Intelligent AI interactive question and answer method and system applied to science popularization and storage medium

Abstract

The invention discloses an intelligent AI interactive question-answering method and system applied to science popularization and a storage medium, and the method comprises the steps: constructing a science popularization knowledge spectrogram based on multi-source science popularization knowledge; extracting key feature data based on a voice or text question input by a user, and generating a question feature vector; similarity matching is carried out on the basis of the feature vectors and a node embedding space of the science popularization knowledge spectrogram, and accurate knowledge sub-graphs related to questions are extracted; after completing the response of the current question, recording a storage positioning index of the knowledge sub-graph corresponding to the question in the popular science knowledge spectrogram, and generating knowledge sub-graph index sequence data; and based on the next question of the user, carrying out correlation analysis, including questions of the same and different knowledge types, and carrying out sub-graph matching or full-graph matching, and if a plurality of knowledge fields are involved, carrying out sub-graph matching and full-graph matching. The method has the advantages that efficient knowledge matching is realized by fusing multi-source popular science knowledge and feature extraction, user questions are quickly responded, and the matching precision is continuously optimized.

Description

Technical Field

[0001] The present invention relates to intelligent AI interaction, and in particular to an intelligent AI interactive question-and-answer method, system, and storage medium for popular science. Background Art

[0002] Intelligent AI interactive question-and-answer methods are innovative applications of natural language processing, machine learning, and deep learning technologies. These technologies primarily integrate speech recognition, text analysis, sentiment understanding, and semantic matching. By training models based on big data, AI can understand and generate natural language, enabling efficient interaction with human users. In recent years, deep neural networks have significantly enhanced AI's ability to understand context, reason, and generate reasonable responses, enabling widespread adoption of intelligent question-and-answer systems across various industries.

[0003] The intelligent AI interactive question-and-answer methods, systems, and storage media currently available on the market for popular science rely solely on preset rules or simple keyword matching, and are unable to flexibly handle complex, context-dependent user questions, resulting in low accuracy and relevance of answers. Secondly, these methods often lack effective knowledge graph support and in-depth knowledge integration capabilities, making them prone to incomplete information or inaccurate answers. Moreover, many existing systems, when dealing with multi-domain problems, are often unable to cover knowledge from different fields at the same time, resulting in answers to interdisciplinary questions that are not comprehensive or in-depth enough. In addition, AI question-and-answer systems on the market often lack continuous learning and optimization mechanisms, and are unable to continuously improve the accuracy of answers based on user feedback, resulting in slow performance improvements after long-term use. Summary of the Invention

[0004] In order to improve existing methods and systems, an intelligent AI interactive question-and-answer method, system, and storage medium for popular science are provided. This method achieves accurate question-and-answer and efficient knowledge matching by integrating multi-source popular science knowledge and deep learning feature extraction. It can quickly respond to user questions and continuously optimize matching accuracy.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] Intelligent AI interactive question-answering methods applied to science popularization include:

[0007] Based on multi-source popular science knowledge, a popular science knowledge spectrum is constructed through heterogeneous data processing of structured scientific research databases and unstructured popular science literature;

[0008] Based on the user's voice or text questions, the pre-trained deep learning feature extraction model extracts key feature data and generates a question feature vector;

[0009] Based on the similarity matching between the feature vector and the node embedding space of the popular science knowledge graph, the precise knowledge subgraph related to the question is extracted;

[0010] After completing the answer to the current question, record the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum, and generate knowledge subgraph index sequence data;

[0011] Based on the user's next question, a correlation analysis is performed. If the knowledge type of consecutive questions is detected, sub-graph matching is performed within the same knowledge category based on the positioning index. If the knowledge type similarity is less than the preset threshold, the full-graph matching process is re-executed. If multiple knowledge fields are involved, sub-graph matching is performed within the same knowledge category while executing the full-graph matching process.

