A method and system for grading books for preschool children

Abstract

The present invention discloses a method and system for grading books for preschool children, and relates to the technical field of book grading. A system for grading books for preschool children includes: a book data acquisition module, a book grading strategy construction module, and a book grading label generation module. The present invention classifies books through multi-level labels, which can more accurately reflect the nature of the book content and the suitable reader group, thereby realizing the refined management of books; not only the text content of the book is considered, but also the image and voice data are covered, so that the book can be comprehensively analyzed from multiple angles to ensure that the grading results are more reasonable; the book grading label sub-level selection library is constructed based on an optimized group algorithm, and can self-adjust according to new information or feedback to adapt to changing needs; by accurately grading books, better services can be provided to users, helping users find books that meet their needs and improving user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of book classification, and in particular to a method and system for classifying books for preschool children. Background Art

[0002] Traditional classification of preschool books usually relies on the prior knowledge of a review team composed of educational experts, psychologists and linguists, who evaluate the books and mark the appropriate age group on the cover or title page. However, different review teams may adopt different standards, which makes it difficult to unify the standards. In addition, the diversity of book content and the wide range of topics make comprehensive coverage complicated. At the same time, considering the differences in children's understanding and acceptance of content in different cultural backgrounds, the classification of preschool books needs to be adjusted from a multicultural perspective. In order to improve the scientificity and applicability of classification, we can explore the use of technologies such as deep learning to analyze book content and reader feedback, so as to establish more objective and comprehensive classification indicators to better meet the needs of different children. Summary of the invention

[0003] The present invention aims to provide a method and system for grading books for preschool children, so as to make the grading of children's books more detailed and humane.

[0004] A method for grading books for preschool children comprises the following steps:

[0005] Step S1: obtaining the content data of the books to be classified; obtaining the first label X1 of the book classification and the preliminary classification book content data set based on the content data of the books to be classified and the book classification initial analysis model; the preliminary classification book content data set includes the image data of the books to be classified, the text data of the books to be classified and the voice data of the books to be classified; the book classification initial analysis model includes a first label analysis layer, a content extraction layer and an initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of the book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model; the book classification label sub-level selection library contains N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1, 2, ..., N; the book classification label sub-level selection library is constructed based on the optimized group algorithm; the primary classification book content data set is input into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2;

[0006] Step S2: Based on the first book classification label X1 and the second book classification label set X2, the classified books are classified in the comprehensive book classification analysis model to obtain a comprehensive book classification label; the comprehensive book classification analysis model includes a classification label matching layer, a comprehensive classification label analysis layer and a comprehensive classification label output layer, which are used to analyze the first book classification label X1 and the second book classification label set X2 of the classified books to obtain a comprehensive book classification label; book classification is marked on the classified books according to the comprehensive book classification label.

[0007] As a preferred technical solution of the present invention, determine the N standard book classification first labels Y in the book classification label sub-level selection library n The specific steps include:

[0008] The preset age stage is used as the age graph to be divided, and the age graph to be divided needs to be divided into Q sub-graphs P q , q=1,2,…,Q, each subgraph P q Indicates the age division of the preset age stages;

[0009] Construct K initial age-classified individuals H to be optimized k ; Each initial age classification individual H to be optimized k is a partition solution of the age graph to be divided; K initial age classification individuals H to be optimized k Combine and obtain the initial age-graded iterative population to be optimized; when performing population iteration, it is divided into a population development stage and a population exploration stage;

[0010] Set parameters R1 and R2, R1=V-((V*t) / t max ), R2=2*R1*W1-R1; W1 is a random number in [0, 1], V is the iteration speed, t is the current iteration number, t max is the maximum number of iterations, t=1, 2, …, t max ; Parameter R1 is used to control parameter R2 for transfer control of population development and population exploration;

[0011] To optimize the initial age classification individual H k Perform simulation calculation to obtain the fitness S k ;

[0012] The specific steps for simulation calculation include:

[0013] Collect several groups of children's growth samples; each group of children's growth samples contains a set of continuous growth indexes of experimental children in a preset age stage; for the initial age classification individual H to be optimized kDivide the children's growth samples to obtain the divided stages of children's growth, calculate the mean value of the corresponding growth indices within all the divided stages of children's growth, and obtain the mean growth indices of the Q groups of divided stages of children's growth; calculate the mean Euclidean distance between the difference sequence values of the mean growth indices of the Q groups of divided stages of children's growth and the preset growth index change threshold to obtain the mean similarity;

