A course design method and system based on artificial intelligence

Abstract

The present invention relates to an artificial intelligence-based course design method and system thereof, and the method steps are: inputting at least one obtained text vector into a Sora model; obtaining a video associated with the text vector derived by the Sora model; utilizing the bidirectionality of the text vector in combination with a database to verify the similarity of the video; receiving voice input and converting it into a text vector, utilizing the Sora model to generate an associated video, and combining the bidirectionality of the text vector with the database to verify the video similarity, so as to ensure that the video content is consistent with the teacher's intention.

Description

Technical Field

[0001] The present invention relates to the field of image artificial intelligence technology, and in particular to an artificial intelligence-based course design method and system thereof. Background Art

[0002] Artificial intelligence (AI) is a branch of computer science that aims to study and develop theories, methods, technologies and application systems that can simulate, extend and expand human intelligence. It combines theories from multiple disciplines such as mathematics, computer science, and psychology, and enables computers to simulate human thinking and behavior processes, realize human-computer interaction, and improve the intelligence level of computers to better serve human society.

[0003] As a coupling product of language model and graph model, Sora overcomes the limitations of traditional language model in the information conversion process. Language model often has a large information loss when describing the physical world, while graph model can more accurately express the relationship between entities; Sora achieves more efficient information conversion and expression by combining these two models.

[0004] The existing course design and teaching have the following drawbacks:

[0005] Lack of practical opportunities and resources: Many schools are unable to provide students with sufficient practical opportunities and resources due to limitations in funding, equipment or teaching staff. This makes it difficult for students to apply what they have learned to actual situations, limiting the development of their practical ability and innovative spirit.

[0006] Therefore, how to use relevant artificial intelligence to overcome the shortcomings of existing curriculum design and teaching is a problem that needs to be solved at present. Summary of the invention

[0007] The main purpose of the present invention is to provide a course design method and system based on artificial intelligence, which receives voice input and converts it into text vectors, generates related videos using the Sora model, and combines the bidirectionality of the text vectors and the database to verify the video similarity to ensure that the video content is consistent with the teacher's intention. Figure 1 To.

[0008] To achieve the above object, the present invention provides a course design method based on artificial intelligence, comprising the following steps:

[0009] Input at least one of the obtained text vectors into the Sora model;

[0010] Obtain a video associated with the text vector exported by the Sora model;

[0011] The bidirectionality of the text vector is used in conjunction with a database to verify the similarity of the videos.

[0012] Furthermore, before the step of inputting the obtained at least one text vector into the Sora model, the following steps are included:

[0013] Acquire the voice data input by the user through the recording device;

[0014] The speech data is input into a language model, and the speech data is vectorized by the language model to obtain at least one text vector.

[0015] Furthermore, the step of vectorizing the speech data using the language model includes:

[0016] Using the language filter configured in the language model to identify the language data, so as to identify nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model;

[0017] Convert the audio that can be recognized after being filtered by the language filter into text to generate corresponding text data, wherein the text data is composed of a plurality of words arranged;

[0018] Importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer;

[0019] The output layer of the language model outputs the vectorized encoded text vector.

[0020] Furthermore, the step of inputting the obtained at least one text vector into the Sora model includes:

[0021] Connecting the output layer interface of the language model to the word filling interface of the sora model;

[0022] The text vector is interpreted according to the forward vector direction, and each word obtained after the interpretation is input into the Sora model through the word filling interface.

[0023] Furthermore, the step of utilizing the bidirectionality of the text vector in combination with a database to verify the similarity of the video includes:

[0024] Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame;

[0025] Align the positive text vector with the long axis of the frame;

[0026] Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area;

[0027] By means of the classification, the long axis of the frame is divided into various regions;

[0028] Append the reverse text vector to the XY structure;

[0029] As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified;

[0030] If they can overlap, the video is recognized to have no ambiguity error and the video is output.