[0012] Preferably, the construction of a popular science knowledge graph based on multi-source popular science knowledge through heterogeneous data processing of structured scientific research databases and unstructured popular science literature specifically includes:

[0013] Based on the acquired structured scientific research database and unstructured popular science literature, knowledge is integrated and matched through keywords, themes, and concepts to effectively connect academic information with popular information.

[0014] Obtain the entities and relationships between academic and popular information, match different expressions across data sources through word alignment and semantic similarity calculation, and integrate the data;

[0015] Based on the results of data integration, a popular science knowledge graph is constructed, where each node represents an entity and each edge represents the relationship between entities.

[0016] Preferably, the step of extracting key feature data based on the voice or text question input by the user through a pre-trained deep learning feature extraction model to generate a question feature vector specifically includes:

[0017] Converting voice signals into text data based on questions input by the user via voice;

[0018] All pre-processed text data is segmented, and the words are mapped into low-dimensional continuous vectors through word embedding methods, and semantically similar words are represented by similar vectors;

[0019] Through the context embedding method, based on the pre-trained feature extraction model, the text is input into the model as a whole to obtain the feature data of the text data and generate a more fine-grained and context-related feature vector.

[0020] Preferably, the similarity matching between the feature vector and the node embedding space of the popular science knowledge graph to extract the precise knowledge subgraph related to the question specifically includes:

[0021] When constructing a popular science knowledge graph, the nodes in the graph are mapped into a continuous low-dimensional vector space to obtain the structural and semantic information between the nodes;

[0022] Based on the acquired feature vector of the user's question and the embedding vector of each node in the knowledge graph, similarity is calculated using cosine similarity;

[0023] By calculating the similarity of all nodes, we select the node that is most similar to the user's question, and extract the nodes and their connected edges to form a subgraph;

[0024] Optimize the extracted subgraph and remove redundant and irrelevant nodes.

[0025] Preferably, after completing the answer to the current question, recording the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum, and generating the knowledge subgraph index sequence data specifically includes:

[0026] Identify each node and its corresponding relationship in the extracted knowledge subgraph, and set a unique identifier for each node and relationship in the knowledge graph;

[0027] Based on the unique identifier, each node and edge is assigned a unique location index in the graph.

[0028] Record the storage location of the knowledge subgraph corresponding to the question, and arrange the storage locations of all related nodes and edges in order to generate an ordered index sequence;

[0029] The index sequence of the knowledge subgraph is checked and updated regularly, and the index sequence of the knowledge subgraph is kept up to date and valid through the incremental update mechanism.

[0030] Preferably, the method performs correlation analysis based on the next question asked by the user. If the knowledge type of consecutive questions is detected, subgraph matching is performed within the same knowledge category based on the positioning index. If the knowledge type similarity is less than a preset threshold, the full-graph matching process is re-executed. If multiple knowledge fields are involved, the full-graph matching process is performed while subgraph matching is performed within the same knowledge category. Specifically, the method includes:

[0031] After completing the answer to the current question, receive the user's next question, perform feature extraction and correlation analysis;

[0032] If the knowledge type of the current question belongs to the same knowledge category as the previous question, the corresponding subgraph is quickly found and loaded based on the knowledge subgraph location index of the previous question;

[0033] Calculate the similarity between the feature vector of the current question and the node embedding vector in the loaded subgraph to determine whether an accurate answer can be given by the loaded subgraph. If the similarity is lower than the preset threshold, execute the full graph matching process;

[0034] If the current problem involves multiple knowledge domains, multiple subgraph matching is performed, with preliminary subgraph matching performed within the same knowledge category, and then matching is performed within the entire graph;

[0035] Based on user feedback, the subgraph matching algorithm is optimized and the threshold for similarity calculation is adjusted.