[0014] Take the reciprocal of the standard deviation of the mean similarities corresponding to several groups of children's growth samples as the fitness S of the initial age classification individual Hk to be optimized k ;

[0015] During the t-th population iteration, if |R2| > V / 2, it is in the population exploration stage, and the initial age classification iteration population to be optimized performs population iteration globally; if |R2| < V / 2, it is in the population exploitation stage, and the initial age classification iteration population to be optimized performs population iteration near the local optimal solution;

[0016] During the t max -th population iteration, output the initial age classification individual H k corresponding to the maximum fitness, which is the optimal initial age classification individual to be optimized; take one division solution of the age map to be divided in the optimal initial age classification individual as the division of the preset age stage to obtain N age divisions, and take the N age divisions as the N standard book classification first labels Y n .

[0017] As a preferred technical solution of the present invention, the specific steps for constructing the book classification label sub-level analysis model M in the book classification label sub-level selection library include: n :

[0018] Generate a book content feature set J n according to the standard book classification first label Y n , and the book content feature set J n contains several book content features that conform to the standard book classification first label Y n ;

[0019] The book classification label sub-level analysis model M n is established based on the standard book classification label sub-level analysis model, and the standard book classification label sub-level analysis model includes a speech content conversion layer, a text content analysis layer B n , an image feature extraction layer, an image feature analysis layer F n and a second label output layer; the text content analysis layer B n and the image feature analysis layer F n of each book classification label sub-level analysis model M n are constructed by different training sets;

[0020] The speech content conversion layer is used to perform speech recognition on the book speech data to obtain the book speech text data;

[0021] Text content analysis layer B n Used to extract features from book voice text data and book text data to obtain book text features; extract features from book text features to obtain a book text content label set;

[0022] The image feature extraction layer includes a zero-filling layer, a 7*7 convolution layer, a normalization layer, an activation layer, and a 3*3 maximum pooling layer; it is used to perform feature processing on the book image data to obtain the book image data features;

[0023] Image feature analysis layer F n It includes an average pooling layer, a dimension reduction layer, and a fully connected layer; it is used to perform feature recognition on the features of book image data to obtain a set of book image content labels;

[0024] The second label output layer is used to combine the book text content label set and the book image content label set to obtain the second label set for book classification.

[0025] As a preferred technical solution of the present invention, the book classification label sub-level analysis model M is trained. n Chinese text content analysis layer B n and image feature analysis layer F n The specific steps include:

[0026] Collection and book content feature set J n Related text feature training set U (J n ) and image feature training set L(J n );

[0027] Using the text feature training set U (J n ) Text content analysis layer B n Perform model training to obtain the initial text content analysis layer B n '; Using the image feature training set L (J n ) Image feature analysis layer F n Perform model training to obtain the initial image feature analysis layer F n ';

[0028] Initial text content analysis layer B n ' and the initial image feature analysis layer F n 'Perform model evaluation. If the model evaluation result is passed, the initial text content analysis layer B n 'As a book rating label sub-level analysis model M nText content analysis layer B n , the initial image feature analysis layer F n 'As a book rating label sub-level analysis model M n The image feature analysis layer F n ; Otherwise, continue model training.

[0029] As a preferred technical solution of the present invention, the book classification initial analysis model in step S1 includes a first label analysis layer, a content extraction layer and an initial analysis output layer;

[0030] The first label analysis layer is used to extract features of the content data of the books to be classified, and obtain features of the content data of the books to be classified; perform initial label analysis on the features of the content data of the books to be classified, and obtain the first label X1 of the book classification;

[0031] The content extraction layer is used to classify and extract the content data of the books to be classified, and obtain the image data of the books to be classified, the text data of the books to be classified, and the voice data of the books to be classified;

[0032] The initial analysis output layer is used to combine the image data of books to be classified, the text data of books to be classified and the voice data of books to be classified to obtain the preliminary classified book content data set, and output the first label X1 of the book classification and the preliminary classified book content data set.

[0033] As a preferred technical solution of the present invention, the specific steps of training the first label analysis layer include:

[0034] Collecting several groups of first label analysis training samples, each group of first label analysis training samples contains classified book content data and annotated standard book classification first labels; combining several groups of first label analysis training samples to obtain a first label analysis training set;

[0035] According to the first label analysis training set, the first label analysis layer is trained with the marked standard book classification first label as the target to obtain an initial first label analysis layer; a model evaluation is performed on the initial first label analysis layer to obtain an initial first label analysis layer model evaluation result; if the initial first label analysis layer model evaluation result is passed, the initial first label analysis layer is used as the first label analysis layer; otherwise, the model training is continued.

