PPT automatic generation method based on AI and textbook resource library

Abstract

The invention belongs to the technical field of PPT generation, and particularly relates to an automatic PPT generation method based on AI and a textbook resource library, and the method comprises the steps: analyzing first input information of a user through a natural language processing technology, recognizing a core concept and a logic structure, and obtaining an analysis result; automatically retrieving teaching materials related to the teaching subject from a teaching material resource library through an AI technology; the method comprises the following steps: selecting layout and design styles from a preset template library according to the characteristics of teaching materials, automatically generating a PPT frame containing characters, pictures and chart elements, and filling the teaching materials into the PPT frame to form a preliminary PPT; performing detail adjustment and optimization on the preliminary PPT by using an intelligent typesetting optimization technology, and generating a preview PPT; and optimizing the preview PPT based on a user feedback learning mechanism to generate a final PPT. According to the invention, intelligent, automatic and personalized generation of PPT can be realized, and the efficiency and quality of teaching resource preparation are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of PPT generation, and specifically relates to a method for automatically generating PPT based on AI and teaching material resource libraries. Background Art

[0002] Currently, PPTs are increasingly widely used in the fields of education and business. Especially in teaching scenarios, PPTs have become an indispensable teaching tool. However, there are many drawbacks in the traditional PPT production process: on the one hand, producers need to spend a lot of time collecting content from various materials, and then design the layout of slides one by one, typeset text, insert elements such as pictures and charts. This process is not only time-consuming and laborious, but also requires high design capabilities and professional knowledge of the producers; on the other hand, although some existing AI tools can achieve simple conversion from text content to PPT pages, in teaching scenarios, due to the lack of in-depth understanding and optimization of teaching material content, these tools cannot meet the personalized needs of teaching demonstrations based on teaching materials, resulting in a large room for improvement in the quality and practicality of the generated PPTs.

[0003] In the field of education, teachers often need to extract key knowledge points from teaching materials and expand and deepen them in combination with relevant teaching resources to ensure the systematicness, logic, and richness of teaching content. However, most current PPT production tools cannot effectively assist teachers in completing this process, that is, they cannot automatically retrieve high-quality teaching materials related to the teaching theme from the teaching material resource library, and even less can intelligently select appropriate layouts and design styles according to the characteristics of teaching content, and reasonably match and fill elements such as text, pictures, and charts into the PPT framework. In addition, after the PPT is generated, there is a lack of effective user feedback mechanisms and intelligent optimization means, and it is impossible to make targeted adjustments and optimizations to the PPT according to the actual use feedback of teachers and students to improve the teaching effect. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for automatically generating PPT based on AI and teaching material resource libraries, which can realize the intelligent, automated, and personalized generation of PPTs, and significantly improve the efficiency and quality of teaching resource preparation.

[0005] The present invention provides the following technical solutions: A method for automatically generating PPT based on AI and teaching material resource libraries, comprising the following steps: S1. Through natural language processing technology, parse the user's first input information, identify the core concepts and logical structures, and obtain the parsing result; S2. According to the parsing result, automatically retrieve teaching materials related to the teaching theme from the teaching material resource library through AI technology; S3. According to the characteristics of teaching materials, select the layout and design style from the preset template library, and automatically generate a PPT framework containing text, picture, and chart elements. Fill the teaching materials into the PPT framework to form a preliminary PPT. S4. Use intelligent typesetting optimization technology to make detailed adjustments and optimizations to the preliminary PPT, and generate a preview PPT. S5. Optimize the preview PPT based on the user feedback learning mechanism to generate the final PPT.

[0006] Preferably, in S1, through natural language processing technology, parse the user's first input information, identify the core concepts and logical structure, and obtain the parsing result. The identification process is as follows: Set the user's first input information as I, where I is text information. Use the adjacency matrix to calculate the parsing function , and the calculation formula is as follows: ; In the formula, C represents the set of core concepts, which is the set composed of the concept elements extracted from I. L represents the logical structure, which is used to reflect the relationship between core concepts. Set the expanded set of core concepts as , and use the shortest path algorithm and based on a predefined knowledge graph to mine associated concepts and derivative concepts, where V represents the set of nodes in the graph, and E represents the set of edges. Perform calculations on , and the calculation formula is as follows: ; In the formula, u represents the concept variable, which is used to traverse each concept in the set of core concepts C. v represents the semantic variable, which represents any node in the knowledge graph. represents the logical AND operator, represents the existential quantifier in logical symbols, represents the function related to C, represents the union operation symbol, and e represents the edge in the knowledge graph.