[0036] Furthermore, an intelligent AI interactive question-answering system for popular science is proposed, including:

[0037] Graph construction module: The graph construction module constructs a multi-source fusion science knowledge graph based on heterogeneous data processing of structured scientific research databases and unstructured science popularization literature;

[0038] Feature extraction module: The feature extraction module is used to pre-process the voice or text questions input by the user, convert them into feature vectors, and extract key question features;

[0039] Subgraph extraction module: The subgraph extraction module performs similarity calculation based on the feature vector of the user question and the node embedding space of the knowledge spectrum graph to extract relevant knowledge subgraphs;

[0040] Subgraph index module: The subgraph index module is used to store the knowledge subgraph corresponding to the question in the knowledge spectrum, and generate a positioning index and an ordered index sequence to ensure efficient access;

[0041] Subgraph matching module: The subgraph matching module is used to perform correlation analysis on the user's continuous questions and match related subgraphs based on indexes. If multiple fields are involved, multi-subgraph matching is performed;

[0042] Optimization feedback module: The optimization feedback module optimizes the subgraph matching algorithm based on user feedback and adjusts the threshold of similarity calculation to continuously improve matching accuracy;

[0043] Processor: The processor is used to process the calculation process of each formula and the construction calculation process of each model.

[0044] Compared with the prior art, the advantages of the present invention are:

[0045] By integrating structured scientific research databases with unstructured popular science literature, a multi-source fusion of popular science knowledge graphs is constructed, ensuring the comprehensiveness and depth of information. When a user asks a question, the method can accurately extract question features through a pre-trained deep learning model, perform similarity matching with the knowledge graph, and quickly extract relevant knowledge subgraphs, thereby providing accurate and targeted answers. In addition, by recording the knowledge subgraph location index of the question and performing correlation analysis when the user asks questions continuously, the system can efficiently match and update relevant knowledge and avoid redundant operations. When questions involve multiple knowledge fields, the system can perform multi-subgraph matching while ensuring accuracy, flexibly responding to complex query requirements. Through the optimization mechanism of user feedback, the matching algorithm will be continuously improved, enhancing the intelligence level of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the method proposed in the present invention;

[0047] Figure 2 This is a schematic diagram of the popular science knowledge spectrum proposed by the present invention;

[0048] Figure 3 Schematic diagram of the question feature vector proposed by the present invention;

[0049] Figure 4 This is a schematic diagram of the knowledge subgraph extraction proposed by the present invention;

[0050] Figure 5 This is a schematic diagram of the knowledge subgraph index sequence proposed by the present invention;

[0051] Figure 6 This is a schematic diagram of the correlation analysis proposed by the present invention;

[0052] Figure 7 This is a schematic diagram of the computer-readable storage medium structure in this solution. DETAILED DESCRIPTION

[0053] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may conceive of other obvious variations.

[0054] Intelligent AI interactive question-answering system for popular science, including:

[0055] Graph construction module: The graph construction module constructs a multi-source fusion science knowledge graph based on heterogeneous data processing of structured scientific research databases and unstructured science popularization literature;

[0056] Feature extraction module: The feature extraction module is used to pre-process the voice or text questions input by the user, convert them into feature vectors, and extract key question features;

[0057] Subgraph extraction module: The subgraph extraction module performs similarity calculation based on the feature vector of the user question and the node embedding space of the knowledge spectrum graph to extract relevant knowledge subgraphs;

[0058] Subgraph index module: The subgraph index module is used to store the knowledge subgraph corresponding to the question in the knowledge spectrum, and generate a positioning index and an ordered index sequence to ensure efficient access;

[0059] Subgraph matching module: The subgraph matching module is used to perform correlation analysis on the user's continuous questions and match related subgraphs based on indexes. If multiple fields are involved, multi-subgraph matching is performed;

[0060] Optimization feedback module: The optimization feedback module optimizes the subgraph matching algorithm based on user feedback and adjusts the threshold of similarity calculation to continuously improve matching accuracy;

[0061] Processor: The processor is used to process the calculation process of each formula and the construction calculation process of each model.

