Intelligent ancient book information processing method and device based on artificial intelligence

Abstract

The invention relates to the related technical field of digital data processing, in particular to an intelligent ancient book information processing method and device based on artificial intelligence, and the method comprises the steps: scanning an ancient book, constructing a knowledge graph, dividing a domain database, and training a vertical processing model; introducing a multi-form historical carrier to form a multi-mode sensing set, establishing a multi-information network, and embedding a processing model; the method comprises the following steps: establishing a dialog box, realizing multi-round interaction and combining threshold optimization output, solving the technical problems of digitization and fragmentation limitation in ancient book information processing and incapability of effectively integrating and utilizing ancient book information due to diversity and complexity of ancient books and literatures, and realizing construction of a knowledge graph and division of a domain database. A knowledge system with a hierarchical structure and logic association is formed, deep learning and optimization are performed according to specific requirements in the field of ancient books by training a vertical processing model, a dialog box is established, multi-round interactive feedback with the user is performed, and the user is assisted to quickly and deeply explore the ancient books.

Description

Technical Field

[0001] The present invention relates to the technical field related to digital data processing, and in particular to an ancient book information intelligent processing method and device based on artificial intelligence. Background Art

[0002] With the rapid development of science and technology, especially the widespread application of artificial intelligence and big data technology, the preservation, organization and utilization of ancient books and documents have ushered in unprecedented opportunities, greatly improving the digitization of ancient books. As the achievements of civilization experienced by the nation, ancient books carry rich historical and cultural information, but conventional organization methods are not only inefficient, but also often limited by material resources, making it difficult to fully cover the vast ancient book resources. In addition, the characters and symbols in ancient books are often complex and diverse, involving multiple languages, fonts and writing styles, which poses a great challenge to the organization and interpretation of ancient book information.

[0003] In summary, the existing technology faces limitations in digitization and fragmentation in the processing of ancient book information, and due to the diversity and complexity of ancient book documents, there is a technical problem that ancient book information cannot be effectively integrated and utilized. Summary of the invention

[0004] This application aims to solve the technical problems faced by the existing technology in the processing of ancient book information due to the limitations of digitization and fragmentation, and the inability to effectively integrate and utilize ancient book information due to the diversity and complexity of ancient book documents, by providing an intelligent processing device for ancient book information based on artificial intelligence.

[0005] In view of the above problems, the technical solution to implement this application is:

[0006] On the one hand, the present application provides an intelligent processing method for ancient book information based on artificial intelligence, wherein the method comprises: scanning ancient books to obtain scanned text, wherein the scanned text is in an editable text format; establishing an ancient book knowledge graph based on the scanned text, wherein the ancient book knowledge graph sets triples with text punctuation as constraints, and the triples are used for text editing with word segmentation and part of speech as limitations; through the ancient book knowledge graph, M domain databases are classified, and a machine learning model is used to determine M vertical pre-trained ancient book information processing models, wherein the classification indicators corresponding to the M domain databases include age, core issues, and representative ideas of the original work; introducing multi-form historical carriers, wherein the multi-form historical carriers include cultural relics, ruins, and intangible cultural heritage; based on the multi-form historical carriers and the age and core issues in the classification indicators, a first multimodal perception set is determined; based on the The multi-form historical carriers are used to determine a second multi-modal perception set with the original thought representatives and core topics in the classification indicators; a third multi-modal perception set is determined based on the multi-form historical carriers and the era and original thought representatives in the classification indicators; based on the first multi-modal perception set, the second multi-modal perception set and the third multi-modal perception set, a streaming generation mode is adopted to configure a multi-layer branch guidance structure, and the M vertical pre-trained ancient book information processing models are embedded in the multi-layer branch guidance structure to establish a multi-information network; an ancient book information processing dialog box is established to receive a first round of input information, interactively connect in the multi-information network, output a first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context joint guidance mechanism, and optimize and adjust in combination with the minimum difference threshold and the maximum difference threshold.

[0007] On the other hand, the present application provides an intelligent processing device for ancient book information based on artificial intelligence, wherein the device includes: an ancient book scanning module, which is used to scan ancient books and obtain scanned texts, wherein the scanned texts are in an editable text format; a knowledge graph establishment module, which is used to establish an ancient book knowledge graph based on the scanned texts, wherein the ancient book knowledge graph sets triples with text punctuation as constraints, and the triples are used for text editing with word segmentation and part of speech as limitations; a database classification module, which is used to classify M domain databases through the ancient book knowledge graph, and use a machine learning model to determine M vertical pre-trained ancient book information processing models, wherein the classification indicators corresponding to the M domain databases include age, core issues, and representative ideas of the original work; a historical carrier introduction module, which is used to introduce multi-form historical carriers, wherein the multi-form historical carriers include cultural relics, relics, and intangible cultural heritage; a multimodal perception module, which is used to classify the multi-form historical carriers and the age and core issues in the classification indicators based on the age and core issues in the classification indicators. Determine a first multimodal perception set; determine a second multimodal perception set based on the multi-form historical carriers and the original thought representatives and core topics in the classification indicators; determine a third multimodal perception set based on the multi-form historical carriers and the era and original thought representatives in the classification indicators; a multi-information network establishment module is used to configure a multi-layer branch guidance structure based on the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set in a streaming generation mode, and embed the M vertical pre-trained ancient book information processing models into the multi-layer branch guidance structure to establish a multi-information network; an optimization and adjustment module is used to establish an ancient book information processing dialog box, receive a first round of input information, perform interactive connections in the multi-information network, output a first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context joint guidance mechanism, and optimize and adjust in combination with the minimum difference threshold and the maximum difference threshold.