[0031] The present invention also proposes a course design system based on artificial intelligence, the system comprising:

[0032] A Sora interface module, used for inputting the obtained at least one text vector into the Sora model;

[0033] A video acquisition module, used to acquire the video associated with the text vector exported by the Sora model;

[0034] The video verification module is used to verify the similarity of the video by utilizing the bidirectionality of the text vector in combination with a database.

[0035] Furthermore, the system comprises:

[0036] The recording module is used to obtain the voice data input by the user through the recording device;

[0037] The CNN interface module is used to input the speech data into a language model, and vectorize the speech data through the language model to obtain at least one text vector.

[0038] Furthermore, the CNN interface module includes:

[0039] a recognition unit, configured to recognize the language data using the language filter configured in the language model, so as to recognize nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model;

[0040] A text unit, used to convert the audio that can be recognized after being filtered by the language filter into text, and generate corresponding text data, wherein the text data is composed of a plurality of words arranged;

[0041] A function unit, used for importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer;

[0042] The vector unit is used to output the vectorized encoded text vector from the output layer of the language model.

[0043] Furthermore, the Sora interface module includes:

[0044] Sora interface module, used to connect the output layer interface of the language model with the word filling interface of the sora model;

[0045] The interpretation module is used to interpret the text vector according to the forward vector direction, and input each word obtained after the interpretation into the Sora model through the word filling interface.

[0046] Furthermore, the video verification module includes:

[0047] Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame;

[0048] Align the positive text vector with the long axis of the frame;

[0049] Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area;

[0050] By means of the classification, the long axis of the frame is divided into various regions;

[0051] Append the reverse text vector to the XY structure;

[0052] As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified;

[0053] If they can overlap, the video is recognized to have no ambiguity error and the video is output.

[0054] The course design method and system based on artificial intelligence provided by the present invention have the following beneficial effects:

[0055] (1) The noun, verb, and adjective recognition functions in the language model can accurately capture the key information in the user's voice, ensuring that the subsequently generated video content closely matches the user's original intention.

[0056] (2) By utilizing the bidirectionality of text vectors and combining them with a database to verify the similarity of videos, the present invention can effectively reduce the problem of inconsistency between the video content and the teacher's actual imagination caused by possible ambiguity during the teacher's explanation. This verification mechanism ensures that the generated video more accurately reflects the teacher's thoughts and intentions.

[0057] (3) By intelligently classifying videos and dividing frame areas, the present invention can more efficiently manage and utilize video resources and reduce unnecessary computing and storage overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1It is a flowchart of a course design method based on artificial intelligence in one embodiment of the present invention;

[0059] Figure 2 is a structural block diagram of a course design system based on artificial intelligence in one embodiment of the present invention;

[0060] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0062] Reference Figure 1 , which is a flow chart of a course design method based on artificial intelligence proposed by the present invention, comprising the following steps:

[0063] S1, input at least one text vector obtained into the Sora model;

[0064] S2, obtaining a video associated with the text vector derived by the Sora model;

[0065] S3, using the bidirectionality of the text vector in combination with a database to verify the similarity of the videos.

[0066] In one embodiment, before the step of inputting the obtained at least one text vector into the Sora model, the method includes:

[0067] Acquire the voice data input by the user through the recording device;

[0068] The speech data is input into a language model, and the speech data is vectorized by the language model to obtain at least one text vector.

[0069] In one embodiment, the step of vectorizing the speech data using the language model includes:

[0070] Using the language filter configured in the language model to identify the language data, so as to identify nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model;

[0071] Convert the audio that can be recognized after being filtered by the language filter into text to generate corresponding text data, wherein the text data is composed of a plurality of words arranged;

[0072] Importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer;

[0073] The output layer of the language model outputs the vectorized encoded text vector.

[0074] Specifically, first, the language filter configured in the neural network language model is used to preprocess and recognize the input voice data. The language filter is designed to accurately recognize key words such as nouns, verbs, and adjectives from the voice data. These words usually carry the main information and intention of the sentence and are an important basis for the subsequent generation of video content.

[0075] Next, the audio part that is filtered and identified by the language filter is converted into text. This step usually involves speech recognition technology, which can convert the audio signal into corresponding text data. These text data are composed of a series of words arranged in a certain order, which together express the information and meaning of the original speech.