[0062] See Figure 1 As shown in Figure 1, the intelligent AI interactive question-answering method applied to science popularization includes:

[0063] Step 1: Based on multi-source popular science knowledge, construct a popular science knowledge spectrum through heterogeneous data processing of structured scientific research databases and unstructured popular science literature;

[0064] Step 2: Based on the user's voice or text question, the pre-trained deep learning feature extraction model extracts key feature data and generates a question feature vector;

[0065] Step 3: Perform similarity matching between the feature vector and the node embedding space of the popular science knowledge graph to extract the precise knowledge subgraph related to the question;

[0066] Step 4: After completing the answer to the current question, record the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum, and generate knowledge subgraph index sequence data;

[0067] Step 5: Based on the user's next question, a correlation analysis is performed. If the knowledge type of consecutive questions is detected, sub-graph matching is performed within the same knowledge category based on the positioning index. If the knowledge type similarity is less than the preset threshold, the full-graph matching process is re-executed. If multiple knowledge fields are involved, sub-graph matching is performed within the same knowledge category while executing the full-graph matching process.

[0068] See Figure 2As shown in the figure, based on multi-source popular science knowledge, through heterogeneous data processing of structured scientific research databases and unstructured popular science literature, the construction of a popular science knowledge spectrum specifically includes:

[0069] Based on the acquired structured scientific research database and unstructured popular science literature, knowledge is integrated and matched through keywords, themes, and concepts to effectively connect academic information with popular information.

[0070] Obtain the entities and relationships between academic and popular information, match different expressions across data sources through word alignment and semantic similarity calculation, and integrate the data;

[0071] Based on the results of data integration, a popular science knowledge graph is constructed, where each node represents an entity and each edge represents the relationship between entities.

[0072] Specifically, we extract metadata of scientific research articles from academic databases, such as titles, authors, abstracts, and keywords, and use SQL queries and API interfaces to extract and organize structured data. We also use natural language processing technology to perform text analysis on unstructured data sources to extract keywords, themes, and concepts.

[0073] Keywords are extracted from text using the TF-IDF method. The formula is:

[0074] TF-IDF(t,d)=TF(t,d)×IDF(t)

[0075] Among them, TF(t,d) is the term frequency, which means the number of times the word t appears in the document d, and IDF is the inverse document frequency. The IDF formula is: N is the total number of documents, df(t) is the number of documents containing word t;

[0076] Different expressions across data sources are aligned through dictionary mapping or synonym dictionaries. By calculating similarities, relationships are established between academic information and popular science information, matching the same concepts.

[0077] Through text mining and natural language processing, entities and their relationships in each text are extracted, and the entities and their relationships extracted from academic information and popular science information are integrated to build a unified knowledge base. The knowledge base can be represented by triples. Based on the results of data integration, a popular science knowledge graph is constructed, where each node represents an entity and each edge represents the relationship between entities.

[0078] See Figure 3 As shown in the figure, based on the voice or text questions input by the user, the key feature data is extracted through the pre-trained deep learning feature extraction model to generate the question feature vector, which specifically includes:

[0079] Converting voice signals into text data based on questions input by the user via voice;

[0080] All pre-processed text data is segmented, and the words are mapped into low-dimensional continuous vectors through word embedding methods, and semantically similar words are represented by similar vectors;

[0081] Through the context embedding method, based on the pre-trained feature extraction model, the text is input into the model as a whole to obtain the feature data of the text data and generate a more fine-grained and context-related feature vector.

[0082] Specifically, the original speech signal is converted into Mel-frequency cepstral coefficient features, and the speech signal is modeled using a deep neural network. The speech signal is decoded based on the probability distribution output by the acoustic model and the sequence output by the language model to finally obtain the text.

[0083] Through word embedding, each word is mapped to a low-dimensional continuous vector space, so that semantically similar words are closer in the vector space. Through contextual embedding, based on the self-attention mechanism of the Transformer structure, the representation of each word in the context is calculated, and the entire text is used as input to generate the context-related feature vector of each word.

