Multi-semantic theme fusion smart education cognitive terminal control method and device

Abstract

The embodiment of the invention discloses a multi-semantic theme fused smart education cognitive terminal control method and device. A specific embodiment of the method comprises the following steps: preprocessing an original education dialogue text set to obtain a processed text set and an evaluated text set; performing topic clustering on the processed text set to obtain a sub-topic text set; performing multi-topic fusion on the sub-topic text set to obtain a topic text set; updating the initial education cognition model to obtain an education cognition model; generating a real-time theme identification set based on the education cognition model and the real-time education conversation; based on the evaluation text set, determining theme indexes of the real-time theme identification set; and according to the real-time theme identification set, generating an education control instruction, and according to the education control instruction, controlling the intelligent education cognition terminal to adjust the output content and the teaching aid transmission device to perform teaching aid transmission. According to the embodiment, response time can be reduced, and wrong transmission of teaching aids is avoided.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a smart education cognitive terminal control method and apparatus for multi-semantic topic fusion. Background Technology

[0002] With the rapid development of artificial intelligence technology, in smart education scenarios, personalized teaching control can be achieved by analyzing educational dialogue text and automatically identifying multiple knowledge points or topic tags from the dialogue. Current technologies often employ methods such as single-label classification or text processing methods using language models pre-trained on specific datasets (e.g., DeepSeek) to identify dialogue intent and then generate control instructions to adjust the output content.

[0003] However, when using the above methods to identify dialogue intent, the following technical problems often arise: Single-label classification struggles to handle complex semantic scenarios in educational dialogues, leading to incomplete knowledge point recognition and consequently, incorrect transmission of teaching aids. Furthermore, the model training requires a large amount of high-quality labeled data, and feature extraction consumes significant computational resources, resulting in slow terminal response speeds and an inability to meet the low-latency requirements of real-time interaction.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion that follows. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a smart education cognitive terminal control method, apparatus, electronic device, and computer-readable medium for multi-semantic topic fusion to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a control method for a smart education cognitive terminal with multi-semantic topic fusion. The method includes: in response to the smart education cognitive terminal receiving an original educational dialogue text set, preprocessing the original educational dialogue text set to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device; performing topic clustering on the processed text set to obtain a sub-topic text set; performing multi-topic fusion on the sub-topic text sets to obtain a topic text set; updating an initial educational cognitive model based on the topic text set to obtain an educational cognitive model; in response to the smart education cognitive terminal receiving a real-time educational dialogue from a target user, generating a real-time topic identifier set based on the educational cognitive model and the real-time educational dialogue; determining topic indicators of the real-time topic identifier set based on the evaluation text set; in response to the topic indicators exceeding a preset indicator threshold, generating an educational control command based on the real-time topic identifier set; and controlling the smart education cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids based on the educational control command.

[0008] Secondly, some embodiments of this disclosure provide a smart education cognitive terminal control device for multi-semantic topic fusion. The device includes: a preprocessing unit configured to preprocess the original educational dialogue text set in response to the smart education cognitive terminal receiving the original educational dialogue text set, to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device; a topic clustering unit configured to perform topic clustering on the processed text set to obtain a sub-topic text set; a multi-topic fusion unit configured to perform multi-topic fusion on the sub-topic text set to obtain a topic text set; and an updating unit configured to update the topic text set according to the topic text set. The system comprises: a text set for updating the initial educational cognitive model to obtain an educational cognitive model; a generation unit configured to generate a real-time topic identifier set based on the educational cognitive model and the real-time educational dialogue received by the smart educational cognitive terminal from the target user; a determination unit configured to determine the topic indicators of the real-time topic identifier set based on the evaluation text set; and a second generation unit configured to generate educational control instructions based on the real-time topic identifier set in response to the topic indicators exceeding a preset indicator threshold, and to control the smart educational cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids based on the educational control instructions.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0011] The various embodiments of this disclosure have the following beneficial effects: the multi-semantic topic fusion smart education cognitive terminal control method of some embodiments of this disclosure can shorten the response time and avoid the erroneous transmission of teaching aids. Specifically, the reasons for long response time and erroneous transmission of teaching aids are: single-label classification makes it difficult to handle scenarios with complex semantic levels in educational dialogues, resulting in incomplete knowledge point recognition and thus erroneous transmission of teaching aids; and model training requires a large amount of high-quality labeled data and feature extraction consumes a large amount of computing resources, resulting in slow terminal response speed and inability to meet the low latency requirements of real-time interaction. Based on this, the multi-semantic topic fusion smart education cognitive terminal control method of some embodiments of this disclosure firstly, in response to the smart education cognitive terminal receiving the original educational dialogue text set, preprocesses the original educational dialogue text set to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device. This can reduce the problem of wasted computing resources caused by data redundancy. Secondly, topic clustering is performed on the processed text set to obtain a sub-topic text set. This can identify potential knowledge points in the educational dialogue and solve the problem of incomplete identification. Next, the aforementioned sub-topic text sets are fused into a multi-topic text set to obtain a topic text set. This enhances the semantic relevance of the dialogue. Then, based on the topic text set, the initial educational cognitive model is updated to obtain an educational cognitive model. This allows for model updates with only a few parameters, improving training efficiency and addressing the slow response time issue. Subsequently, in response to the smart educational cognitive terminal receiving a real-time educational dialogue from the target user, a real-time topic identifier set is generated based on the educational cognitive model and the real-time educational dialogue. This allows for topic identification of the real-time dialogue. Then, based on the aforementioned evaluation text set, the topic indicators of the real-time topic identifier set are determined. This reduces the risk of subsequent misoperations through accurate identification results. Finally, in response to the topic indicators exceeding a preset threshold, educational control instructions are generated based on the real-time topic identifier set. These instructions then control the smart educational cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids. This converts the identification results into controllable instructions, ultimately reducing response time and preventing erroneous transmission of teaching aids. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the intelligent education cognitive terminal control method based on the multi-semantic topic fusion of this disclosure; Figure 2 This is a schematic diagram of the structure of some embodiments of the intelligent education cognitive terminal control device for multi-semantic topic fusion according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1A flowchart 100 is shown illustrating some embodiments of a smart education cognitive terminal control method based on multi-semantic topic fusion according to the present disclosure. This smart education cognitive terminal control method based on multi-semantic topic fusion includes the following steps: Step 101: In response to the smart education cognitive terminal receiving the original educational dialogue text set, the original educational dialogue text set is preprocessed to obtain the processed text set and the evaluation text set.