[0036] As a preferred technical solution of the present invention, the comprehensive book classification analysis model in step S2 includes a classification label matching layer, a comprehensive classification label analysis layer and a comprehensive classification label output layer;

[0037] The classification label matching layer is used to select the corresponding book classification content weight matching library according to the first label X1 of the book classification;

[0038] The comprehensive classification label analysis layer is used to match the scores of the book classification second label set X2 according to the corresponding book classification content weight matching library to obtain the book classification label score; perform weighted addition of all book classification label scores to obtain the book comprehensive evaluation score; determine the comprehensive book classification label according to the book comprehensive evaluation score;

[0039] The comprehensive classification label output layer is used to output comprehensive book classification labels.

[0040] A preschool children's book rating system, comprising:

[0041] The book data acquisition module includes a data acquisition unit; the data acquisition unit is used to acquire the content data of the book to be classified.

[0042] The book classification strategy construction module includes a strategy construction unit; the strategy construction unit is used to construct a book classification label sub-level selection library, and the book classification label sub-level selection library is constructed based on the optimized group algorithm; the book classification label sub-level selection library includes N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1,2,…,N;

[0043] The book classification label generation module includes a first classification label generation unit, a second classification label generation unit and a comprehensive classification label generation unit;

[0044] The first classification label generation unit is used to obtain the first label X1 of book classification and the preliminary classification book content data set based on the content data of the book to be classified and the initial analysis model of book classification; the preliminary classification book content data set includes the image data of the book to be classified, the text data of the book to be classified and the voice data of the book to be classified; the initial analysis model of book classification includes the first label analysis layer, the content extraction layer and the initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model;

[0045] The second classification label generation unit is used to input the initial classification book content data set into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2;

[0046] The comprehensive grading label generation unit is used to grade the to-be-graded books in the comprehensive book grading analysis model based on the book grading first label X1 and the book grading second label set X2 to obtain a comprehensive book grading label; the comprehensive book grading analysis model includes a grading label matching layer, a comprehensive grading label analysis layer and a comprehensive grading label output layer, which are used to analyze the book grading first label X1 and the book grading second label set X2 of the to-be-graded books to obtain a comprehensive book grading label; and the to-be-graded books are marked with book grading labels according to the comprehensive book grading labels.

[0047] The present invention has the following advantages:

[0048] 1. The present invention classifies books through multi-level labels, which can more accurately reflect the nature of the book content and the suitable reader group, thereby realizing the refined management of books; not only the text content of the book is considered, but also the image and voice data are covered, so that the book can be comprehensively analyzed from multiple angles to ensure that the grading results are more reasonable; the book grading label sub-level selection library is constructed based on the optimized group algorithm, which is more in line with the actual situation, while improving the construction speed, and can self-adjust according to new information or feedback to adapt to changing needs; by accurately grading books, better services can be provided to users, helping users find books that meet their needs and improving user experience.

[0049] 2. The present invention uses children's growth samples and their growth indexes to make the division of age stages more scientific and reasonable, rather than simply following administrative or traditional methods; it adopts a swarm intelligence algorithm, which can dynamically adjust strategies in different iterations to find the most suitable age division scheme; by constructing a book grading label sub-level analysis model, in-depth analysis of the book content can more accurately identify the theme, style, emotional tendency and other characteristics of the book, thereby helping to formulate a more accurate book classification; different book grading label sub-level analysis models can be adjusted and optimized according to different book content feature sets, which increases the application scope and flexibility of the model, and at the same time, selection based on the first label X1 of the book classification is more targeted. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention is a schematic diagram of the structure of a preschool children's book grading system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0052] Embodiment 1, a method for grading books for preschool children, comprising the following steps:

[0053] Step S1: obtaining the content data of the books to be classified; obtaining the first label X1 of the book classification and the preliminary classification book content data set based on the content data of the books to be classified and the book classification initial analysis model; the preliminary classification book content data set includes the image data of the books to be classified, the text data of the books to be classified and the voice data of the books to be classified; the book classification initial analysis model includes a first label analysis layer, a content extraction layer and an initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of the book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model; the book classification label sub-level selection library contains N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1, 2, ..., N; the book classification label sub-level selection library is constructed based on the optimized group algorithm; the primary classification book content data set is input into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2;