[0084] After modeling the text as a whole, the contextual representation of each word is obtained through the context embedding method. At the same time, the static representation in the word embedding is combined to construct a richer text representation. Each text can ultimately be represented as a multi-dimensional feature vector. This vector not only contains the static information of the vocabulary, but also combines the contextual relationship in the text through weighted averaging, pooling, etc. to combine the vector of each word into a feature vector for the entire text.

[0085] See Figure 4 As shown in the figure, based on the similarity matching between the feature vector and the node embedding space of the popular science knowledge graph, the precise knowledge subgraph related to the question is extracted, which specifically includes:

[0086] When constructing a popular science knowledge graph, the nodes in the graph are mapped into a continuous low-dimensional vector space to obtain the structural and semantic information between the nodes;

[0087] Based on the acquired feature vector of the user's question and the embedding vector of each node in the knowledge graph, similarity is calculated using cosine similarity;

[0088] By calculating the similarity of all nodes, we select the node that is most similar to the user's question, and extract the nodes and their connected edges to form a subgraph;

[0089] Optimize the extracted subgraph and remove redundant and irrelevant nodes.

[0090] Specifically, we use the Node2Vec graph embedding algorithm to generate node sequences through random walks, and then input these sequences into the Skip-gram model to learn the low-dimensional vector representation of the nodes. In this way, each node is mapped to a low-dimensional vector space.

[0091] Each entity in the knowledge graph is mapped to a vector space, preserving the structure and semantic information between nodes. The embedded nodes not only consider their graph structure but also capture the semantic information between nodes.

[0092] The similarity between the feature vector of the user's question and the embedding vector of each node in the knowledge graph is calculated through cosine similarity, and the node with the highest similarity is selected as the best matching node;

[0093] After calculating the similarity between all nodes and the user's question, the node most similar to the question is selected, and the nodes and edges directly connected to the most similar node are extracted from the knowledge graph to form a subgraph. The node set of the subgraph includes the most similar node and its adjacent nodes.

[0094] In the extracted subgraph, there may be some redundant nodes. By further calculating the similarity between these nodes and the questions, the nodes with low similarity are deleted to reduce the complexity of the graph.

[0095] See Figure 5 As shown in the figure, after completing the answer to the current question, the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum is recorded, and the knowledge subgraph index sequence data is generated, specifically including:

[0096] Identify each node and its corresponding relationship in the extracted knowledge subgraph, and set a unique identifier for each node and relationship in the knowledge graph;

[0097] Based on the unique identifier, each node and edge is assigned a unique location index in the graph.

[0098] Record the storage location of the knowledge subgraph corresponding to the question, and arrange the storage locations of all related nodes and edges in order to generate an ordered index sequence;

[0099] The index sequence of the knowledge subgraph is checked and updated regularly, and the index sequence of the knowledge subgraph is kept up to date and valid through the incremental update mechanism.

[0100] Specifically, each node in the graph requires a unique identifier, and each relationship also requires a unique identifier to identify the type and direction of the relationship, so that it can be uniquely identified and efficiently searched in the graph;

[0101] Each node and relationship is assigned a unique storage location index in the graph storage system. This index determines the specific location of the node and relationship in the storage system and can be implemented through a hash table, database index, or row or column number in the graph database.

[0102] Based on the user's question, a related subgraph is extracted. Each node and relationship in the subgraph has a unique storage location, and the storage location of the subgraph will record the storage locations of all related nodes and relationships;

[0103] Arrange the nodes and relationships in the subgraph in a certain order to generate an ordered index sequence, which can be sorted by the relative importance of nodes, the connection order of relationships, etc.

[0104] Regularly check the existing index sequence to determine whether there are redundant or irrelevant nodes and edges. If new nodes or relationships are added to the knowledge graph, or the relationships between certain nodes change, update the corresponding storage location and index sequence in an incremental update manner.