[0021] In some embodiments, the execution entity (e.g., a server) of the multi-semantic topic fusion smart education cognitive terminal control method can preprocess the original educational dialogue text set received by the smart education cognitive terminal to obtain a processed text set and an evaluation text set. The smart education cognitive terminal may include an interactive display, audio equipment, and an associated teaching aid transmission device. The teaching aid transmission device may be a physical device capable of selecting, displaying, and moving physical teaching aids. The physical teaching aids may be subject-specific experimental equipment, teaching demonstration model experimental equipment, or models. The physical device may be a robotic arm or a conveyor belt. The original educational dialogue text in the original educational dialogue text set may be dialogue records from different channels, such as chat logs from online education platforms or classroom transcription texts.

[0022] In practice, the aforementioned implementing entities can preprocess the original educational dialogue text set using a preset ratio (e.g., 8:2) to obtain a processed text set and an evaluation text set. The processed text set and the evaluation text set can be obtained using the following formula: .

[0023] in, This refers to a collection of original educational dialogue texts. This indicates the evaluation text set. This indicates that the text set is being processed.

[0024] In some optional implementations of certain embodiments, the aforementioned executing entity may preprocess the original educational dialogue text set through the following steps to obtain a processed text set and an evaluation text set: Step one involves standardizing the original educational dialogue text set to obtain a standard text set. This standardization process may include, but is not limited to, removing irrelevant symbols, replacing privacy information, and text normalization. Irrelevant symbols can be characters unrelated to semantics, including but not limited to spaces, carriage returns, and garbled characters. Privacy information may include, but is not limited to, mobile phone numbers, ID card numbers, and names. As an example, text normalization may convert full-width characters to half-width characters and convert traditional Chinese characters to simplified Chinese characters.

[0025] Step two involves semantic segmentation of the aforementioned standard text set to obtain a sequence of semantic units. In practice, the executing entity can use a text chunking algorithm to perform semantic segmentation on the aforementioned standard text set to obtain a sequence of semantic units. This sequence of semantic units can be derived by dividing continuous educational dialogue text into complete, independent, minimal dialogue units.

[0026] Step 3: Based on the aforementioned semantic unit sequence, generate a scene feature set. In practice, the aforementioned executing entity can use named entity recognition technology to identify the topic of each semantic unit in the semantic unit sequence, obtaining an educational terminology set and a corresponding dialogue behavior label set. The scene feature set includes the educational terminology set and the dialogue behavior label set. The educational terms in the educational terminology set can represent professional vocabulary in the field of education. The dialogue behavior label set can represent the behavioral categories of the dialogue (such as asking questions, answering questions, and providing feedback).

[0027] Step four involves stratified sampling of the aforementioned scene feature set to obtain the processing text set and the evaluation text set. In practice, the executing entity can use the StratifiedShuffleSplit class in the Scikit-Learn library to perform stratified sampling of the scene feature set to obtain the processing text set and the evaluation text set.

[0028] Step 102: Perform topic clustering on the processed text set to obtain subtopic text sets.

[0029] In some embodiments, the executing entity may perform topic clustering on the processed text set to obtain sub-topic text sets. In practice, the executing entity can perform topic clustering on the processed text set using the following formula to obtain sub-topic text sets: .

[0030] in, Indicates the number of categories. This represents a category index variable, with a range of [0, ...]. ]. Indicates the corresponding first Subtopic text for each category. This represents the clustering function for categories. This indicates that a text set is being processed. This represents the union operator.

[0031] In some optional implementations of certain embodiments, the aforementioned execution entity may perform topic clustering on the processed text set to obtain sub-topic text sets through the following steps: Step one: Determine the semantic vectors of each processed text in the above-mentioned text set to obtain a semantic feature vector set. In practice, the execution entity can use a Transformer encoder to convert each processed text in the above-mentioned text set into semantic vectors to obtain a semantic feature vector set.

[0032] Step two: Based on the silhouette coefficients, determine the optimal number of clusters for the aforementioned semantic feature vector set. In practice, the execution entity can first set the range of values ​​for the number of clusters (which can be [2, N]). Second, for each cluster number n within the range, use a clustering algorithm to cluster the semantic feature vector set and determine the sample silhouette coefficients of all semantic feature vectors. Then, sum all the sample silhouette coefficients and take the average to obtain the average silhouette coefficient for the current number of clusters. Third, determine the n value corresponding to the largest average silhouette coefficient as the optimal number of clusters.

[0033] The sample silhouette coefficients mentioned above can be obtained through the following steps: For each semantic feature vector (sample), its average distance (a) to other semantic feature vectors in the same category cluster and its average distance (b) to all semantic feature vectors in the nearest category cluster can be determined to obtain the sample silhouette coefficient. The sample silhouette coefficient is calculated as (ba) / max(a, b). The clustering algorithm mentioned above can be the K-Means clustering algorithm. N can be preset and can be 10.

[0034] Step 3: Based on the optimal number of clusters and the semantic feature vector set mentioned above, generate a topic relation set. In practice, the executing entity can cluster the semantic feature vectors in the semantic feature vector set according to the optimal number of clusters and the clustering algorithm mentioned above to generate a topic relation set. The topic relations in the topic relation set can be used to represent knowledge topics.

[0035] Step four: Based on the aforementioned topic relationship set and the aforementioned processed text set, generate a subtopic text set. In practice, the executing entity can merge the processed texts in the aforementioned processed text set with the topic relationships corresponding to the aforementioned topic relationship set to form a subtopic text set.

[0036] Step 103: Perform multi-topic fusion on the subtopic text sets to obtain the topic text sets.

[0037] In some embodiments, the aforementioned executing entity may perform multi-topic fusion on the aforementioned subtopic text sets to obtain a topic text set.