[0007] Preferably, when the input information is multimodal information, set the input multimodal information as , where Q represents image information and A represents audio information. The output fused semantics is represented as SQ, and perform calculations on the fused semantics SQ. The calculation formula is as follows: ; In the formula, represents the fusion function, , and respectively represent semantic feature extraction for text, image, and audio information, and convert them into corresponding semantic feature vectors. It means that the input M containing multimodal information such as text, image, and audio is passed to this function, and after arithmetic processing, the fused semantics SQ is finally output.

[0008] Preferably, in S2, according to the parsing result, teaching materials related to the teaching theme are automatically retrieved from the teaching material resource library through AI technology, and the retrieval process is as follows: The core concept set C is represented as , where each concept element corresponds to the core content in the teaching theme, i = 1, 2,..., n, and n represents the total number of concept elements; Define the association degree function to measure the association degree between the concept element and any node v in the knowledge graph. Calculate the association degree function , and the calculation formula is as follows: ; In the formula, represents the shortest path length between the concept element and any node v in the knowledge graph, represents the edge weight adjustment factor, represents the multimodal semantic weight factor, , represents the semantic similarity between the fused semantics SQ and any node v in the knowledge graph; Set the set of materials related to the teaching theme retrieved through the knowledge graph as , and each material corresponds to a set of nodes in the knowledge graph. k = 1, 2,..., m, where m is the total number of materials, and the associated knowledge point set is ; For each material , calculate its comprehensive association degree score , and the calculation formula is as follows: ; In the formula, represents the weight coefficient of the concept element ; Based on the reinforcement learning algorithm, adaptive screening is performed on each material .

[0009] Preferably, the adaptive screening of each material based on the reinforcement learning algorithm is as follows: For each material , extract multiple feature attributes of it to construct a feature vector , kp represents the total number of feature vectors, and collect the feedback data of teachers on the use of materials in the previous PPT to form a feedback vector , kq represents the total number of feedback data features, and set the current teaching scenario information vector as , kr represents the total number of teaching scenario information features; Set a screening value function for measuring materials whether it should be screened and retained in the current teaching scenario, and the screening formula is as follows: ; In the formula, represents an activation function, represents the weight coefficient of the material feature attribute , represents a feedback data element , represents a teaching scenario information element , represents a feature attribute vector , represents the dimension of the feedback vector UG, represents the dimension of the current teaching scenario information vector IE, represents the weight coefficient of reinforcement learning dynamic optimization, represents the semantic matching degree of the fusion semantics SQ and the material knowledge point set ; Through the reinforcement learning algorithm, according to the new feedback data of teachers and the actual teaching effect feedback information after each generation of PPT, continuously adjust , and the values of, so that the screening value function can more accurately reflect the adaptability of materials in different teaching situations; Set the screening threshold as ; When , it is determined that the material meets the current teaching situation and is retained; When , it is determined that the material does not meet the current teaching situation and is discarded.

[0010] Preferably, in S3, automatically generate a PPT framework containing text, picture, and chart elements, and fill the teaching materials into the PPT framework to form a preliminary PPT. The formation process is as follows: Set that the preliminary PPT contains multiple pages , ad = 1, 2, …, pd, where pd is the total number of pages of the PPT, and each page is generated depending on the layout selected from the preset template library and the filling operation of teaching materials. The generation formula is as follows: ; In the formula, represents the operation function of filling teaching materials into the selected layout; Establish a layout dynamic generation model based on content semantics to determine the layout of each page and calculate the adaptation score of the mv-th layout style for the pd-th page . The calculation formula is: ; In the formula, represents the semantic weight of the is-th type of teaching material in the ad-th page, represents the semantic correlation degree between the is-th teaching material and the js-th teaching material in the ad-th page. nf represents the number of types of teaching materials, and mv represents the number of layout styles that can be selected for the page layout. represents the semantic weight of the js-th type of teaching material in the ad-th page, represents the fitness factor of the is-th teaching material for the mv-th layout style in the ad-th page, and its value range is between 0 and 1. represents the matching degree weight factor. , represents the matching degree of fusing semantic SQ and the number of layout styles mv; Finally select the layout style with the maximum value. The selection formula is as follows: ; Adopt the collaborative technology of image understanding and text analysis to realize the intelligent combination of text, pictures and charts and complete the filling operation.