[0076] Then, the obtained text data is imported into the hidden layer of the neural network language model. The hidden layer is an important component of the neural network, which is responsible for deep processing and handling of the input data. In this process, the nonlinear function in the hidden layer will perform forward and reverse vector encoding on the text data. Forward vector encoding refers to encoding in the order of words in the text data, while reverse vector encoding is encoding in the opposite order. This bidirectional encoding method can capture more contextual information and semantic relationships in the text data.

[0077] Finally, the output layer of the neural network language model outputs the vectorized encoded text vectors. These text vectors are vectorized representations of the original speech data. They contain the key information and semantic features of the original speech and can be used for subsequent tasks such as video generation and verification.

[0078] in:

[0079] In neural networks, common nonlinear functions include Sigmoid, Tanh, ReLU (Rectified Linear Unit), etc. These functions have their own characteristics and are suitable for different scenarios. For example, the Sigmoid function maps the output of a neuron to between 0 and 1, which is suitable for binary classification problems; the ReLU function outputs 0 for negative input values ​​and directly outputs the value for positive input values.

[0080] Input layer: First, the text data (the sequence of words after recognition and conversion) is input into the input layer of the neural network.

[0081] Hidden layer: In the hidden layer, each neuron receives inputs from the input layer and calculates an intermediate value through weighted summation. Then, this intermediate value is passed to a non-linear function for processing. The role of the non-linear function is to generate a new value based on this intermediate value, and this new value will be used as the output of the neuron. Through the combination of multiple such neurons and layers, the neural network can gradually learn and extract complex features from the text data.

[0082] Output layer: Finally, in the output layer, the neural network further processes the output of the hidden layer and converts it into the required vector form. These vectors encode the semantic information and structural features in the original text data and can be used for subsequent tasks such as video generation or similarity verification.

[0083] For a language filter, it is a component in the field of natural language processing and text processing, used to screen, clean, or transform text data to meet specific requirements or improve the effect of subsequent processing.

[0084] The main function of the language filter is to preprocess and analyze the input text or speech data and can perform the following tasks:

[0085] Lexical recognition: Identify specific types of words in the text, such as nouns, verbs, adjectives, etc. These words usually contain important semantic information and are crucial for understanding the content and intention of the text.

[0086] Stop word removal: Remove common words in the text that are not helpful for understanding the content, such as "de", "le", "zai", etc. These stop words appear frequently in the text but usually do not carry key information.

[0087] Stemming / lemmatization: Restore words to their basic form or root for unified processing and comparison. For example, restore "running" to "run".

[0088] Text cleaning: Remove errors, non-standard characters, special symbols, extra spaces, etc. in the text to improve the quality and consistency of the text.

[0089] Sentiment analysis preprocessing: In sentiment analysis tasks, the language filter may be used to identify and filter sentiment-related words or expressions.

[0090] Multilingual processing: For multilingual texts, the language filter can identify texts in different languages and perform corresponding processing.

[0091] The working principle of the language filter involves the following steps:

[0092] Input Processing: First, the voice data is converted into text form (through speech recognition technology). Then, this text is input into the language filter.

[0093] Vocabulary Identification: Language filters use predefined rules, dictionaries, or machine learning models to identify nouns, verbs, and adjectives in text. These rules may be based on grammatical properties of vocabulary, contextual information, or statistical patterns.

[0094] Filtering and output: The recognized words are filtered out and used as the output of the filter. These filtered words will be used in the subsequent text vectorization or video generation process.

[0095] Optimization and Adjustment: The performance and accuracy of language filters can be improved through continuous training and adjustment. For example, more recognition rules can be added, dictionaries can be updated, or more advanced machine learning algorithms can be used to improve the effect of the filter.

[0096] In one embodiment, the step of inputting the obtained at least one text vector into the Sora model includes:

[0097] Connecting the output layer interface of the language model to the word filling interface of the sora model;

[0098] The text vector is interpreted according to the forward vector direction, and each word obtained after the interpretation is input into the Sora model through the word filling interface.

