from the false for the image generated by the generator; the image generated by the generator is used as a new template for frame selection screenshots.

4. A method for constructing a knowledge base based on standard digital processing as claimed in claim 3, characterized in that: The extracting of key information from the image includes using a data processing unit (4) to convert the selected image into text data, and filling the text data into a corresponding filling space according to the number mark to obtain the text data; Using the semantic analysis model, we can perform semantic analysis on the selected images, extract key information from the images, and use the information as a supplement to the text data, expressed as: h (l) =f(W (l) x (l-1) +b (l) ) Among them, x (l) represents the input feature map of the first layer, h (l) represents the hidden feature map of layer 1, represents the output feature map of the first layer, W(l) represents the weight parameter of the first layer, b(l) represents the bias parameter of the first layer, f represents the activation function, x represents the pixel category distribution of the real image, Represents the pixel category distribution of the predicted image, represents the loss function of semantic segmentation, log represents logarithmic operation, and ∑ represents summation operation; Semantic analysis includes inputting the cropped image into the SSN model, extracting the high-dimensional feature vector in the image after being processed by the encoder; inputting the feature vector into the decoder, restoring the original size of the image after being processed by the deconvolution layer and the upsampling layer, and outputting the pixel category distribution of the image; segmenting the image into different areas according to the pixel category distribution, each area corresponds to a semantic category, and performing semantic analysis on each area according to the segmentation results to extract information in the area as a supplement to the text data.

5. A method for constructing a knowledge base based on standard digital processing as claimed in claim 4, characterized in that: The master-slave association of the title space and the filling space includes using a tag sorting unit (5) to perform master-slave association of the title space and the filling space, establishing a master-slave association between the number tag of the selected picture corresponding to the filling space and the title space, and filling the tag name in the title space according to the standard template as a keyword; Using the knowledge graph, keywords are expanded, including using keywords as query conditions, retrieving entities and relationships related to the keywords from GDB, and constructing a subgraph, which is a graph composed of nodes and edges, where nodes represent entities and edges represent relationships. The subgraph is analyzed and information in the subgraph is extracted as a supplement to the keywords. The subgraph is expanded, and a new subgraph is generated based on the entities and relationships in the subgraph and the corresponding entities and relationships retrieved from GDB. At the same time, the entities and relationships in the subgraph are used as key words and set as retrieval conditions for the knowledge base.

6. A method for constructing a knowledge base based on standard digital processing as claimed in claim 5, characterized in that: The keyword screening includes using a keyword classification unit (6) to screen the keywords, obtain text data corresponding to the selected pictures marked with corresponding subordinate numbers, sort the text data according to time sequence, generate time-text data, and store it in the knowledge base (2) for storage; The semantic processing model is used to perform grammar checking and semantic understanding on the text data, and to correct or prompt the errors or unclear expressions in the text data, which can be expressed as: in, represents the input feature vector of layer 1, represents the forward hidden feature vector of layer 1, represents the backward hidden feature vector of layer 1, represents the bidirectional hidden feature vector of layer 1, y t represents the output feature vector, W y represents the weight parameter, b y represents the bias parameter, f represents the activation function, y represents the actual grammatical and semantic labels, represents the predicted grammatical and semantic labels, represents the loss function of natural language processing, log represents logarithmic operation, ∑ represents summation operation, [;] represents vector concatenation operation; The semantic processing step includes segmenting the text data, converting each word into a word vector, inputting the word vector into the BiRNN model, and extracting the context feature vector in the text data through the forward RNN and the backward RNN processing; The feature vector is input into the fully connected layer, and after being processed by the activation function and the softmax function, the grammatical and semantic labels of the text data are output. According to the grammatical and semantic labels, the text data is grammatically checked and semantically understood, and errors or unclear expressions in the text data are corrected and prompted.

7. A method for constructing a knowledge base based on standard digital processing as claimed in claim 6, characterized in that: The optimization of text data according to feedback includes taking the knowledge base as the environment, taking the optimizer as the intelligent agent, defining the state, action and reward of the optimizer, wherein the state is the text data in the knowledge base, the action is to optimize the text data, and the reward is the user's feedback and evaluation; and using a deep Q network to realize the learning of the optimizer.

8. A system using the knowledge base construction method based on standard digital processing as claimed in any one of claims 1 to 7, characterized in that: It includes a text conversion system (1) and a knowledge base (2); The text conversion system (1) comprises a frame selection unit (3), a data processing unit (4), a mark arrangement unit (5) and a keyword classification unit (6), wherein the frame selection unit (3) is connected to the data processing unit (4) and the mark arrangement unit (5) respectively, and the data processing unit (4) and the mark arrangement unit (5) are both connected to the keyword classification unit (6); the frame selection unit (3) comprises a template reference module (7), a comparison correction module (8) and a frame selection label module (9), wherein the template reference module (7) and the comparison correction module (8) docking, the comparison and correction module (8) is docked with the frame selection and labeling module (9); the template reference module (7) is used to set a standard template, and map the standard template on the shooting section to generate a shooting frame, and reserve a title space and a filling space; the comparison and correction module (8) is used to use the title space and the filling space to compare and correct the text data, and obtain a plurality of groups of filling texts on the standard text; the frame selection and labeling module (9) is used to cut out the filling text and then number the cut out pictures in sequence; the data processing The unit (4) comprises an optical character recognition module (10) and a comparison and filling module (11), wherein the optical character recognition module (10) is connected to the comparison and filling module (11); the optical character recognition module (10) is used to convert the selected image into text data; the comparison and filling module (11) is used to sequentially fill the text data into the corresponding filling space according to the number mark to obtain the text data; the tag sorting unit (5) comprises a module number comparison module (12) and a tag definition module (13), wherein the module number comparison module (12) is connected to the tag definition module (13); the keyword classification unit (6) comprises a keyword screening module (14) and a time sorting module (15), wherein the keyword screening module (14) is connected to the time sorting module (15); the keyword screening module (14) is used to screen keywords and obtain text data corresponding to the selected image under the corresponding subordinate number mark; the time sorting module (15) is used to sort the text data according to the time sequence, generate time-text data, and transmit the time-text data to the knowledge base (2).

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images