[0105] See Figure 6 As shown, based on the user's next question, a correlation analysis is performed. If the knowledge type of consecutive questions is detected, sub-graph matching is performed within the same knowledge category based on the positioning index. If the knowledge type similarity is less than the preset threshold, the full-graph matching process is re-executed. If multiple knowledge fields are involved, sub-graph matching is performed within the same knowledge category while executing the full-graph matching process. The specific process includes:

[0106] After completing the answer to the current question, receive the user's next question, perform feature extraction and correlation analysis;

[0107] If the knowledge type of the current question belongs to the same knowledge category as the previous question, the corresponding subgraph is quickly found and loaded based on the knowledge subgraph location index of the previous question;

[0108] Calculate the similarity between the feature vector of the current question and the node embedding vector in the loaded subgraph to determine whether an accurate answer can be given by the loaded subgraph. If the similarity is lower than the preset threshold, execute the full graph matching process;

[0109] If the current problem involves multiple knowledge domains, multiple subgraph matching is performed, with preliminary subgraph matching performed within the same knowledge category, and then matching is performed within the entire graph;

[0110] Based on user feedback, the subgraph matching algorithm is optimized and the threshold for similarity calculation is adjusted.

[0111] Specifically, when the current problem involves multiple knowledge fields, the system needs to perform multi-subgraph matching, perform subgraph matching for each knowledge field, and within each field, match the feature vector of the current problem with the node embedding vector in the subgraph of that field. Integrate and associate subgraphs from different fields to ensure that cross-domain knowledge can also be effectively queried and answered. Find the most relevant subgraph among the subgraphs of multiple fields, and then perform matching and verification at the full-graph level.

[0112] Figure 7 This is a schematic diagram of the computer-readable storage medium structure provided by an embodiment of the present application. Figure 7 As shown, a computer-readable storage medium 600 according to one embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, the intelligent AI interactive question-answering method, system and storage medium for popular science according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0113] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0114] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent AI interactive question-answering method for popular science, characterized by: include: Based on multi-source popular science knowledge, a popular science knowledge spectrum is constructed through heterogeneous data processing of structured scientific research databases and unstructured popular science literature; Based on the user's voice or text questions, the pre-trained deep learning feature extraction model extracts key feature data and generates a question feature vector; Based on the similarity matching between the feature vector and the node embedding space of the popular science knowledge graph, the precise knowledge subgraph related to the question is extracted; After completing the answer to the current question, record the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum, and generate knowledge subgraph index sequence data; Based on the user's next question, a correlation analysis is performed. If the knowledge type of consecutive questions is detected, sub-graph matching is performed within the same knowledge category based on the positioning index. If the knowledge type similarity is less than the preset threshold, the full-graph matching process is re-executed. If multiple knowledge fields are involved, sub-graph matching is performed within the same knowledge category while executing the full-graph matching process.

2. The intelligent AI interactive question-answering method for popular science according to claim 1, characterized in that: The construction of a popular science knowledge graph based on multi-source popular science knowledge through heterogeneous data processing of structured scientific research databases and unstructured popular science literature specifically includes: Based on the acquired structured scientific research database and unstructured popular science literature, knowledge is integrated and matched through keywords, themes, and concepts to effectively connect academic information with popular information. Obtain the entities and relationships between academic and popular information, match different expressions across data sources through word alignment and semantic similarity calculation, and integrate the data; Based on the results of data integration, a popular science knowledge graph is constructed, where each node represents an entity and each edge represents the relationship between entities.

3. The intelligent AI interactive question-answering method for popular science according to claim 1, characterized in that: The method of extracting key feature data based on the voice or text question input by the user through a pre-trained deep learning feature extraction model to generate a question feature vector specifically includes: Converting voice signals into text data based on questions input by the user via voice; All pre-processed text data is segmented, and the words are mapped into low-dimensional continuous vectors through word embedding methods, and semantically similar words are represented by similar vectors; Through the context embedding method, based on the pre-trained feature extraction model, the text is input into the model as a whole to obtain the feature data of the text data and generate a more fine-grained and context-related feature vector.