[0099] In one embodiment, the step of utilizing the bidirectionality of the text vector in combination with a database to verify the similarity of the video includes:

[0100] Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame;

[0101] Align the positive text vector with the long axis of the frame;

[0102] Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area;

[0103] By means of the classification, the long axis of the frame is divided into various regions;

[0104] Append the reverse text vector to the XY structure;

[0105] As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified;

[0106] If they can overlap, the video is recognized to have no ambiguity error and the video is output.

[0107] In one embodiment, the step of using the bidirectionality of text vectors combined with a database to verify video similarity is a complex but critical process. Each step is explained in detail below:

[0108] First, in order to handle the relationship between video and text vectors, a three-dimensional data structure needs to be created. In this structure, the XY axis is used to construct a frame of the video, that is, the width and height of each frame; and the Z axis is the frame length axis, representing the time dimension of the video, that is, the number of frames or playback time of the video.

[0109] Next, the forward text vector (the vector encoded in the original text order) is overlapped with the frame long axis. This means that each element or feature of the text vector corresponds to a frame or a period of time in the video. This correspondence is the key to understanding the content of the video and verifying the similarity of the video.

[0110] The video is then classified based on the words in the forward text vector. This means dividing the video into multiple frames, each of which corresponds to a word in the text vector. The purpose of this step is to establish a detailed correspondence between the text and the video so that the subsequent verification process can be carried out accurately.

[0111] Based on the classification, the frame long axis is further divided. This means that according to the structure and content of the text vector, the time axis of the video is divided into multiple regions, each of which corresponds to one or more elements in the text vector. This division helps to match the content of the text and video more accurately.

[0112] Next, the reverse text vector (a vector encoded in the opposite order of the original text) is appended to the XY architecture. This step is to use the bidirectionality of the text vector to capture possible ambiguities or errors in the video. The reverse text vector provides a way to look at the video content from the opposite perspective, which helps to discover information that the forward text vector may overlook.

[0113] As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified. This means that during the video playback, the elements of the reverse text vector are constantly compared with the corresponding frame area to check whether they are consistent or similar. If they are consistent or similar, it means that the video content is consistent with the text vector of that part; otherwise, there may be ambiguity or error.

[0114] Finally, if they can overlap, that is, the comparison results of the reverse text vector and each divided frame area are consistent or highly similar, the video is recognized without ambiguity and error, and the video is output. The verified video content matches the bidirectionality of the original text vector, accurately expressing the intent and content of the original text. If they do not overlap or have obvious errors, further processing or adjustment of the video content may be required to match the intent of the original text.

[0115] like Figure 2 , a course design system based on artificial intelligence, the system includes:

[0116] A Sora interface module, used for inputting the obtained at least one text vector into the Sora model;

[0117] A video acquisition module, used to acquire the video associated with the text vector exported by the Sora model;

[0118] The video verification module is used to verify the similarity of the video by utilizing the bidirectionality of the text vector in combination with a database.

[0119] The system includes:

[0120] The recording module is used to obtain the voice data input by the user through the recording device;

[0121] The CNN interface module is used to input the speech data into a language model, and vectorize the speech data through the language model to obtain at least one text vector.

[0122] The CNN interface module includes:

[0123] a recognition unit, configured to recognize the language data using the language filter configured in the language model, so as to recognize nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model;

[0124] A text unit, used to convert the audio that can be recognized after being filtered by the language filter into text, and generate corresponding text data, wherein the text data is composed of a plurality of words arranged;

[0125] A function unit, used for importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer;

[0126] The vector unit is used to output the vectorized encoded text vector from the output layer of the language model.

[0127] Sora interface modules include:

[0128] Sora interface module, used to connect the output layer interface of the language model with the word filling interface of the sora model;

[0129] The interpretation module is used to interpret the text vector according to the forward vector direction, and input each word obtained after the interpretation into the Sora model through the word filling interface.