[0011] Preferably, for the collaborative technology of image understanding and text analysis to realize the intelligent combination of text, pictures and charts and complete the filling operation, the combination process is as follows: In the operation function, for each group of text zw and picture pz in the ad-th page, set to represent the optimal combination position and actual proportion information of picture pz relative to text zw in the ad-th page. represents the theme content vector described by text zw in the ad-th page. ​It represents the theme content vector contained in the picture pz on the ad-th page; Calculate the content fitness score between the text zw and the picture pz on the ad-th page , and the calculation formula is as follows: ; Finally, fill all the determined texts and pictures with their combined positions and display ratios into the ad-th page according to the layout to complete the generation of the page and generate a complete preliminary PPT.

[0012] Preferably, in S4, use intelligent layout optimization technology to perform detailed adjustment and optimization on the preliminary PPT, and generate a preview PPT. The generation process is as follows: Set the total width of the page as WY and the total height as HY. The horizontal coordinate range occupied by the picture in the page is , and the vertical coordinate range is . Its width is wy and its height is hy. Set the number of visual blocks divided in the text area as nfi. For the fe-th visual block, fe = 1, 2,..., nfi, its width and height are calculated as follows: ; In the formula, represents the width distribution coefficient; ; In the formula, represents the height distribution coefficient; Set the basic text font size as , and calculate the actual displayed text font size sa. The calculation formula is as follows: ; In the formula, da represents the text distribution density adjustment coefficient; Set the number of data series in the chart as msf, and msf is an integer. For each data series jf, jf = 1, 2,..., msf, its interaction response function is expressed as follows: If , then ; If , then ; In the formula, ck represents the coordinate of the user's interaction operation position on the page, represents the specific animation display function of the data series jf; Set the initial height of the column of the bar chart data series as , the target animation display height is , the animation duration is , the time variable for animation playback is , where , calculate the specific animation display function , and the calculation formula is as follows: ; Finally, after re - typesetting, a preview PPT is generated.

[0013] Preferably, in S5, optimize the preview PPT based on the user feedback learning mechanism to generate the final PPT, and the generation process is as follows: Set the comprehensive evaluation index of the final PPT as QZZ, and calculate the comprehensive evaluation index QZZ. The calculation formula is as follows: ; In the formula, nfd represents the total number of user feedbacks, ijq represents the current feedback number, , represents the user satisfaction quantization value corresponding to the ijq - th user feedback, represents the page content detail adjustment coefficient calculated according to the user operation during the ijq - th user feedback, represents the total number of iterative optimizations for the ijq - th user feedback, represents the probability of optimizing strategy adjustment during the tty - th iteration, ; Among them, set the user feedback number threshold as ; When , stop collecting the number of user feedbacks; When , continue to collect the number of user feedbacks; Calculate the page content detail adjustment coefficient through the front - end interaction technology. The calculation formula is as follows: ; In the formula, represents the quantization value corresponding to the mouse wheel scrolling speed, represents the quantization value corresponding to the zoom operation on the element, represents the quantization value corresponding to the page switching operation, represents the quantization value corresponding to the mouse click frequency, mos represents the number of items of user operation feedback, , , …, are respectively corresponding to , , …, The weight coefficient of , , Corresponding to , , The weight coefficient of right The calculation formula is as follows: ; In the formula, represents the satisfaction evaluation of user feedback after the tyy-1th iteration, represents the difference in user satisfaction scores between two adjacent iterations, where two adjacent iterations represent the tyy-1th and tyy-2th iterations. and Corresponding to and The influence weight coefficient of In each iteration, the optimization strategy is adjusted according to user feedback and the comprehensive evaluation index QZZ to generate the final PPT.

[0014] Technical effects and advantages of the present invention: The present invention uses the semantic association expansion algorithm in natural language processing technology to explore potential related concepts and derived concepts of teaching topics, breaking through the limitation of traditional recognition of only explicit concepts, and providing more comprehensive clues for material retrieval. At the same time, it uses AI technology to build a dynamic knowledge graph to assist in retrieval, and performs intelligent path search and association matching based on the complex relationship between knowledge points. Compared with traditional keyword matching retrieval, it can accurately explore deep and systematic teaching materials, avoid one-sided and scattered retrieval results, greatly enrich the reserve of teaching resources, and help teachers obtain materials that are more in line with teaching needs, thereby improving teaching quality and effectiveness.

[0015] The present invention realizes adaptive screening of materials through reinforcement learning algorithm. The system can dynamically evaluate and screen the retrieved materials according to the feedback data of users' historical use of PPT and current teaching scenario information, such as the grade of teaching objects, subject difficulty requirements, etc. This method can flexibly adjust the material selection strategy according to the actual teaching situation, significantly improve the fit between the materials and teaching objectives, and accurately push appropriate materials for different teaching objects and teaching requirements. Whether it is the explanation of basic concepts or the analysis of complex knowledge, it can provide the most suitable content, meet diversified teaching needs, and improve the pertinence and effectiveness of teaching.