[0038] In addressing the aforementioned technical challenges by employing technical solutions, and considering the application scenario—the construction of an intelligent educational resource database—which requires building clearly labeled texts from massive amounts of multi-topic educational texts, often presents the following technical problems: Traditional clustering algorithms typically create independent clusters based on simple semantic similarity, failing to identify core and peripheral content in dialogues. This results in overly homogeneous or dispersed data, leading to poor model recognition capabilities and slow recognition speeds, consequently increasing the response time of smart educational cognitive terminals and hindering real-time responses. Given the specific requirements of this application scenario—deep semantic association and balanced data distribution—we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned execution entity may perform multi-topic fusion on the aforementioned subtopic text set through the following steps to obtain a topic text set: Step one: Determine the similarity matrix of each subtopic text in the aforementioned subtopic text set, thus obtaining a subtopic matrix set. In practice, the executing entity can determine the cosine similarity between every two semantic feature vectors in each subtopic text within the aforementioned subtopic text set based on the aforementioned semantic feature vector set, thereby generating a similarity matrix and obtaining a subtopic matrix set. The elements in the aforementioned similarity matrix can represent the cosine similarity values ​​between texts.

[0039] Step two involves performing a graph structure transformation on the aforementioned subtopic matrix set to generate a topic text relationship graph set. In practice, the executing entity can perform a graph structure transformation on each subtopic matrix in the aforementioned subtopic matrix set to generate a topic text relationship graph, thus obtaining the topic text relationship graph set. The aforementioned topic text relationship graph can be an undirected weighted graph. Nodes in the aforementioned topic text relationship graph represent subtopic texts. Edges in the aforementioned topic text relationship graph can be obtained by connecting two subtopic text sets whose cosine similarity exceeds a preset similarity threshold. The preset similarity threshold can be 0.75.

[0040] Step 3: Based on graph theory algorithms, perform node identification on the aforementioned topic text relationship graph to obtain a representative text set and a diversity text set. In practice, the executing entity can use graph theory algorithms to determine the degree centrality of each node in the aforementioned topic text relationship graph. Then, based on the degree centrality, sort the nodes in the aforementioned topic text relationship graph in descending order to obtain a node sequence. Next, determine the first x nodes in the node sequence as the representative text set, and the remaining nodes as the diversity text set. These nodes include central nodes and edge nodes. The representative text set can be a text set composed of central nodes. The diversity text set can be a text set composed of edge nodes. The degree centrality can be the number of edges directly connected to a node. x can be preset, such as 20.

[0041] Step four involves determining the cross-category semantic relevance of the aforementioned representative text set, resulting in a cross-category relevance matrix. In practice, the executing entity can determine the average semantic vector of each representative text in the aforementioned representative text set. Then, the cosine similarity between the average semantic vectors of any two different representative texts is determined, yielding the cross-category relevance matrix. The rows and columns in the aforementioned cross-category relevance matrix can represent representative texts. The elements in the aforementioned cross-category relevance matrix can represent cosine similarity.

[0042] Step 5: Based on the aforementioned cross-category correlation matrix, construct text sampling indicators. These indicators may include a diversity text reward item and a representative text coverage item. The representative text coverage item encourages the selected texts to maintain semantic consistency with the selected representative sample set. The diversity text reward item encourages the selection of texts from other sub-topics that have lower relevance to the sub-topics of the selected texts.

[0043] Step Six: Based on the aforementioned text sampling metrics and the greedy algorithm, sample the diverse text set to obtain a sampled text set. In practice, the executing entity can use a greedy algorithm to traverse the diverse text set and determine the sampling metric gain of each diverse text in the set based on the aforementioned text sampling metrics. Then, the top y diverse texts with the largest sampling metric gains are selected as the sampled texts, thus obtaining the sampled text set. Here, y can be a preset value, or it can be 100.

[0044] Step seven involves merging the labels of the sampled text sets to obtain an initial topic text set. In practice, the executing entity can determine the sub-topic categories of each sampled text in the sampled text set. Then, based on the cross-category correlation matrix, sub-topic categories with cosine similarity exceeding a preset merging value (which could be 0.6) are identified as new sub-topic categories, and the sampled text set is updated according to these new sub-topic categories to obtain the initial topic text set. As an example, the executing entity can merge "solving quadratic equations" and "solving functions" into "algebraic foundations".

[0045] Specifically, the aforementioned executing entity can perform tag merging on the sampled text set to obtain the initial topic text set through the following steps: .

[0046] .

[0047] .

[0048] .

[0049] .

[0050] in, Indicates the category from which the current text originates. Indicates the number of texts involved in the splicing. This represents the component index, used to identify the first component in a new sample. Each part has a value range from 0 to... . Indicates the corresponding first Subtopic text for each category. Indicates from the first The sample taken from the dataset of the nth category is used as the first sample to be spliced. Each part. This represents a function for randomly selecting samples. This represents the initial topic text after splicing. This indicates a concatenation function. This indicates the label corresponding to the extracted sample. This represents the newly constructed sample label. This represents the initial topic text set obtained by merging the tags of the text set processed above. This represents the union operator.

[0051] Step 8: Perform deduplication filtering on the initial topic text set to obtain the topic text set. In practice, the executing entity can determine the cosine similarity of every two initial topic texts in the initial topic text set, and remove the initial topic texts with a cosine similarity exceeding a preset threshold (which can be 0.75) to obtain the topic text set.

[0052] Steps one through eight and their related content, as an inventive point of this disclosure, combined with the steps below, solve the technical problem that "traditional clustering algorithms are usually based on simple semantic similarity to divide independent category clusters, which cannot identify the core and peripheral content in a dialogue, making the data too homogeneous or too divergent, resulting in poor model recognition ability and slow recognition speed, which in turn increases the response time of the smart education cognitive terminal and makes it impossible to achieve real-time response." The reason for the inability to achieve real-time response is that traditional clustering algorithms are usually based on simple semantic similarity to divide independent category clusters, which cannot identify the core and peripheral content in a dialogue, making the data too homogeneous or too divergent, resulting in poor model recognition ability and slow recognition speed, which in turn increases the response time of the smart education cognitive terminal and makes it impossible to achieve real-time response. If the above factors are solved, the problem of not being able to achieve real-time response can be solved. To achieve this effect, the first step is to determine the similarity matrix of each subtopic text in the above subtopic text set, obtaining a subtopic matrix set. Thus, the semantic relationship between texts can be determined through the similarity matrix. The second step is to perform graph structure transformation on the above subtopic matrix set to generate a topic text relationship graph set. Therefore, the position of text in the graph structure network can be quickly identified. The third step, based on graph theory algorithms, is to identify nodes in the aforementioned topic text relationship graph to obtain a representative text set and a diverse text set. This solves the problem of uneven data distribution in traditional clustering methods. The fourth step is to determine the cross-category semantic correlation of the aforementioned representative text set, obtaining a cross-category correlation matrix. This allows for the identification of potential connections between different knowledge topics. The fifth step is to construct text sampling indicators based on the aforementioned cross-category correlation matrix. This allows for the construction of quantitative demand indicators. The sixth step is to sample the aforementioned diverse text set based on the aforementioned text sampling indicators and a greedy algorithm, obtaining a sampled text set. This results in a text set with balanced data distribution. The seventh step is to merge the labels of the aforementioned sampled text set to obtain an initial topic text set. This solves the problems of label redundancy and semantic fragmentation caused by traditional algorithms. The eighth step is to perform deduplication filtering on the aforementioned initial topic text set to obtain a topic text set. This results in a non-redundant text set, improving the model's data processing efficiency and performance. Combined with step 107 below, in response to the topic indicator exceeding the preset indicator threshold, an educational control command is generated based on the real-time topic identifier set. Based on the educational control command, the smart education cognitive terminal adjusts its output content, and the teaching aid transmission device transmits the teaching aids. This ultimately solves the problem of not being able to achieve real-time response.

