Ai in scaffolding learning method

Abstract

An interactive learning method is used to guide a user to think and obtain an answer to a question, which is implemented by connecting a server device of a user device through a signal, and includes：(A) obtaining at least one target knowledge data related to the answer according to a question string related to the question; (B) obtaining an answer string related to the answer; (C) generating and sending a question string to the user device using a LLM based on the at least one target knowledge data; (D) determining whether a first response string related to the question string is the answer; (E) if not, using the LLM to generate and send a prompt string to the user device; (F) determining whether a second response string related to the prompt string is the answer; (G) If not, send the answer string to the user device.

Claims

1. An AI scaffolding learning method is suitable for guiding a user to think about and obtain an answer to a question, and is implemented by connecting a server device to a user device operated by the user via a signal. The server device stores a large language model and multiple knowledge data corresponding to multiple knowledge texts. The AI ​​scaffolding learning method includes the following steps: (A) after receiving a question string related to the question from the user device, the server device obtains at least one target knowledge data related to the answer from the knowledge data based on the question string; (B) using the large language model, the server device obtains an answer string related to the answer based on the at least one target knowledge data; (C) The server device generates and transmits a question string for guiding the user to think about the answer to the user device based on the at least one target knowledge data and the large language model, so that the user device generates and transmits a first response string related to the user's response to the question string to the server device, and the question string is not the answer to the question string; (D) The server device determines whether the first response string is the answer to the question string using the large language model; (E) When it is determined that the first response string is not the answer to the question string, the server device generates and transmits a prompt string for guiding the user to think about the answer to the user device based on the first response string and the at least one target knowledge data and the large language model, so that the user device generates and transmits a second response string related to the user's response to the prompt string to the server device, and the prompt string is not the answer to the question string; (F) The server device uses the large language model to determine whether the second reply string is the answer to the question string; and (G) when it is determined that the second reply string is not the answer to the question string, the server device transmits the answer string to the user device.

2. The AI ​​scaffolding learning method of claim 1, wherein: Before step (A), the method further includes the following steps: (H) for each knowledge data, obtaining knowledge graph data corresponding to the knowledge data by the server-side device according to the knowledge data; After step (G), the method further includes the following steps: (I) determining whether the at least one target knowledge data is relevant by the server-side device according to the at least one knowledge graph data corresponding to the at least one target knowledge data; (J) when it is determined that the at least one target knowledge data is relevant, dividing the at least one target knowledge data into at least one cluster by the server-side device according to the at least one knowledge graph data corresponding to the at least one target knowledge data; and (K) for each cluster, generating and transmitting a segmented string about the at least one target knowledge data contained in the cluster to the user-side device by the server-side device according to the at least one target knowledge data contained in the cluster and utilizing the large language model.

3. The AI ​​scaffolding learning method of claim 2, wherein: In step (I), the server device determines whether the at least one target knowledge data is relevant by determining whether there is any relationship between multiple entities contained in the at least one knowledge graph data corresponding to the at least one target knowledge data, each relationship corresponds to two target entities and is used to indicate that the two target entities are relevant, and each target entity is one of the entities.

4. The AI ​​scaffolding learning method of claim 2, wherein: After step (I), the following steps are also included: (L) when it is determined that the at least one target knowledge data does not have relevance, the server device uses the large language model to generate and transmit an auxiliary memory string to the user device for assisting the user in memorizing the at least one knowledge text corresponding to the at least one target knowledge data based on the at least one knowledge graph data corresponding to the at least one target knowledge data.

5. The AI ​​scaffolding learning method of claim 4, wherein: Step (L) further includes the following steps: (L-1) when it is determined that the at least one target knowledge data does not have relevance, the server-end device determines whether the number of words in the at least one target knowledge data is greater than a preset number of words; (L-2) when it is determined that the number of words in the at least one target knowledge data is greater than the preset number of words, the server-end device generates and transmits the auxiliary memory string to the user-end device based on the at least one knowledge graph data corresponding to the at least one target knowledge data, using the large language model, and the auxiliary memory string includes a storyline; and (L-3) when it is determined that the number of words in the at least one target knowledge data is not greater than the preset number of words, the server-end device generates and transmits the auxiliary memory string to the user-end device based on the at least one knowledge graph data corresponding to the at least one target knowledge data, using the large language model and a mnemonic method.