[0016] The present invention adopts a layout dynamic generation model based on content semantics, which can automatically generate a page layout adapted to the semantic weights and mutual relationships of texts, pictures, and charts in teaching materials, change the traditional layout method based on the selection of fixed template types, enhance the coherent display of content, and utilize the collaborative technology of image understanding and text analysis to realize the intelligent combination of texts, pictures, and charts, accurately match and place them, improve the integration degree of page information, and is superior to the traditional simple filling type of element combination. These technologies make the PPT page layout more reasonable, the element combination more coordinated, and the visual effect better, which helps students better understand and absorb teaching content and improve the efficiency of teaching information transmission.

[0017] Based on the user feedback learning mechanism, the present invention can capture the user operation feedback in real time through front-end interaction technology when the user previews the PPT and convert it into an optimization requirement signal to achieve real-time interactive optimization with the user, which is difficult for traditional PPT generation technology to achieve. An iterative optimization model based on reinforcement learning is constructed, with user satisfaction as the reward signal. Through multiple iterative trainings, the PPT continuously approaches the ideal state of the user, improving the stability and reliability of the final quality of the PPT. It can deeply understand and optimize the material content, meet the needs of different scenarios, especially for teaching demonstrations based on textbook content, and is different from the traditional single optimization or model-free optimization method. This interactive optimization process can fully consider the personalized needs and aesthetic preferences of users, ensuring that the generated PPT has higher quality and practicality. Description of the Drawings

[0018] Figure 1 It is a flowchart of the steps of the method of the present invention. Detailed Embodiments

[0019] The following further describes the present invention with specific embodiments.

[0020] Embodiment 1: As Figure 1 shown, a PPT automatic generation method based on AI and a textbook resource library includes the following steps: S1. Through natural language processing technology, parse the user's first input information, identify the core concepts and logical structures, and obtain the parsing result; S2. According to the parsing result, automatically retrieve teaching materials related to the teaching theme from the textbook resource library through AI technology; S3. According to the characteristics of the teaching materials, select the layout and design style from the preset template library, and automatically generate a PPT framework containing text, picture, and chart elements, and fill the teaching materials into the PPT framework to form a preliminary PPT; S4. Use intelligent layout optimization technology to make detailed adjustments and optimizations to the preliminary PPT and generate a preview PPT; S5. Optimize the preview PPT based on the user feedback learning mechanism to generate the final PPT.

[0021] In S1, through natural language processing technology, parse the user's first input information, identify the core concepts and logical structures, and obtain the parsing result. The identification process is as follows: Set the user's first input information as I, where I is text information, and use the adjacency matrix to calculate the parsing function , and the calculation formula is as follows: ; In the formula, C represents the set of core concepts, which is the set composed of the concept elements extracted from I, and L represents the logical structure, which is used to reflect the relationship between core concepts; Set the expanded set of core concepts as , and use the shortest path algorithm and based on a predefined knowledge graph to mine associated concepts and derivative concepts, where V represents the set of nodes in the graph, and E represents the set of edges; The shortest path algorithm can calculate the shortest path between two concepts. The shorter the path, the stronger the semantic association between the concepts. Associated concepts can be mined by finding nodes closer to the target concept; this function can also be implemented using algorithms based on graph traversal (breadth-first search, depth-first search), etc.; Perform calculations on , and the calculation formula is as follows: ; In the formula, u represents the concept variable, which is used to traverse each concept in the set of core concepts C, v represents the semantic variable, which represents any node in the knowledge graph, represents the logical AND operator, represents the existential quantifier in logical symbols, represents the function related to C, represents the union operation symbol, and e represents the edge in the knowledge graph.

[0022] When the input information is multimodal information, set the input multimodal information as , where Q represents the image information, A represents the audio information, and the output fused semantics is SQ. Perform calculations on the fused semantics SQ, and the calculation formula is as follows: ; In the formula, represents the fusion function, , and respectively represent semantic feature extraction of text, image, and audio information, and convert them into corresponding semantic feature vectors. It means that the input M containing multimodal information such as text, images, and audio is passed to this function, and after arithmetic processing, the fused semantics SQ is finally output.