[0053] Step 104: Update the initial educational cognitive model based on the thematic text set to obtain the educational cognitive model.

[0054] In some embodiments, the aforementioned implementing entity may update the initial educational cognitive model based on the aforementioned thematic text set to obtain an educational cognitive model.

[0055] In some optional implementations of certain embodiments, the aforementioned executing entity may update the initial educational cognitive model based on the aforementioned topic text set through the following steps to obtain the educational cognitive model: Step one: Based on the preset instruction template, the above-mentioned topic text set is formatted to obtain an instruction dataset. In practice, the executing entity can select the corresponding preset instruction template based on the educational scenario (topic tag) of each topic text in the above-mentioned topic text set, and fill the blank positions of the preset instruction template with the topic text to obtain the instruction dataset. The preset instruction template can be a template used to convert the above-mentioned topic text set into a unified input format, used to convert the topic text into a standard input form that the model can recognize.

[0056] Step two involves inputting the aforementioned instruction dataset into the initial educational cognitive model to obtain a text feature vector set. In practice, the executing entity can use masking techniques to cover words in the instruction dataset that are irrelevant to the knowledge point recognition task. Then, a Transformer encoder is used to convert the masked instruction dataset into semantic vectors, resulting in a text vector set. These irrelevant words can be greetings such as "Please answer," "Okay," or "Thank you." The initial educational model can be a Large Language Model (LLM).

[0057] Specifically, the aforementioned executing entity can convert the masked instruction dataset into a text feature vector set using the following formula: .

[0058] in, This represents the text feature vector. This indicates that semantic features can be extracted using the base model through the mapping function of the text mapping part. This represents the instruction dataset.

[0059] Step three involves adding a weight matrix to the initial educational cognitive model to obtain the first educational cognitive model. In practice, the implementing entity can first be constructed based on the longest topic text in the aforementioned topic text set. For other topic texts, a two-dimensional matrix can be constructed using the interpolation completion algorithm. A two-dimensional matrix. Next, the matrices are merged to obtain... A three-dimensional matrix. It can be achieved using the following formula: .

[0060] in, Represents a three-dimensional matrix. This represents a function that converts text into a matrix. This indicates a set of texts related to a specific topic. The dimension of the constructed two-dimensional matrix is ​​determined by the longest topic text in the topic text set. Indicates the total number of topic texts.

[0061] Furthermore, the aforementioned execution entity can extract the relational features of the aforementioned three-dimensional matrix using a multimodal large model to obtain a relational feature set. This relational feature set is then merged with the aforementioned text feature vector set to obtain a merged vector set. This can be achieved using the following formula: .

[0062] .

[0063] in, Represents a set of relational features. This represents a multimodal large model. This represents the set of text feature vectors. This indicates addition by place value. This indicates a merged vector set. Represents a three-dimensional matrix.

[0064] Next, the LoRA fine-tuning method is used to fine-tune the pre-trained weight matrix. The incremental parameter matrix of full parameter fine-tuning Represented as a matrix with two smaller parameters and The low-rank similarity matrix can be obtained using the following formula: .

[0065] in, This represents the weight matrix. This represents the incremental parameter matrix. Let represent a dimension-reduced matrix with dimensions d×r and rank d greater than r. This represents an upgraded matrix with dimensions r×d.

[0066] Step four: Input the above text feature vector set into the first educational cognitive model to obtain the output vector set. This can be achieved using the following formula: .

[0067] in, This represents the output vector. This represents the bias term, with a dimension of llm. This represents a dimension reduction matrix. This represents an upgraded matrix. This represents the weight matrix. This indicates a merged vector set. This represents the learnable weight matrix, with dimensions llm×d.

[0068] Step 5: Determine the similarity loss value based on the output vector set and the instruction dataset described above. In practice, the executing entity can determine the similarity loss value using the following formula: .

[0069] in, This indicates a topic index. This indicates a text index. This represents the index variable, used to iterate through all samples. This indicates the total number of text samples. Indicates the first Output vectors. Indicates the first Instruction data. This represents the similarity loss value. Represents the similarity function (which can be a cosine similarity function). Represents the natural constant.

[0070] Step six: Based on the aforementioned similarity loss value, the first educational cognitive model is updated in reverse to obtain the educational cognitive model. In practice, the aforementioned executing entity can use the AdamW optimizer to update the low-rank similarity matrix of the first educational cognitive model according to the aforementioned similarity loss value to obtain the educational cognitive model.

[0071] The aforementioned educational cognitive model can be a deep learning model trained on a Large Language Model (LLM) capable of identifying dialogue topics based on the dialogue text. This model may include a semantic encoding and feature extraction layer, a multimodal feature fusion layer, and an output layer. The semantic encoding and feature extraction layer can be a Transformer-based neural network that takes the educational dialogue text as input and outputs semantic vectors. The multimodal feature fusion layer can be a neural network that takes semantic feature vectors and topological relationship feature vectors as input and outputs a comprehensive feature representation. This multimodal feature fusion layer may include Graph Convolutional Networks (GCNs) to convert the keyword association matrix into feature vectors. The output layer can be a neural network that takes the comprehensive feature representation as input and outputs multiple topic identifiers. This output layer may consist of multiple fully connected layers and a sigmoid activation function to determine the probability values ​​of the comprehensive feature representation and each topic.