4. The intelligent AI interactive question-answering method for popular science according to claim 1, characterized in that: The similarity matching between the feature vector and the node embedding space of the popular science knowledge graph is performed to extract the precise knowledge subgraph related to the question, specifically including: When constructing a popular science knowledge graph, the nodes in the graph are mapped into a continuous low-dimensional vector space to obtain the structural and semantic information between the nodes; Based on the acquired feature vector of the user's question and the embedding vector of each node in the knowledge graph, similarity is calculated using cosine similarity; By calculating the similarity of all nodes, we select the node that is most similar to the user's question, and extract the nodes and their connected edges to form a subgraph; Optimize the extracted subgraph and remove redundant and irrelevant nodes.

5. The intelligent AI interactive question-answering method for popular science according to claim 1, characterized in that: After completing the answer to the current question, the storage location index of the knowledge subgraph corresponding to the question in the popular science knowledge spectrum is recorded, and the knowledge subgraph index sequence data is generated, specifically including: Identify each node and its corresponding relationship in the extracted knowledge subgraph, and set a unique identifier for each node and relationship in the knowledge graph; Based on the unique identifier, each node and edge is assigned a unique location index in the graph. Record the storage location of the knowledge subgraph corresponding to the question, and arrange the storage locations of all related nodes and edges in order to generate an ordered index sequence; The index sequence of the knowledge subgraph is checked and updated regularly, and the index sequence of the knowledge subgraph is kept up to date and valid through the incremental update mechanism.

6. The intelligent AI interactive question-answering method for popular science according to claim 1, characterized in that: The process of performing correlation analysis based on the user's next question and performing subgraph matching within the same knowledge category based on the positioning index if the knowledge type similarity is less than a preset threshold is repeated. If multiple knowledge fields are involved, subgraph matching within the same knowledge category is repeated while performing the full-graph matching process. After completing the answer to the current question, receive the user's next question, perform feature extraction and correlation analysis; If the knowledge type of the current question belongs to the same knowledge category as the previous question, the corresponding subgraph is quickly found and loaded based on the knowledge subgraph location index of the previous question; Calculate the similarity between the feature vector of the current question and the node embedding vector in the loaded subgraph to determine whether an accurate answer can be given by the loaded subgraph. If the similarity is lower than the preset threshold, execute the full graph matching process; If the current problem involves multiple knowledge domains, multiple subgraph matching is performed, with preliminary subgraph matching performed within the same knowledge category, and then matching is performed within the entire graph; Based on user feedback, the subgraph matching algorithm is optimized and the threshold for similarity calculation is adjusted.

7. An intelligent AI interactive question-answering system for popular science, used to implement the intelligent AI interactive question-answering method for popular science as described in any one of claims 1 to 6, characterized in that: include: Graph construction module: The graph construction module constructs a multi-source fusion science knowledge graph based on heterogeneous data processing of structured scientific research databases and unstructured science popularization literature; Feature extraction module: The feature extraction module is used to pre-process the voice or text questions input by the user, convert them into feature vectors, and extract key question features; Subgraph extraction module: The subgraph extraction module performs similarity calculation based on the feature vector of the user question and the node embedding space of the knowledge spectrum graph to extract relevant knowledge subgraphs; Subgraph index module: The subgraph index module is used to store the knowledge subgraph corresponding to the question in the knowledge spectrum, and generate a positioning index and an ordered index sequence to ensure efficient access; Subgraph matching module: The subgraph matching module is used to perform correlation analysis on the user's continuous questions and match related subgraphs based on indexes. If multiple fields are involved, multi-subgraph matching is performed; Optimization feedback module: The optimization feedback module optimizes the subgraph matching algorithm based on user feedback and adjusts the threshold of similarity calculation to continuously improve matching accuracy; Processor: The processor is used to process the calculation process of each formula and the construction calculation process of each model.

8. A computer-readable storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a processor, the intelligent AI interactive question-answering method for popular science described in any one of claims 1 to 6 is implemented.

Images