[0023] In S2, according to the parsing result, teaching materials related to the teaching topic are automatically retrieved from the teaching material resource library through AI technology, and the retrieval process is as follows: The core concept set C is represented as , where each concept element corresponds to the core content in the teaching topic, i = 1, 2,..., n, and n represents the total number of concept elements; Define the association degree function to measure the association degree between the concept element and any node v in the knowledge graph. Calculate the association degree function , and the calculation formula is as follows: ; In the formula, represents the shortest path length between the concept element and any node v in the knowledge graph, represents the edge weight adjustment factor, represents the multimodal semantic weight factor, , represents the semantic similarity between the fused semantics SQ and any node v in the knowledge graph; Set the set of teaching materials related to the teaching topic retrieved through the knowledge graph (based on AI) as , and each teaching material corresponds to a set of nodes in the knowledge graph. k = 1, 2,..., m, where m is the total number of teaching materials, and the set of associated knowledge points is ; For each teaching material , calculate its comprehensive association degree score , and the calculation formula is as follows: ; In the formula, represents the weight coefficient of the concept element ; Based on the reinforcement learning algorithm, perform adaptive screening on each teaching material .

[0024] Based on the reinforcement learning algorithm, perform adaptive screening on each teaching material , and the screening process is as follows: For each teaching material , extract its multiple feature attributes to construct a feature vector , kp represents the total number of feature vectors, and the feedback data collected from teachers on the use of materials in the previous PPT constitutes the feedback vector , kq represents the total number of features of the feedback data, and the current teaching scenario information vector is set as , kr represents the total number of features of the teaching scenario information; Set the screening value function to measure the material whether it should be screened and retained in the current teaching scenario. The screening formula is as follows: ; In the formula, represents the activation function, represents the weight coefficient of the material feature attribute , represents the feedback data element 's weight coefficient, represents the teaching scenario information element 's weight coefficient, represents the feature attribute vector 's dimension, represents the dimension of the feedback vector UG, represents the dimension of the current teaching scenario information vector IE, represents the weight coefficient of the reinforcement learning dynamic optimization, represents the semantic matching degree of the fusion semantics SQ and the material knowledge point set ; Through the reinforcement learning algorithm, according to the new feedback data of the teacher and the actual teaching effect feedback information after each generation of PPT, continuously adjust , and 's values, so that the screening value function can more accurately reflect the adaptability of the material in different teaching situations; Set the screening threshold as ; When , it is determined that the material meets the current teaching situation and is retained; When , it is determined that the material does not meet the current teaching situation and is discarded.

[0025] In S3, automatically generate a PPT framework containing text, picture, and chart elements, and fill the teaching materials into the PPT framework to form a preliminary PPT. The formation process is as follows: Set the preliminary PPT to contain multiple pages , ad = 1, 2,..., pd, where pd is the total number of pages of the PPT. Each page Its generation depends on the layout selected from the preset template library and the filling operation of teaching materials. The generation formula is as follows: ; In the formula, represents the operation function of filling teaching materials into the selected layout; Establish a dynamic layout generation model based on content semantics to determine the layout of each page and calculate the fitness score of the mv-th layout style for the pd-th page , and the calculation formula is: ; In the formula, represents the semantic weight of the is-th type of teaching material in the ad-th page, represents the semantic correlation degree between the is-th teaching material and the js-th teaching material in the ad-th page, nf represents the number of types of teaching materials, mv represents the number of layout styles that can be selected for the page layout, represents the semantic weight of the js-th type of teaching material in the ad-th page, represents the fitness factor of the is-th teaching material for the mv-th layout style in the ad-th page, and its value range is between 0 and 1, represents the matching weight factor, , represents the matching degree of integrating semantic SQ and the number of layout styles mv; Finally select the layout style with the maximum value, and the selection formula is as follows: ; Adopt the collaborative technology of image understanding and text analysis to realize the intelligent combination of text, pictures and charts and complete the filling operation.

[0026] Adopt the collaborative technology of image understanding and text analysis to realize the intelligent combination of text, pictures and charts and complete the filling operation. The combination process is as follows: In the operation function, for each group of text zw and picture pz in the ad-th page, set to represent the best combination position and actual proportion information of picture pz relative to text zw in the ad-th page, to represent the theme content vector described by text zw in the ad-th page, to represent the theme content vector contained in picture pz in the ad-th page; Calculate the content fit score between the text zw and the picture pz on the ad-th page , and the calculation formula is as follows: ; Finally, fill all the texts and pictures with determined combined positions and display ratios into the ad-th page according to the layout to complete the generation of the page and generate a complete preliminary PPT.