[0072] Step 105: In response to the smart education cognitive terminal receiving the target user's real-time educational dialogue, a real-time topic identifier set is generated based on the educational cognitive model and the real-time educational dialogue.

[0073] In some embodiments, the aforementioned executing entity may, in response to the smart education cognitive terminal receiving a real-time educational dialogue from a target user, generate a real-time topic identifier set based on the aforementioned educational cognitive model and the real-time educational dialogue. The aforementioned real-time educational dialogue may be a raw dialogue data stream generated and acquired in real-time within an educational interaction scenario. The aforementioned educational interaction scenario may be an online classroom or an intelligent tutoring system.

[0074] In some optional implementations of certain embodiments, the aforementioned executing entity can generate a real-time topic identifier set based on the aforementioned educational cognitive model and the aforementioned real-time educational dialogue through the following steps: Step one involves cleaning the real-time educational dialogue to obtain a standard educational dialogue. In practice, the executing entity can use a spectral subtraction algorithm (noisereduce) to denoise the real-time educational dialogue and obtain the standard educational dialogue.

[0075] Step two involves converting the aforementioned standard educational dialogue into text, resulting in the standard educational dialogue text. In practice, the executing entity can use a word segmentation algorithm to convert the standard educational dialogue into a text sequence. Then, based on a preset text length, the word sequence is aligned to obtain the standard educational dialogue text. Specifically, the executing entity can dynamically pad or truncate the word sequence to reach the preset text length, and add marker characters (such as "CLS" and "SEP") to the word sequence to indicate the start and end of the dialogue. The word segmentation algorithm can be the BPE (Byte Pair Encoding) algorithm. The preset text length can be a pre-set fixed upper limit for the number of words (which can be no more than 8192 words).

[0076] Step three involves converting the aforementioned standard educational dialogue text into a keyword association matrix. In practice, firstly, the executing entity can use the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm to identify keywords in the standard educational dialogue text, obtaining a keyword set. Then, Word2Vec can be used to convert the keyword set into a vector representation, obtaining a semantic vector set. Finally, the keyword association matrix can be obtained by determining the cosine similarity between every two semantic vectors in the semantic vector set. This keyword association matrix can be a two-dimensional matrix, where rows and columns represent keywords, and elements represent cosine similarity.

[0077] Step four: Input the aforementioned standard educational dialogue text into the aforementioned educational cognitive model to obtain a semantic feature vector. In practice, the executing entity can input the aforementioned standard educational dialogue text into the Transformer encoder part of the aforementioned educational cognitive model to obtain a semantic feature vector.

[0078] Step five: Input the keyword association matrix into the educational cognitive model to obtain the topological relationship feature vector. In practice, the executing entity can convert the keyword association matrix into a graph data structure. Then, through the graph neural network of the educational cognitive model, perform a global pooling operation on the graph data structure to obtain the topological relationship feature vector.

[0079] Step six involves concatenating the semantic feature vectors and topological relationship feature vectors to generate a comprehensive feature representation. In practice, the executing entity can use the NumPy library to concatenate the semantic feature vectors and topological relationship feature vectors to generate the comprehensive feature representation.

[0080] Step 7: Generate an initial topic identifier set based on the above comprehensive feature representation. In practice, the executing entity can input the above comprehensive feature representation into the fully connected layer of the above educational cognitive model to determine the probability values ​​of the above comprehensive feature representation and each topic, and identify topics with probability values ​​exceeding a preset probability threshold (which can be 0.5) as the initial topic identifier set.

[0081] Step 8: Based on the preset educational knowledge base, verify the initial topic identifier set to obtain the real-time topic identifier set. In practice, the executing entity can query each initial topic identifier in the preset educational knowledge base. Then, determine the cosine similarity between the initial topic identifier and the educational knowledge entities in the preset educational knowledge base, and identify the initial topic identifiers with a cosine similarity exceeding a preset value (0.75) as real-time topic identifiers, thus obtaining the real-time topic identifier set.

[0082] Step 106: Based on the evaluation text set, determine the topic metrics of the real-time topic identifier set.

[0083] In some embodiments, the executing entity may determine the topic metrics of the real-time topic identifier set based on the evaluation text set.

[0084] In practice, the aforementioned implementing entities can determine the topic metrics of the aforementioned real-time topic identifier set based on the aforementioned evaluation text set through the following steps: The first step is to generate a predicted label set based on the aforementioned assessment text set and the aforementioned educational cognitive model. In practice, the implementing entity can input the aforementioned assessment text set into the aforementioned educational cognitive model to obtain the predicted label set.

[0085] The second step involves generating a label comparison result based on the predicted label set and the real-time topic identifier set. In practice, the executing entity can compare the predicted label set (set P) and the real-time topic identifier set (set T) to obtain the label comparison result. This comparison result can include correctly predicted labels (True Positive, TP), incorrectly predicted labels (False Positive, FP), unpredicted labels (False Negative, FN), and their count. The correctly predicted labels can be the set of labels that are both predicted and actually exist, i.e., TP = P. T. The above incorrectly predicted labels can be the set of labels predicted to exist but actually not existing, i.e., FP=PT. The above unpredicted labels can be the set of labels predicted not to exist but actually existing, i.e., FN=TP.

[0086] The third step is to generate topic metrics for the aforementioned real-time topic identifier set based on the above tag comparison set. These topic metrics include precision and recall. The topic metrics can be calculated using the following formula: .

[0087] .

[0088] .

[0089] in, Indicates accuracy. This indicates the number of correctly predicted labels. This indicates the number of incorrectly predicted labels. This indicates the recall rate. This indicates the number of unpredicted labels. This indicates a summation operation. Indicates thematic indicators.

[0090] Step 107: In response to the topic indicator exceeding the preset indicator threshold, an educational control instruction is generated based on the real-time topic identifier set, and the smart education cognitive terminal is controlled to adjust the output content and the teaching aid transmission device is controlled to transmit teaching aids according to the educational control instruction.