[0054] Determine the sub-level of the book classification label and select the first label Y of N standard book classifications in the library n The specific steps include:

[0055] The preset age stage is used as the age graph to be divided, and the age graph to be divided needs to be divided into Q sub-graphs P q , q=1,2,…,Q, each subgraph P q Indicates the age division of the preset age stages; abstracting the age stages into a diagram is more vivid and specific; making the book classification more accurate and more adaptable to the actual needs of readers of different age groups; the preset age stages are set by professional technicians according to actual conditions. For example, in the preschool stage, the preset age stages are generally 1-7 years old;

[0056] Construct K initial age-classified individuals H to be optimized k ; Each initial age classification individual H to be optimized k is a partition solution of the age graph to be divided; K initial age classification individuals H to be optimized k Combine and obtain the initial age-graded iterative population to be optimized; when performing population iteration, it is divided into a population development stage and a population exploration stage;

[0057] Set parameters R1 and R2, R1=V-((V*t) / t max), R2 = 2 * R1 * W1 - R1; W1 is a random number within [0, 1], V is the iteration speed, t is the current iteration number, t max is the maximum iteration number, t = 1, 2,..., t max ; The parameter R1 is used to control the transfer control of the parameter R2 for population development and population exploration; The iteration speed and the maximum iteration number are set by professional technicians according to the actual situation;

[0058] For the initial age - graded individual H to be optimized k perform simulation calculations to obtain the fitness S k ;

[0059] The specific steps for performing simulation calculations include:

[0060] Collect several groups of children's growth samples; Each group of children's growth samples contains a set of continuous growth indices of experimental children within a preset age stage; For the initial age - graded individual H to be optimized k Divide the children's growth samples to obtain the children's growth division stages, calculate the mean of the corresponding growth indices within all children's growth division stages to obtain the mean growth indices of Q groups of children's growth division stages; Calculate the Euclidean distance mean between the difference sequence values of the mean growth indices of Q groups of children's growth division stages and the preset growth index change threshold to obtain the similarity mean; Several groups of children's growth samples are obtained by open and compliant means, and the growth indices of different children are recorded by professionals. The growth index represents the degree of knowledge absorption and acceptance of children;

[0061] Take the reciprocal of the standard deviation of the similarity means corresponding to several groups of children's growth samples as the fitness S of the initial age - graded individual Hk to be optimized k ;

[0062] At the t - th population iteration, if |R2| > V / 2, then it is in the population exploration stage, and the initial age - graded iteration population to be optimized iterates globally; If |R2| < V / 2, then it is in the population development stage, and the initial age - graded iteration population to be optimized iterates near the local optimal solution;

[0063] At the t max - th population iteration, output the initial age - graded individual H corresponding to the maximum fitness k , which is the optimal initial age - graded individual to be optimized; Take one partition solution of the age map to be partitioned in the optimal initial age - graded individual as the partition of the preset age stage to obtain N age partitions, and take the N age partitions as the first labels Y of N standard book classifications n ;

[0064] By dividing the preset age stages into more detailed sub-graphs, it is possible to more accurately match the reading abilities and interests of readers of different age groups, thereby providing more personalized reading materials; by constructing the initial age-grading individuals to be optimized and combining them with children's growth sample data, the book grading is more based on empirical research, which improves the scientificity and effectiveness of book grading; through the iterative population optimization process, the book grading standards can be dynamically adjusted according to the changes in children's growth and the needs of social development, maintaining the timeliness and adaptability of the grading system; through simulation calculation and fitness evaluation, educational resources can be allocated more effectively to ensure that books and other reading materials can meet the growth needs of children of different age groups; by setting parameters R1 and R2, as well as the iteration speed V and the maximum number of iterations t max , making the book grading process more controllable and operable, and easier for professionals to manage and adjust; through accurate book grading, readers can more easily find books that suit their reading level and interests, thereby improving their reading experience and satisfaction;

[0065] Constructing the book classification label sub-level analysis model M in the book classification label sub-level selection library n The specific steps include:

[0066] According to the standard book classification first label Y n , generate book content feature set J n , book content feature set J n Contains the first label Y that meets the standard book classification n Several book content features;

[0067] Book rating label sub-level analysis model M n The sub-level analysis model of standard book classification labels is established, which includes a speech content conversion layer and a text content analysis layer B. n , image feature extraction layer, image feature analysis layer F n and the second label output layer; each book rating label sub-level analysis model M n Text content analysis layer B n and image feature analysis layer F n Constructed from different training sets;