[0027] In S4, use intelligent layout optimization technology to make detailed adjustments and optimizations to the preliminary PPT and generate a preview PPT. The generation process is as follows: Set the total width of the page to WY, the total height to HY, the horizontal coordinate range occupied by the picture in the page is , and the vertical coordinate range is . Set its width to wy and height to hy. Set the number of visual blocks for text area division to nfi. For the fe-th visual block, fe = 1, 2,..., nfi, its width and height are calculated as follows: ; In the formula, represents the width distribution coefficient; ; In the formula, represents the height distribution coefficient; Set the basic font size of the text to , and calculate the actually displayed font size sa. The calculation formula is as follows: ; In the formula, da represents the text distribution density adjustment coefficient; Set the number of data series in the chart to msf, where msf is an integer. For each data series jf, jf = 1, 2,..., msf, its interactive response function is expressed as follows: If , then ; If , then ; In the formula, ck represents the coordinate of the user's interactive operation position on the page, represents the specific animation display function of the data series jf; A data series refers to a set of data points or data items used to display data in a chart, usually used to represent a specific category, group, or trend. For example, in a bar chart, each bar represents a data series, which is used to show the data values of that series at different categories or time points.

[0028] Set the initial height of the bars in the bar chart data series to , and the target animation display height to , the animation duration to , and the time variable for the animation playback to , where , calculate for the specific animation display function , and the calculation formula is as follows: ; Finally, after re - typesetting, a preview PPT is generated.

[0029] In S5, optimize the preview PPT based on the user feedback learning mechanism to generate the final PPT, and the generation process is as follows: Set the comprehensive evaluation index of the final PPT to QZZ, and calculate the comprehensive evaluation index QZZ. The calculation formula is as follows: ; In the formula, nfd represents the total number of user feedbacks, ijq represents the current number of feedbacks, , represents the user satisfaction quantization value corresponding to the ijq - th user feedback, represents the page content detail adjustment coefficient calculated according to the user operation during the ijq - th user feedback, represents the total number of iterative optimizations for the ijq - th user feedback, represents the probability of optimizing strategy adjustment during the tty - th iteration, ; lim represents the limit; Among them, set the user feedback number threshold to ; When , then stop collecting the number of user feedbacks; When , then continue to collect the number of user feedbacks; Calculate the page content detail adjustment coefficient through the front - end interaction technology. The calculation formula is as follows: ; In the formula, represents the quantization value corresponding to the mouse wheel scrolling speed, represents the quantization value corresponding to the zoom operation on the element Represents the quantization value corresponding to the page switching operation, Represents the quantization value corresponding to the mouse click frequency, and mos represents the number of items of user operation feedback. , , …, are respectively corresponding to , , …, weight coefficients; 、 、 are respectively corresponding to 、 、 weight coefficients; Perform calculations on , and the calculation formula is as follows: ; In the formula, represents the satisfaction evaluation of user feedback after the (tyy - 1)-th iteration, represents the difference in user satisfaction scores between two adjacent iterations. Two adjacent iterations refer to the (tyy - 1)-th and (tyy - 2)-th iterations, and are respectively corresponding to and influence weight coefficients; In each iteration, according to user feedback and the comprehensive evaluation index QZZ, adjust the optimization strategy to generate the final PPT.

[0030] The weight coefficients in this embodiment can be set according to requirements or obtained by a method based on expert knowledge; when calculating each formula, parameters can be normalized or dimensionless processed as needed.

[0031] Embodiment 2: A PPT automatic generation device based on AI and a teaching material resource library, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor. The method in Embodiment 1 is implemented by the processor executing the computer program.

Claims

1. A PPT automatic generation method based on AI and teaching material resource library, characterized in that: The following steps are involved: S1. Analyze the user's first input information through natural language processing technology, identify the core concepts and logical structure, and obtain the analysis results; S2. Based on the analysis results, AI technology is used to automatically retrieve teaching materials related to the teaching topic from the teaching resource library; S3. According to the characteristics of the teaching materials, select the layout and design style from the preset template library, and automatically generate a PPT frame containing text, pictures, and chart elements, fill the teaching materials into the PPT frame, and form a preliminary PPT; S4. Use intelligent layout optimization technology to adjust and optimize the preliminary PPT in detail and generate a preview PPT; S5. Optimize the preview PPT based on the user feedback learning mechanism to generate the final PPT.