[0091] In some embodiments, the execution entity may, in response to the topic indicator exceeding a preset indicator threshold, generate an educational control instruction based on the real-time topic identifier set, and control the smart education cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids based on the educational control instruction. The preset indicator threshold may be 0.85.

[0092] In addressing the technical problems mentioned above by adopting technical solutions, and considering the application scenario—high-precision educational practice scenarios such as remote medical surgery teaching—the following technical issues often arise: traditional rule-based or simple retrieval-based recommendation methods cannot deeply analyze knowledge points, leading to overly generalized and unspecific results. This, in turn, causes educational auxiliary terminals to output incorrectly and call up teaching aids, resulting in wasted teaching aids. Considering the following requirements for this application scenario: high timeliness, execution accuracy, and decision reliability, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned execution entity may generate educational control instructions based on the aforementioned real-time topic identifier set, and control the aforementioned smart education cognitive terminal to adjust its output content and the aforementioned teaching aid transmission device to transmit teaching aids based on the aforementioned educational control instructions: Step one: In response to the aforementioned topic indicators exceeding a preset threshold, determine the semantic relevance of each real-time topic identifier in the aforementioned real-time topic identifier set to obtain a topic weight vector set. In practice, the aforementioned executing entity can obtain the topic weight vector set by determining the cosine similarity of each real-time topic identifier in the aforementioned real-time topic identifier set.

[0093] Step two: Based on the aforementioned topic weight vector set and teaching strategy knowledge base, a candidate strategy set is generated. In practice, the executing entity can convert each teaching strategy knowledge in the teaching strategy knowledge base into a vector representation, obtaining a teaching strategy vector set. Then, by determining the cosine similarity between the aforementioned topic weight vector set and each teaching strategy vector in the aforementioned teaching strategy vector set, the teaching strategy knowledge corresponding to the teaching strategy vectors with a cosine similarity exceeding a preset similarity threshold is identified as candidate strategies, thus obtaining the candidate strategy set. The aforementioned teaching strategy knowledge base can be an external database storing best practice strategies in the education field. The preset similarity threshold can be 0.75.

[0094] Step 3: Based on the historical dialogue memory, extract historical dialogue information and construct a context feature vector set based on this information. In practice, the executing entity can extract historical dialogues within a preset time window from the historical dialogue memory to obtain historical dialogue information. Then, convert the historical dialogue information into a vector representation to obtain the context feature vector set. The historical dialogue memory can be a time-series database storing complete records of historical educational dialogues. These complete records can include text and timestamps. The preset time window can be 10 days.

[0095] Step four involves inputting the aforementioned candidate strategy set, context feature vector set, and topic weight vector set into the causal inference model to obtain the inference report set. In practice, the implementing entity can determine the topic weight vector set as a confounding variable, the context feature vector set as a state variable, and the candidate strategy set as an intervention variable, and obtain the inference report set through a counterfactual inference algorithm. The causal inference model can be a deep learning model based on a structural causal graph (SCM) to simulate the potential causal effects under the intervention of teaching strategies. For example, the causal inference model can be a CausalVAE model. The inference report set can include the teaching prediction results generated after the candidate strategies are implemented. These teaching prediction results can be used to characterize the degree of knowledge mastery.

[0096] Step 5: Based on the aforementioned context feature vector set, construct a teaching simulation environment and input the aforementioned candidate strategy set into the teaching simulation environment to obtain a behavior trajectory set. In practice, the aforementioned executing entity can simulate the student's knowledge mastery state and emotional state using the aforementioned context feature vectors to construct a teaching simulation environment. Then, using the Monte Carlo Simulation algorithm, simulate each candidate strategy in the aforementioned candidate strategy set within the aforementioned teaching simulation environment to obtain a behavior trajectory set. The aforementioned teaching simulation environment can be a virtual student agent interaction environment simulated by the Proximal Policy Optimization (PPO) algorithm. The aforementioned behavior trajectory set can be a set of records generated by executing candidate strategies within the aforementioned teaching simulation environment. The aforementioned behavior trajectory set can include a state sequence, an action sequence, a reward signal sequence, and a final state. The aforementioned state sequence can represent the student's knowledge mastery state and emotional state. The aforementioned action sequence can represent the executed teaching actions, such as "explaining new knowledge points" or "pushing practice questions." The aforementioned reward signal sequence can represent the reward value of student state changes, such as giving a positive reward for improved knowledge mastery. The aforementioned final state is used to represent the student's learning state after the dialogue ends.

[0097] Step Six: Perform a fusion analysis on the aforementioned behavioral trajectory set and the aforementioned deduction report set to obtain the optimal teaching strategy. In practice, the implementing entity can perform a weighted summation of the aforementioned behavioral trajectory set and the aforementioned deduction report set to obtain the optimal teaching strategy. Specifically, weights can be assigned to the key indicators in the aforementioned behavioral trajectory set, and a weighted score can be obtained. The candidate strategy corresponding to the behavioral trajectory with the highest weight score is determined as the optimal teaching strategy. The aforementioned key indicators may include teaching efficiency, robustness, and causal effect. The weight of the aforementioned teaching efficiency may be 0.5. The weight of the aforementioned robustness may be 0.3. The weight of the aforementioned causal effect may be 0.2.

[0098] Step 7: Generate educational control instructions based on the aforementioned preferred teaching strategy. In practice, the executing entity can map the preferred teaching strategy to a preset instruction template to obtain educational control instructions. The preset instruction template can be a structured template used to convert abstract teaching strategies into machine instructions that can be parsed and executed by the machine. The educational control instructions can be machine-readable instructions used to control the specific behavior of the smart educational cognitive terminal, and can be in JSON format. The educational control instructions also include specific answers to user questions and the parameters required to answer them. As an example, the educational control instructions may include, but are not limited to, "slow down the explanation speed," "highlight knowledge points," and "push teaching aids."

[0099] Step eight: Send the aforementioned educational control commands to the aforementioned smart education cognitive terminal to adjust the teaching output content and the aforementioned teaching aid transmission device for teaching aid transmission. In practice, the aforementioned executing entity can send the aforementioned educational control commands to the aforementioned smart education cognitive terminal through a preset communication protocol. The preset communication protocol can be the WebSocket protocol.