[0068] The speech content conversion layer is used to perform speech recognition on the book speech data to obtain the book speech text data;

[0069] Text content analysis layer B n Used to extract features from book voice text data and book text data to obtain book text features; extract features from book text features to obtain a book text content label set;

[0070] The image feature extraction layer includes a zero-filling layer, a 7*7 convolution layer, a normalization layer, an activation layer, and a 3*3 maximum pooling layer; it is used to perform feature processing on the book image data to obtain the book image data features;

[0071] Image feature analysis layer F n It includes an average pooling layer, a dimension reduction layer, and a fully connected layer; it is used to perform feature recognition on the features of book image data to obtain a set of book image content labels;

[0072] The second label output layer is used to combine the book text content label set and the book image content label set to obtain a second label set for book classification;

[0073] Training book classification label sub-level analysis model M n Chinese text content analysis layer B n and image feature analysis layer F n The specific steps include:

[0074] Collection and book content feature set J n Related text feature training set U (J n ) and image feature training set L(J n );

[0075] Using the text feature training set U (J n ) Text content analysis layer B n Perform model training to obtain the initial text content analysis layer B n '; Using the image feature training set L (J n ) Image feature analysis layer F n Perform model training to obtain the initial image feature analysis layer F n ';

[0076] Initial text content analysis layer B n ' and the initial image feature analysis layer F n 'Perform model evaluation. If the model evaluation result is passed, the initial text content analysis layer B n 'As a book rating label sub-level analysis model M n Text content analysis layer B n , the initial image feature analysis layer F n 'As a book rating label sub-level analysis model M n The image feature analysis layer F n ; Otherwise, continue model training;

[0077] Through deep learning technology, such as CNN models, books can be classified into hierarchical multi-label categories more accurately, thereby improving the accuracy and effectiveness of book grading; accurate grading helps to provide personalized reading material recommendations for children of different ages to meet their reading needs and interests; using multiple convolutional layers for analysis can better grasp image features and determine image content characteristics;

[0078] The book classification initial analysis model in step S1 includes a first label analysis layer, a content extraction layer and an initial analysis output layer;

[0079] The first label analysis layer is used to extract features of the content data of the books to be classified, and obtain features of the content data of the books to be classified; perform initial label analysis on the features of the content data of the books to be classified, and obtain the first label X1 of the book classification;

[0080] The content extraction layer is used to classify and extract the content data of the books to be classified, and obtain the image data of the books to be classified, the text data of the books to be classified, and the voice data of the books to be classified;

[0081] The initial analysis output layer is used to combine the image data of books to be classified, the text data of books to be classified and the voice data of books to be classified to obtain the preliminary classified book content data set, and output the first label X1 of the book classification and the preliminary classified book content data set;

[0082] The specific steps for training the first label analysis layer include:

[0083] Collecting several groups of first label analysis training samples, each group of first label analysis training samples contains classified book content data and annotated standard book classification first labels; combining several groups of first label analysis training samples to obtain a first label analysis training set;

[0084] According to the first label analysis training set, the first label analysis layer is trained with the first label of the standard book classification marked as the target to obtain an initial first label analysis layer; a model evaluation is performed on the initial first label analysis layer to obtain an initial first label analysis layer model evaluation result; if the initial first label analysis layer model evaluation result is passed, the initial first label analysis layer is used as the first label analysis layer; otherwise, the model training is continued;

[0085] By training the first label analysis layer, accurate feature extraction and initial label analysis of book content data can be achieved, thereby obtaining an accurate first label for book grading and ensuring the initial accuracy of book grading. A large number of first label analysis training samples are used for model training, so that the model learning process is based on actual data, which enhances the reliability and practicality of the model. The trained first label analysis layer can automatically complete the tasks of feature extraction and label analysis, reducing the need for manual intervention and improving the efficiency of book grading.