2. The method for automatically generating PPT based on AI and teaching material resource library according to claim 1, characterized in that: In S1, the user's first input information is parsed through natural language processing technology to identify the core concepts and logical structure and obtain the parsing results. The identification process is as follows: Assume that the user's first input information is represented as I, where I is text information, and use the adjacency matrix to calculate the analytical function , the calculation formula is as follows: ； In the formula, C represents the core concept set, which represents the set of concept elements extracted from I, and L represents the logical structure, which is used to reflect the relationship between core concepts; Set the expanded core concept set to , through the shortest path algorithm and based on a predefined knowledge graph To mine related concepts and derived concepts, where V represents the node set in the graph and E represents the edge set; right The calculation formula is as follows: ； In the formula, u represents the concept variable, which is used to traverse each concept in the core concept set C, and v represents the semantic variable, which represents any node in the knowledge graph. Represents the logical AND operator, Represents the existential quantifier in logical symbols, Represents functions related to C, represents the union operator, and e represents an edge in the knowledge graph.

3. The method for automatically generating PPT based on AI and teaching material resource library according to claim 2, characterized in that: When the input information is multimodal information, the input multimodal information is set to be expressed as , where Q represents image information, A represents audio information, and the output fusion semantics is represented as SQ. The fusion semantics SQ is calculated using the following formula: ； In the formula, represents the fusion function, , and Respectively represent the semantic feature extraction of text, image and audio information, and convert them into corresponding semantic feature vectors. It means that the input M containing multimodal information such as text, image and audio is passed to the function, and after calculation and processing, the fused semantics SQ is finally output.

4. The method for automatically generating PPT based on AI and teaching material resource library according to claim 3 is characterized in that: In S2, according to the analysis results, the teaching materials related to the teaching topic are automatically retrieved from the teaching material resource library through AI technology. The retrieval process is as follows: The core concept set C is represented as , where each concept element They all correspond to the core content of the teaching topic, i=1, 2,…, n, n represents the total number of concept elements; Define the correlation function To measure the conceptual elements The degree of association with any node v in the knowledge graph, the association function The calculation formula is as follows: ； In the formula, Representing conceptual elements The shortest path length to any node v in the knowledge graph, represents the edge weight adjustment factor, represents the multimodal semantic weight factor, , Represents the semantic similarity between the fused semantics SQ and any node v in the knowledge graph; Set the collection of materials related to the teaching topic retrieved through the knowledge graph as , each material In the knowledge graph, it corresponds to a set of nodes, k=1, 2, ..., m, where m is the total number of materials, and the associated knowledge point set is ; For each material , calculate its comprehensive correlation score , the calculation formula is as follows: ； In the formula, Representing conceptual elements The weight coefficient of Based on the reinforcement learning algorithm, Perform adaptive screening.

5. The method for automatically generating PPT based on AI and teaching material resource library according to claim 4 is characterized in that: The reinforcement learning algorithm is used to Perform adaptive screening. The screening process is as follows: For each material , extract multiple feature attributes to construct feature vectors , kp represents the total number of feature vectors, and collects the teacher's feedback data on the materials used in the previous PPT to form a feedback vector , kq represents the total number of feedback data features, and the current teaching scene information vector is set to , kr represents the total number of teaching scene information features; Setting the filter value function To measure material Whether it should be screened and retained in the current teaching scenario, the screening formula is as follows: ； In the formula, represents the activation function, Represents material feature attributes The weight coefficient of Represents feedback data element The weight coefficient of Indicates the information element of the teaching scene The weight coefficient of Represents the feature attribute vector The dimension of represents the dimension of the feedback vector UG, Represents the dimension of the current teaching scene information vector IE, represents the weight coefficient of dynamic optimization of reinforcement learning, Represents the fusion of semantic SQ and material knowledge point set The semantic matching degree of Through the reinforcement learning algorithm, we continuously adjust the system according to the new feedback data from teachers and the actual teaching effect feedback information after each PPT is generated. , and The value of makes the screening value function More accurately reflect the adaptability of the materials in different teaching situations; Set the screening threshold to ; when When In line with the current teaching situation, it will be retained; when When If it does not meet the current teaching situation, it will be discarded.

6. The method for automatically generating PPT based on AI and teaching material resource library according to claim 4, characterized in that: In S3, a PPT frame including text, pictures, and chart elements is automatically generated, and teaching materials are filled into the PPT frame to form a preliminary PPT. The formation process is as follows: Set up a preliminary PPT with multiple pages , ad=1, 2, ..., pd, where pd is the total number of pages in the PPT, and each page The generation depends on the layout selected from the preset template library And the filling operation of teaching materials, the generation formula is as follows: ； In the formula, Indicates the operation function of filling the teaching materials into the selected layout; Establish a dynamic layout generation model based on content semantics to determine the layout of each page Layout , calculate the fitness score of the mvth layout style for the pdth page , the calculation formula is: ； In the formula, represents the semantic weight of the is-th type of teaching material in the ad-th page, represents the semantic association between the isth teaching material and the jsth teaching material in the adth page, nf represents the number of types of teaching materials, and mv represents the number of layout styles that can be selected for the page layout. represents the semantic weight of the jsth type of teaching material in the adth page, It indicates the fitness factor of the is-th teaching material in the ad-th page for the mv-th layout style, and its value range is between 0 and 1. represents the matching weight factor, , Indicates the matching degree between the fusion semantics SQ and the number of layout styles mv; final choose The largest layout style, the selection formula is as follows: ； The collaborative technology of image understanding and text analysis is used to realize the intelligent combination of text, pictures and charts to complete the filling operation.