[0130] The video verification module includes:

[0131] Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame;

[0132] Align the positive text vector with the long axis of the frame;

[0133] Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area;

[0134] By means of the classification, the long axis of the frame is divided into various regions;

[0135] Append the reverse text vector to the XY structure;

[0136] As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified;

[0137] If they can overlap, the video is recognized to have no ambiguity error and the video is output.

[0138] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A course design method based on artificial intelligence, characterized in that: The following steps are involved: Input at least one of the obtained text vectors into the Sora model; Obtain a video associated with the text vector exported by the Sora model; Verify the similarity of the videos by using the bidirectionality of the text vector in combination with a database; The step of utilizing the bidirectionality of the text vector in combination with a database to verify the similarity of the video includes: Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame; Align the positive text vector with the long axis of the frame; Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area; By means of the classification, the long axis of the frame is divided into various regions; Append the reverse text vector to the XY structure; As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified; If they can overlap, the video is recognized to have no ambiguity error and the video is output.

2. The course design method based on artificial intelligence according to claim 1, characterized in that: Before the step of inputting the obtained at least one text vector into the Sora model, the method includes: Acquire the voice data input by the user through the recording device; The speech data is input into a language model, and the speech data is vectorized by the language model to obtain at least one text vector.

3. The course design method based on artificial intelligence according to claim 2, characterized in that: The step of vectorizing the speech data by using the language model comprises: Using the language filter configured in the language model to identify the language data, so as to identify nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model; Convert the audio that can be recognized after being filtered by the language filter into text to generate corresponding text data, wherein the text data is composed of a plurality of words arranged; Importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer; The output layer of the language model outputs the vectorized encoded text vector.

4. The course design method based on artificial intelligence according to claim 2, characterized in that: The step of inputting the obtained at least one text vector into the Sora model comprises: Connecting the output layer interface of the language model to the word filling interface of the sora model; The text vector is interpreted according to the forward vector direction, and each word obtained after the interpretation is input into the Sora model through the word filling interface.

5. A course design system based on artificial intelligence, characterized in that: The system includes: A Sora interface module, used for inputting the obtained at least one text vector into the Sora model; A video acquisition module, used to acquire the video associated with the text vector exported by the Sora model; A video verification module, used to verify the similarity of the video by utilizing the bidirectionality of the text vector in combination with a database; The video verification module includes: Create a three-dimensional data structure; wherein the three-dimensional data structure includes: XY is used to construct a frame of video, and Z is the long axis of the frame; Align the positive text vector with the long axis of the frame; Classifying the video; wherein the classification is to divide the video according to each word of the forward text vector, so as to distinguish the words in the forward text vector corresponding to each frame area; By means of the classification, the long axis of the frame is divided into various regions; Append the reverse text vector to the XY structure; As the time of the frame long axis moves, the coincidence of the reverse text vector and each divided frame area is gradually verified; If they can overlap, the video is recognized to have no ambiguity error and the video is output.

6. The course design system based on artificial intelligence according to claim 5, characterized in that: The system includes: The recording module is used to obtain the voice data input by the user through the recording device; The CNN interface module is used to input the speech data into a language model, and vectorize the speech data through the language model to obtain at least one text vector.

7. The course design system based on artificial intelligence according to claim 6, characterized in that: The CNN interface module includes: a recognition unit, configured to recognize the language data using the language filter configured in the language model, so as to recognize nouns, verbs and adjectives in the language data; wherein the language model is a neural network language model; A text unit, used to convert the audio that can be recognized after being filtered by the language filter into text, and generate corresponding text data, wherein the text data is composed of a plurality of words arranged; A function unit, used for importing the text data into a hidden layer in a language model, so as to perform forward and reverse vectorization encoding on the text data through a nonlinear function in the hidden layer; The vector unit is used to output the vectorized encoded text vector from the output layer of the language model.

8. The course design system based on artificial intelligence according to claim 6, characterized in that: Sora interface modules include: Sora interface module, used to connect the output layer interface of the language model with the word filling interface of the sora model; The interpretation module is used to interpret the text vector according to the forward vector direction, and input each word obtained after the interpretation into the Sora model through the word filling interface.

Images