6. The AI ​​scaffolding learning method of claim 1, wherein the server device further stores a plurality of positive words for encouraging the user and a plurality of feedback categories with different feedback styles, each feedback category comprising a plurality of feedback strings, wherein: Prior to step (C), the following steps are further included: (M) after receiving the question string, the server-side device utilizes a semantic sentiment analysis algorithm based on the question string to obtain a first analysis result indicating whether the user has a negative emotion; Step (C) further includes the following steps: (C-1) when the first analysis result indicates that the user does not have a negative emotion, the server-side device utilizes the large language model based on any feedback string in a currently set feedback category and the at least one target knowledge data to generate and transmit to the user-side device the question string belonging to a feedback style corresponding to the current feedback category and used to guide the user to think about obtaining the answer, wherein the current feedback category is one of the feedback categories; and (C-2) When the first analysis result indicates that the user has a negative emotion, the server-end device, based on any positive word, any feedback string in a first feedback category, and the at least one target knowledge data, uses the large language model to generate and transmit to the user-end device the question string for encouraging the user and belonging to the feedback style corresponding to the first feedback category and for guiding the user to think to obtain the answer, wherein the first feedback category is one of the feedback categories other than the current feedback category; and between steps (C) and (D), the following steps are also included: (N) using the server-end device to use the first feedback category as the new current feedback category.

7. The AI ​​scaffolding learning method according to claim 6, wherein the user device is further configured to periodically capture the user's face and transmit multiple facial images of the user's face to the server device, wherein: Step (M) further includes the following steps: (M-1) after receiving the question string, the server device uses the semantic sentiment analysis algorithm to obtain a first text analysis result indicating whether the user has a negative emotion based on the question string; (M-2) the server device obtains a plurality of first facial images to be analyzed from the facial images, and the first facial images to be analyzed are facial images received by the server device within a first preset time interval before receiving the question string; (M-3) based on the first facial images to be analyzed, the server device uses a facial recognition algorithm to obtain a first image analysis result indicating whether the user has a negative emotion; and (M-4) based on the first text analysis result and the first image analysis result, the server device obtains the first analysis result.

8. The AI ​​scaffolding learning method of claim 1, wherein the server device further stores a plurality of positive words for encouraging the user and a plurality of feedback categories with different feedback styles, each feedback category comprising a plurality of feedback strings, wherein: Prior to step (E), the following steps are also included: (O) obtaining, by the server-side device, a second analysis result indicating whether the user has a negative emotion based on the first reply string using a semantic sentiment analysis algorithm; Step (E) further includes the following steps: (E-1) when it is determined that the first reply string is not the answer to the question string and the second analysis result indicates that the user does not have a negative emotion, generating and transmitting, by the server-side device, a prompt string belonging to a feedback style corresponding to the current feedback category and used to guide the user to think and obtain the answer based on any feedback string in a currently set current feedback category, the first reply string, and the at least one target knowledge data using the large language model, and transmitting to the user-side device; (E-2) When it is determined that the first reply string is not the answer to the question string and the second analysis result indicates that the user has a negative emotion, the server-end device, based on any positive word, any feedback string in a second feedback category, the first reply string, and the at least one target knowledge data, uses the large language model to generate and transmit to the user-end device the prompt string for encouraging the user and belonging to the feedback style corresponding to the current feedback category and for guiding the user to think to obtain the answer, wherein the second feedback category is one of the feedback categories other than the current feedback category; and between steps (E) and (F), the following steps are also included: (P) using the server-end device to use the second feedback category as the new current feedback category.

9. The AI ​​scaffolding learning method of claim 8, wherein the user device is further configured to periodically capture the user's face and transmit multiple facial images of the user's face to the server device, wherein: Step (O) further includes the following steps: (O-1) obtaining, by the server device, a second text analysis result indicating whether the user has a negative emotion based on the first reply string and using the semantic sentiment analysis algorithm; (O-2) obtaining, by the server device, a plurality of second facial images to be analyzed from the facial images, the second facial images to be analyzed being facial images received by the server device within a second preset time interval before receiving the first reply string; (O-3) obtaining, by the server device, a second image analysis result indicating whether the user has a negative emotion based on the second facial images to be analyzed and using the facial recognition algorithm; and (O-4) obtaining the second analysis result based on the second text analysis result and the second image analysis result.

10. The AI ​​scaffolding learning method of claim 1, wherein the server device further stores a learning record of the user, a plurality of learning suggestion strings corresponding to a plurality of different learning styles, and a style classification algorithm for marking the user as having one of the learning styles, wherein: The applied AI scaffolding learning method also includes the following steps: (Q) using the server-end device, after receiving a questionnaire response data regarding a learning status questionnaire from the user-end device, based on the questionnaire response data and the learning record, using the style classification algorithm, marking the user as a target learning style, where the target learning style is one of the learning styles; and (R) using the server-end device, based on the learning plan suggestion string corresponding to the target learning style, using the large language model, generating and transmitting a learning reminder string for reminding the user of things to pay attention to in learning to the user-end device.

Images