7. The method for automatically generating PPT based on AI and teaching material resource library according to claim 6, characterized in that: The image understanding and text analysis collaborative technology is used to realize the intelligent combination of text, pictures and charts to complete the filling operation. The combination process is as follows: exist In the operation function, for each set of text zw and image pz in the ad-th page, set Indicates the optimal combination position and actual ratio information of the image pz relative to the text zw in the ad-th page, represents the subject content vector described by the text zw in the ad-th page, Represents the subject content vector contained in the image pz in the ad-th page; Calculate the content fit score between the text zw and the image pz in the ad-th page , the calculation formula is as follows: ； Finally, all the texts and pictures with determined combination positions and display ratios are laid out according to the layout. Fill in the ad-th page to complete the page Generate a complete preliminary PPT.

8. The method for automatically generating PPT based on AI and teaching material resource library according to claim 7, characterized in that: In S4, the intelligent layout optimization technology is used to adjust and optimize the preliminary PPT in detail, and a preview PPT is generated. The generation process is as follows: Settings Page The total width is WY, the total height is HY, and the horizontal coordinate range of the image on the page is , the vertical coordinate range is , its width is wy, its height is hy, and the number of visual blocks divided into the text area is set to nfi. For the fe-th visual block, fe=1, 2, ..., nfi, its width and height The calculation formula is as follows: ； In the formula, represents the width allocation coefficient; ； In the formula, represents the height distribution coefficient; Set the base text size to , calculate the actual display text size sa, the calculation formula is as follows: ； In the formula, da represents the text distribution density adjustment coefficient; Set the number of data series in the chart to msf, where msf is an integer. For each data series jf, jf=1, 2, ..., msf, its interactive response function It is expressed as follows: if ,but ; if ,but ; In the formula, ck represents the coordinates of the user's interactive operation position on the page, Represents the specific animation display function of the data series jf; Set the initial height of the bar chart data series columns to , the target animation display height is , the animation duration is , the time variable for animation playback is ,in , for specific animation display functions The calculation formula is as follows: ； Finally, after reformatting, a preview PPT is generated.

9. The method for automatically generating PPT based on AI and teaching material resource library according to claim 7, characterized in that: In S5, the preview PPT is optimized based on the user feedback learning mechanism to generate the final PPT. The generation process is as follows: The comprehensive evaluation index of the final PPT is set to QZZ, and the comprehensive evaluation index QZZ is calculated. The calculation formula is as follows: ； In the formula, nfd represents the total number of user feedbacks, ijq represents the number of current feedbacks, , represents the quantified value of user satisfaction corresponding to the ijqth user feedback, It represents the adjustment coefficient of the detail level of the page content calculated based on the user operation at the ijqth user feedback. represents the total number of iterative optimizations for ijq user feedbacks, represents the probability of optimizing the strategy adjustment at the tty-th iteration, ; The threshold of user feedback times is set as ; when , then stop collecting user feedback; when , then continue to collect user feedback times; Adjust the detail level of page content through front-end interactive technology The calculation formula is as follows: ； In the formula, Indicates the quantized value corresponding to the mouse wheel scrolling speed. Represents the quantized value corresponding to the scaling operation of the element. Indicates the quantized value corresponding to the page switching operation, Indicates the quantitative value corresponding to the mouse click frequency, mos indicates the number of user operation feedback items, , , …, Corresponding to , , …, The weight coefficient of , , Corresponding to , , The weight coefficient of right The calculation formula is as follows: ； In the formula, represents the satisfaction evaluation of user feedback after the tyy-1th iteration, represents the difference in user satisfaction scores between two adjacent iterations, where two adjacent iterations represent the tyy-1th and tyy-2th iterations. and Corresponding to and The influence weight coefficient of In each iteration, the optimization strategy is adjusted according to user feedback and the comprehensive evaluation index QZZ to generate the final PPT.

Images