[0100] Steps one through eight and their related content, as an inventive point of this disclosure, solve the technical problem that "traditional rule-based or simple retrieval-based recommendation methods cannot perform in-depth analysis of knowledge points, resulting in overly generalized and unspecific results, leading to incorrect output and use of teaching aids by educational auxiliary terminals, resulting in waste of teaching aids." The reason for this waste is that traditional rule-based or simple retrieval-based recommendation methods cannot perform in-depth analysis of knowledge points, resulting in overly generalized and unspecific results, leading to incorrect output and use of teaching aids by educational auxiliary terminals, resulting in waste of teaching aids. Solving these factors can resolve the problem of wasted teaching aids. To achieve this, the first step involves determining the semantic relevance of each real-time topic identifier in the real-time topic identifier set in response to the aforementioned topic indicators exceeding a preset threshold, thus obtaining a topic weight vector set. This allows for the identification of key knowledge points or operational problems. The second step involves generating a candidate strategy set based on the aforementioned topic weight vector set and the teaching strategy knowledge base. This avoids blind screening and reduces response time. The third step involves extracting historical dialogue information from the historical dialogue memory and constructing a contextual feature vector set based on this information. This enhances the relevance of the output content. The fourth step involves inputting the candidate strategy set, the contextual feature vector set, and the topic weight vector set into the causal reasoning model to obtain a deduction report set. This allows for simulating the impact of the current strategy, improving the feasibility and security of decision-making. The fifth step involves constructing a teaching simulation environment based on the contextual feature vector set and inputting the candidate strategy set into this environment to obtain a behavior trajectory set. This quantifies the impact of the strategy, making the output content more objective. The sixth step involves fusing and analyzing the behavior trajectory set and the deduction report set to obtain the optimal teaching strategy. This allows for the selection of the best output strategy. The seventh step involves generating educational control instructions based on the optimal teaching strategy. This transforms the strategy into executable instructions. The eighth step involves sending the educational control instructions to the smart education cognitive terminal to adjust the teaching output content and the teaching aid transmission device for teaching aid transmission. This ensures the accuracy of output content and teaching aid recall, ultimately solving the problem of teaching aid waste.

[0101] Optionally, after step 107, the above method further includes: In response to the completion of the output of the above-mentioned smart education cognitive terminal content, the feedback data of the above-mentioned target users is collected, and the above-mentioned teaching strategy knowledge base is updated based on federated learning and the above-mentioned feedback data.

[0102] In some embodiments, the executing entity may, in response to the completion of content output by the smart education cognitive terminal, collect feedback data from the target user and update the teaching strategy knowledge base based on federated learning and the feedback data. In practice, the executing entity may use the feedback data as a supervisory signal to update the teaching strategy knowledge base through a federated learning algorithm. The feedback data may include user ratings of the teaching output content and operation results (such as repeatedly viewing the output content).

[0103] The various embodiments of this disclosure have the following beneficial effects: the multi-semantic topic fusion smart education cognitive terminal control method of some embodiments of this disclosure can shorten the response time and avoid the erroneous transmission of teaching aids. Specifically, the reasons for long response time and erroneous transmission of teaching aids are: single-label classification makes it difficult to handle scenarios with complex semantic levels in educational dialogues, resulting in incomplete knowledge point recognition and thus erroneous transmission of teaching aids; and model training requires a large amount of high-quality labeled data and feature extraction consumes a large amount of computing resources, resulting in slow terminal response speed and inability to meet the low latency requirements of real-time interaction. Based on this, the multi-semantic topic fusion smart education cognitive terminal control method of some embodiments of this disclosure firstly, in response to the smart education cognitive terminal receiving the original educational dialogue text set, preprocesses the original educational dialogue text set to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device. This can reduce the problem of wasted computing resources caused by data redundancy. Secondly, topic clustering is performed on the processed text set to obtain a sub-topic text set. This can identify potential knowledge points in the educational dialogue and solve the problem of incomplete identification. Next, the aforementioned sub-topic text sets are fused into a multi-topic text set to obtain a topic text set. This enhances the semantic relevance of the dialogue. Then, based on the topic text set, the initial educational cognitive model is updated to obtain an educational cognitive model. This allows for model updates with only a few parameters, improving training efficiency and addressing the slow response time issue. Subsequently, in response to the smart educational cognitive terminal receiving a real-time educational dialogue from the target user, a real-time topic identifier set is generated based on the educational cognitive model and the real-time educational dialogue. This allows for topic identification of the real-time dialogue. Then, based on the aforementioned evaluation text set, the topic indicators of the real-time topic identifier set are determined. This reduces the risk of subsequent misoperations through accurate identification results. Finally, in response to the topic indicators exceeding a preset threshold, educational control instructions are generated based on the real-time topic identifier set. These instructions then control the smart educational cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids. This converts the identification results into controllable instructions, ultimately reducing response time and preventing erroneous transmission of teaching aids.

[0104] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a smart education cognitive terminal control device that integrates multiple semantic topics. These device embodiments are similar to... Figure 2 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0105] like Figure 2As shown, a multi-semantic topic fusion smart education cognitive terminal control device 200 in some embodiments includes: a preprocessing unit 201, a topic clustering unit 202, a multi-topic fusion unit 203, an updating unit 204, a generation unit 205, a determination unit 206, and a second generation unit 207. The preprocessing unit 201 is configured to preprocess the original educational dialogue text set in response to the smart education cognitive terminal receiving the original educational dialogue text set, obtaining a processed text set and an evaluation text set. The smart education cognitive terminal includes a display and an associated teaching aid transmission device. The topic clustering unit 202 is configured to perform topic clustering on the processed text set, obtaining a sub-topic text set. The multi-topic fusion unit 203 is configured to perform multi-topic fusion on the sub-topic text sets, obtaining a topic text set. The updating unit 204 is configured to update the initial educational cognitive model based on the topic text set, obtaining... The system includes an educational cognitive model; a generation unit 205, configured to generate a real-time topic identifier set based on the educational cognitive model and the real-time educational dialogue received by the smart educational cognitive terminal; a determination unit 206, configured to determine the topic indicators of the real-time topic identifier set based on the evaluation text set; and a second generation unit 207, configured to generate an educational control command based on the real-time topic identifier set in response to the topic indicators exceeding a preset threshold, and to control the smart educational cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids based on the educational control command.

[0106] It is understandable that the units described in the device 200 are related to the reference. Figure 2 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the device 200 and the units contained therein, and will not be repeated here.