[0086] Step S2: Based on the first book classification label X1 and the second book classification label set X2, the classified books are classified in the comprehensive book classification analysis model to obtain a comprehensive book classification label; the comprehensive book classification analysis model includes a classification label matching layer, a comprehensive classification label analysis layer and a comprehensive classification label output layer, which are used to analyze the first book classification label X1 and the second book classification label set X2 of the classified books to obtain a comprehensive book classification label; the classified books are marked with book classification according to the comprehensive book classification label;

[0087] The comprehensive book classification analysis model in step S2 includes a classification label matching layer, a comprehensive classification label analysis layer and a comprehensive classification label output layer;

[0088] The classification label matching layer is used to select the corresponding book classification content weight matching library according to the book classification first label X1; the book classification content weight matching library indicates the importance and difficulty of different content features for children under the corresponding book classification first label; for example, for the same content feature, different age groups have different acceptance levels, and different standards need to be adopted for classification; the book classification content weight matching library is constructed using prior knowledge and is set by professional technicians after review according to actual conditions;

[0089] The comprehensive classification label analysis layer is used to match the scores in the book classification second label set X2 according to the corresponding book classification content weight matching library to obtain the book classification label score; weighted addition of all book classification label scores to obtain the book comprehensive evaluation score; determine the comprehensive book classification label according to the book comprehensive evaluation score; make a final evaluation based on the book comprehensive evaluation score to determine the comprehensive book classification label to ensure flexibility and non-limitation; different children at the same stage have different degrees of acceptance of knowledge and may need to receive higher-level knowledge at the current age stage. The comprehensive book classification label can provide instructions for different children;

[0090] The comprehensive classification label output layer is used to output comprehensive book classification labels;

[0091] Through the grading label matching layer, it can ensure that the first label of the book grading is accurately matched with the corresponding content weight matching library, thereby improving the accuracy of grading; comprehensively considering the different acceptance levels of different children for the same content feature, the book grading is more personalized and flexible, and better adapted to the needs of different children; the book grading content weight matching library can assign weights according to the importance and difficulty of content features for children, ensuring that the grading is more scientific and reasonable; the comprehensive grading label analysis layer obtains a more comprehensive book comprehensive evaluation score by matching and weighted addition of the scores of the second label set of book grading, making the grading results more comprehensive and objective; the comprehensive book grading label can adapt to the differences in the degree of knowledge acceptance of different children at the same development stage, and provide more suitable reading materials for each child; through precise and personalized grading, it helps children find reading materials that suit them, thereby promoting the development of reading interest and reading ability.

[0092] Example 2, a preschool children's book rating system, see Figure 1 As shown, including:

[0093] The book data acquisition module includes a data acquisition unit; the data acquisition unit is used to acquire the content data of the book to be classified.

[0094] The book classification strategy construction module includes a strategy construction unit; the strategy construction unit is used to construct a book classification label sub-level selection library, and the book classification label sub-level selection library is constructed based on the optimized group algorithm; the book classification label sub-level selection library includes N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1,2,…,N;

[0095] The book classification label generation module includes a first classification label generation unit, a second classification label generation unit and a comprehensive classification label generation unit;

[0096] The first classification label generation unit is used to obtain the first label X1 of book classification and the preliminary classification book content data set based on the content data of the book to be classified and the initial analysis model of book classification; the preliminary classification book content data set includes the image data of the book to be classified, the text data of the book to be classified and the voice data of the book to be classified; the initial analysis model of book classification includes the first label analysis layer, the content extraction layer and the initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model;

[0097] The second classification label generation unit is used to input the initial classification book content data set into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2;

[0098] The comprehensive grading label generation unit is used to grade the to-be-graded books in the comprehensive book grading analysis model based on the book grading first label X1 and the book grading second label set X2 to obtain a comprehensive book grading label; the comprehensive book grading analysis model includes a grading label matching layer, a comprehensive grading label analysis layer and a comprehensive grading label output layer, which are used to analyze the book grading first label X1 and the book grading second label set X2 of the to-be-graded books to obtain a comprehensive book grading label; and the to-be-graded books are marked with book grading labels according to the comprehensive book grading labels.