[0107] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0108] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0109] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0110] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0111] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0112] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0113] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs. When the electronic device executes the aforementioned one or more programs, the electronic device causes the following actions: In response to the smart education cognitive terminal receiving an original educational dialogue text set, the electronic device preprocesses the original educational dialogue text set to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device; performs topic clustering on the processed text set to obtain a sub-topic text set; performs multi-topic fusion on the sub-topic text set to obtain a topic text set; updates the initial educational cognitive model based on the topic text set to obtain an educational cognitive model; In response to the smart education cognitive terminal receiving a real-time educational dialogue from a target user, the electronic device generates a real-time topic identifier set based on the educational cognitive model and the real-time educational dialogue; determines topic indicators of the real-time topic identifier set based on the evaluation text set; In response to the topic indicators exceeding a preset indicator threshold, the electronic device generates an educational control command based on the real-time topic identifier set, and controls the smart education cognitive terminal to adjust its output content and the teaching aid transmission device to transmit teaching aids based on the educational control command.

[0114] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0116] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including a preprocessing unit, a topic clustering unit, a multi-topic fusion unit, an update unit, a generation unit, a determination unit, and a second generation unit. The names of these units do not necessarily limit the specific unit itself; for example, the preprocessing unit may also be described as "a unit that, in response to a smart education cognitive terminal receiving an original educational dialogue text set, preprocesses the original educational dialogue text set to obtain a processed text set and an evaluation text set."

[0117] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0118] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A control method for a smart education cognitive terminal that integrates multiple semantic themes, comprising: In response to the smart education cognitive terminal receiving the original educational dialogue text set, the original educational dialogue text set is preprocessed to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device. The processed text set is subjected to topic clustering to obtain subtopic text sets; The sub-topic text sets are fused into multiple themes to obtain thematic text sets; Based on the aforementioned thematic text set, the initial educational cognitive model is updated to obtain the new educational cognitive model. In response to the smart education cognitive terminal receiving a real-time educational dialogue from a target user, a real-time topic identifier set is generated based on the educational cognitive model and the real-time educational dialogue; Based on the evaluation text set, determine the topic metrics of the real-time topic identifier set; In response to the topic indicator exceeding a preset indicator threshold, an educational control instruction is generated based on the real-time topic identifier set, and the intelligent educational cognitive terminal is controlled to adjust its output content and the teaching aid transmission device is controlled to transmit teaching aids based on the educational control instruction.

2. The method according to claim 1, wherein, The response to the smart education cognitive terminal receiving the original educational dialogue text set involves preprocessing the original educational dialogue text set to obtain a processed text set and an evaluation text set, including: The original educational dialogue text set was standardized to obtain a standard text set. The standard text set is semantically segmented to obtain a sequence of semantic units; Based on the semantic unit sequence, a scene feature set is generated, wherein the scene feature set includes an educational terminology set and a corresponding dialogue behavior tag set; The scene feature set is subjected to stratified sampling to obtain the processing text set and the evaluation text set.

3. The method according to claim 1, wherein, The step of performing topic clustering on the processed text set to obtain sub-topic text sets includes: The semantic vectors of each processed text in the processed text set are determined to obtain a set of semantic feature vectors; Based on the contour coefficient, the optimal number of clusters for the semantic feature vector set is determined; Based on the optimal number of clusters and the semantic feature vector set, a topic relation set is generated; Based on the topic relationship set and the processed text set, a subtopic text set is generated.

4. The method according to claim 1, wherein, The step of updating the initial educational cognitive model based on the topic text set to obtain the educational cognitive model includes: The topic text set is formatted according to a preset instruction template to obtain an instruction dataset; The instruction dataset is input into the initial educational cognitive model to obtain a set of text feature vectors; Add a weight matrix to the initial educational cognitive model to obtain the first educational cognitive model; The text feature vector set is input into the first educational cognitive model to obtain the output vector set; The similarity loss value is determined based on the output vector set and the instruction dataset; Based on the similarity loss value, the first educational cognitive model is updated in reverse to obtain the educational cognitive model.

5. The method according to claim 1, wherein, In response to the smart education cognitive terminal receiving a real-time educational dialogue from the target user, a real-time topic identifier set is generated based on the educational cognitive model and the real-time educational dialogue, including: The real-time educational dialogue is cleaned to obtain a standard educational dialogue; The standard educational dialogue is converted into text to obtain the standard educational dialogue text. Convert the standard educational dialogue text into a keyword association matrix; The standard educational dialogue text is input into the educational cognitive model to obtain a semantic feature vector; The keyword association matrix is ​​input into the educational cognitive model to obtain the topological relationship feature vector; The semantic feature vector and the topological relationship feature vector are concatenated to generate a comprehensive feature representation; Based on the comprehensive feature representation, an initial topic identifier set is generated; Based on a pre-set educational knowledge base, the initial topic identifier set is verified to obtain a real-time topic identifier set.

6. The method according to claim 1, wherein, The method further includes: In response to the completion of content output by the smart education cognitive terminal, feedback data from the target user is collected, and the teaching strategy knowledge base is updated based on federated learning and the feedback data.

7. A smart educational cognitive terminal control device that integrates multiple semantic themes, comprising: A preprocessing unit is configured to preprocess the original educational dialogue text set in response to the smart education cognitive terminal receiving the original educational dialogue text set, to obtain a processed text set and an evaluation text set, wherein the smart education cognitive terminal includes a display and an associated teaching aid transmission device. A topic clustering unit is configured to perform topic clustering on the processed text set to obtain a subtopic text set; The multi-topic fusion unit is configured to perform multi-topic fusion on the sub-topic text set to obtain a topic text set; The update unit is configured to update the initial educational cognitive model based on the topic text set to obtain the educational cognitive model. The generation unit is configured to generate a real-time topic identifier set based on the educational cognitive model and the real-time educational dialogue in response to the smart education cognitive terminal receiving a real-time educational dialogue from a target user. The determining unit is configured to determine the topic metrics of the real-time topic identifier set based on the evaluation text set; The second generation unit is configured to, in response to the topic indicator exceeding a preset indicator threshold, generate an educational control instruction based on the real-time topic identifier set, and control the smart education cognitive terminal to adjust the output content and the teaching aid transmission device to transmit teaching aids based on the educational control instruction.

8. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images