[0099] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for grading books for preschool children, characterized in that: It includes the following steps: Step S1: Obtain the content data of the book to be classified; Based on the content data of books to be classified and the initial analysis model of book classification, the first label X1 of book classification and the data set of initially classified book content are obtained; the data set of initially classified book content includes image data of books to be classified, text data of books to be classified and voice data of books to be classified; the initial analysis model of book classification includes a first label analysis layer, a content extraction layer and an initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model; the book classification label sub-level selection library contains N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1, 2, ..., N; the book classification label sub-level selection library is constructed based on the optimized group algorithm; the primary classification book content data set is input into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2; Step S2: Classify the book to be classified in the comprehensive book classification analysis model based on the first book classification label X1 and the second book classification label set X2 of the book classification, and obtain the comprehensive book classification label; the comprehensive book classification analysis model includes a classification label matching layer, a comprehensive classification label analysis layer, and a comprehensive classification label output layer, which are used to analyze the first book classification label X1 and the second book classification label set X2 of the book to be classified, and obtain the comprehensive book classification label; perform book classification annotation on the book to be classified according to the comprehensive book classification label; Determine the sub-level of the book classification label and select the first label Y of N standard book classifications in the library n The specific steps include: The preset age stage is used as the age graph to be divided, and the age graph to be divided needs to be divided into Q sub-graphs P q , q=1,2,…,Q, each subgraph P q Indicates the age division of the preset age stages; Construct K initial age-classified individuals H to be optimized k ; Each initial age classification individual H to be optimized k is a partition solution of the age graph to be divided; K initial age classification individuals H to be optimized k Combine and obtain the initial age-graded iterative population to be optimized; when performing population iteration, it is divided into a population development stage and a population exploration stage; Set parameters R1 and R2, R1=V-((V*t) / t max ), R2=2*R1*W1-R1; W1 is a random number in [0, 1], V is the iteration speed, t is the current iteration number, t max is the maximum number of iterations, t=1, 2, …, t max ; Parameter R1 is used to control parameter R2 for transfer control of population development and population exploration; To optimize the initial age classification individual H k Perform simulation calculation to obtain the fitness S k ; The specific steps for performing simulation calculations include: Collect several groups of children's growth samples; each group of children's growth samples contains a set of continuous growth indexes of experimental children in a preset age stage; for the initial age classification individual H to be optimized k The child growth samples are divided to obtain the child growth stages, and the means of the corresponding growth indexes in all the child growth stages are calculated to obtain the mean growth index of the Q group of children in the growth stages; the Euclidean distance mean is calculated between the difference sequence values ​​of the mean growth index of the Q group of children in the growth stages and the preset growth index change threshold to obtain the similarity mean; The reciprocal of the standard deviation of the mean similarity corresponding to several groups of children's growth samples is used as the fitness S of the initial age-graded individual Hk to be optimized. k ; In the t-th population iteration, if |R2| > V / 2, it is in the population exploration stage, and the initial age classification iteration population to be optimized performs population iteration globally; if |R2| < V / 2, it is in the population exploitation stage, and the initial age classification iteration population to be optimized performs population iteration near the local optimal solution; In the t max In the population iteration, the output of the initial age-classified individual H to be optimized corresponding to the maximum fitness k , which is the optimal initial age classification individual to be optimized; a partition solution of the age graph to be divided in the optimal initial age classification individual to be optimized is used as the division of the preset age stage to obtain N age divisions, and the N age divisions are used as the first label Y of N standard book classifications n .

2. A method for grading books for preschool children according to claim 1, characterized in that: Constructing the book classification label sub-level analysis model M in the book classification label sub-level selection library n The specific steps include: According to the standard book classification first label Y n , generate book content feature set J n , book content feature set J n Contains the first label Y that meets the standard book classification n Several book content features; Book rating label sub-level analysis model M n The sub-level analysis model of standard book classification labels is established, which includes a speech content conversion layer and a text content analysis layer B. n , image feature extraction layer, image feature analysis layer F n and the second label output layer; each book rating label sub-level analysis model M n Text content analysis layer B n and image feature analysis layer F n Constructed from different training sets; The speech content conversion layer is used to perform speech recognition on the book speech data to obtain the book speech text data; Text content analysis layer B n Used to extract features from book voice text data and book text data to obtain book text features; extract features from book text features to obtain a book text content label set; The image feature extraction layer includes a zero-padding layer, a 7*7 convolutional layer, a normalization layer, an activation layer, and a 3*3 max pooling layer; it is used to process the features of the book image data to obtain the book image data features; Image feature analysis layer F n It includes an average pooling layer, a dimension reduction layer, and a fully connected layer; it is used to perform feature recognition on the features of book image data to obtain a set of book image content labels; The second label output layer is used to combine the book text content label set and the book image content label set to obtain the second book classification label set.

3. A method for grading books for preschool children according to claim 2, characterized in that: Training book classification label sub-level analysis model M n Chinese text content analysis layer B n and image feature analysis layer F n The specific steps include: Collection and book content feature set J n Related text feature training set U (J n ) and image feature training set L(J n ); Using the text feature training set U (J n ) Text content analysis layer B n Perform model training to obtain the initial text content analysis layer B n '; Using the image feature training set L (J n ) Image feature analysis layer F n Perform model training to obtain the initial image feature analysis layer F n '; Initial text content analysis layer B n ' and the initial image feature analysis layer F n 'Perform model evaluation. If the model evaluation result is passed, the initial text content analysis layer B n 'As a book rating label sub-level analysis model M n Text content analysis layer B n , the initial image feature analysis layer F n 'As a book rating label sub-level analysis model M n The image feature analysis layer F n ; Otherwise, continue model training.

4. A method for grading books for preschool children according to claim 3, characterized in that: The initial book classification analysis model in Step S1 includes a first label analysis layer, a content extraction layer, and an initial analysis output layer; The first label analysis layer is used to extract features from the content data of the book to be classified to obtain the content data features of the book to be classified; perform initial label analysis on the content data features of the book to be classified to obtain the first book classification label X1; The content extraction layer is used to perform content classification extraction on the content data of the book to be classified to obtain the book image data to be classified, the book text data to be classified, and the book speech data to be classified; The initial analysis output layer is used to combine the book image data to be classified, the book text data to be classified, and the book speech data to be classified to obtain the initial classified book content data set, and output the first book classification label X1 and the initial classified book content data set.

5. A method for grading books for preschool children according to claim 4, characterized in that: The specific steps for training the first label analysis layer include: Collect several groups of first label analysis training samples, and each group of first label analysis training samples includes the content data of the classified book and the marked standard first book classification label; combine several groups of first label analysis training samples to obtain the first label analysis training set; Train the first label analysis layer according to the first label analysis training set with the marked standard first book classification label as the target to obtain the initial first label analysis layer; perform model evaluation on the initial first label analysis layer to obtain the model evaluation result of the initial first label analysis layer; if the model evaluation result of the initial first label analysis layer is passed, use the initial first label analysis layer as the first label analysis layer; otherwise, continue model training.

6. A method for grading books for preschool children according to claim 5, characterized in that: The comprehensive book classification analysis model in step S2 includes a classification label matching layer, a comprehensive classification label analysis layer and a comprehensive classification label output layer; The classification label matching layer is used to select the corresponding book classification content weight matching library according to the first label X1 of the book classification; The comprehensive classification label analysis layer is used to match the scores of the book classification second label set X2 according to the corresponding book classification content weight matching library to obtain the book classification label score; perform weighted addition of all book classification label scores to obtain the book comprehensive evaluation score; determine the comprehensive book classification label according to the book comprehensive evaluation score; The comprehensive classification label output layer is used to output comprehensive book classification labels.

7. A preschool children's book rating system, characterized in that: The system applies a method for grading books for preschool children as described in any one of claims 1 to 6, comprising: The book data acquisition module includes a data acquisition unit; the data acquisition unit is used to acquire the content data of the book to be classified. The book classification strategy construction module includes a strategy construction unit; the strategy construction unit is used to construct a book classification label sub-level selection library, and the book classification label sub-level selection library is constructed based on the optimized group algorithm; the book classification label sub-level selection library includes N standard book classification first labels Y n and the corresponding book rating label sub-level analysis model M n , n=1,2,…,N; The book classification label generation module includes a first classification label generation unit, a second classification label generation unit and a comprehensive classification label generation unit; The first classification label generation unit is used to obtain the first label X1 of book classification and the preliminary classification book content data set based on the content data of the book to be classified and the initial analysis model of book classification; the preliminary classification book content data set includes the image data of the book to be classified, the text data of the book to be classified and the voice data of the book to be classified; the initial analysis model of book classification includes the first label analysis layer, the content extraction layer and the initial analysis output layer, which are used to perform initial label matching on the books to be classified and provide subsequent classification support; based on the first label X1 of book classification, a model is selected in the book classification label sub-level selection library to obtain the corresponding book classification label sub-level analysis model; The second classification label generation unit is used to input the initial classification book content data set into the corresponding book classification label sub-level analysis model for analysis to obtain the book classification second label set X2; The comprehensive grading label generation unit is used to grade the to-be-graded books in the comprehensive book grading analysis model based on the book grading first label X1 and the book grading second label set X2 to obtain a comprehensive book grading label; the comprehensive book grading analysis model includes a grading label matching layer, a comprehensive grading label analysis layer and a comprehensive grading label output layer, which are used to analyze the book grading first label X1 and the book grading second label set X2 of the to-be-graded books to obtain a comprehensive book grading label; and the to-be-graded books are marked with book grading labels according to the comprehensive book grading labels.

Images