[0008] In summary, one or more technical solutions provided in this application solve the limitations of digitization and fragmentation faced in the processing of ancient book information, as well as the technical problem that ancient book information cannot be effectively integrated and utilized due to the diversity and complexity of ancient books and documents. It realizes the construction of knowledge graphs and the division of domain databases on the basis of digitization of ancient books and documents, forming a knowledge system with a hierarchical structure and logical associations, and through the training of vertical processing models, deep learning and optimization are carried out for the specific needs of the ancient book field, and dialog boxes are established to assist users in quickly and deeply exploring the technical effects of ancient books through multiple rounds of interactive feedback with users. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of an intelligent processing method for ancient book information based on artificial intelligence is provided for this application;

[0010] Figure 2 A structural schematic diagram of an ancient book information intelligent processing device based on artificial intelligence is provided for this application.

[0011] Explanation of the accompanying drawings: ancient book scanning module M100, knowledge graph establishment module M200, database classification module M300, historical carrier introduction module M400, multimodal perception module M500, multi-information network establishment module M600, optimization and adjustment module M700. DETAILED DESCRIPTION

[0012] Embodiment 1

[0013] The present application is described in detail below with reference to the accompanying drawings. Figure 1 As shown, the present application provides an ancient book information intelligent processing method based on artificial intelligence, wherein the method comprises:

[0014] S1: Scan ancient books to obtain scanned text, which is in an editable text format; S2: Build an ancient book knowledge graph based on the scanned text, wherein the ancient book knowledge graph sets triples based on text punctuation constraints, and the triples are used for text editing based on word segmentation and part of speech.

[0015] Choose equipment suitable for scanning ancient books to ensure that every detail of the ancient books can be captured with high quality; place the ancient book on the scanner and scan it page by page or in batches to obtain the scanned image of the ancient book; use optical character recognition (OCR) technology to convert the scanned image into an editable text format (such as TXT, DOCX, etc.). It is necessary to select or train an OCR model suitable for ancient book texts to improve the recognition accuracy; since there is a certain probability of errors in OCR recognition, the recognized text needs to be proofread to correct typos, format errors and other problems to ensure the accuracy of the text.

[0016] Remove irrelevant information in the text, such as page numbers, notes, illustrations, etc., and only keep the main text content of the ancient book; use the word segmentation tool to segment the cleaned text, and divide the continuous text into independent words or phrases. This is especially important for ancient Chinese books because Chinese text does not have natural spaces to separate words; perform part-of-speech tagging on the results after word segmentation, and assign a part-of-speech label (such as noun, verb, adjective, etc.) to each word or phrase, which will help understand the role and relationship of words in sentences when constructing the knowledge graph later.

[0017] Define the triple structure. In the ancient book knowledge graph, triples are the basic knowledge representation units, usually composed of "entity-relationship-entity" or "entity-attribute-value". Triples are set with text punctuation as constraints, which means that punctuation marks (such as periods, commas, semicolons, etc.) can be used to identify independent information units in sentences or clauses, and then construct triples; based on the results of word segmentation and part-of-speech tagging, combined with text punctuation, triples are extracted from the text. It is necessary to develop special algorithms or use natural language processing (NLP) technology to identify entities, relationships and attributes.

[0018] The extracted triples are stored and displayed in the form of a graph structure to form an ancient book knowledge graph. In the process of establishing the ancient book knowledge graph, it is necessary to assign a unique identifier to each entity and establish connections for the relationships between entities; the constructed knowledge graph is optimized, such as removing redundant information, merging duplicate entities, etc. At the same time, it is verified through automated means to ensure the accuracy and completeness of the knowledge graph.

[0019] S3: Through the ancient book knowledge graph, M domain databases are classified, and a machine learning model is used to determine M vertical pre-trained ancient book information processing models. The classification indicators corresponding to the M domain databases include age, core issues, and representative ideas of the original works.

[0020] Clarifying the specific categories of the M fields can be based on subject classification (such as history, literature, philosophy, medicine, etc.), time span (such as dynasties, eras, etc.), subject content (such as politics, economy, culture, etc.) or any other standards that help to subdivide the content of ancient books. The classification indicators corresponding to the M field databases include era (divided according to the dynasty in which the ancient books were created, which helps to understand the background and value of ancient books in different historical periods), core issues (analyzing the main discussion content of ancient books and classifying them into corresponding topics, such as political systems, economic thoughts, literature and art, etc.), and original thought representatives (evaluating the representativeness of ancient books in a certain ideological system or school, such as Confucian classics, Taoist works, etc.); according to the defined fields and classification indicators, the information in the ancient book knowledge map is classified into the corresponding field database.

[0021] Clean the data in each domain database to remove duplicate, erroneous or irrelevant information; extract features useful for machine learning models from the cleaned data, such as keywords, subject terms, sentiment tendencies, etc.; if a supervised learning model is required, some data needs to be annotated to provide the labels required for training; according to the specific tasks of ancient book information processing (such as classification, clustering, summary generation, sentiment analysis, etc.), select appropriate machine learning models, such as neural networks, decision trees, support vector machines, etc.; use the data in the domain database to pre-train the model. The purpose of pre-training is to enable the model to learn the knowledge and patterns unique to each domain, thereby improving the accuracy and efficiency of subsequent task processing; evaluate the performance of the model through methods such as cross-validation, and adjust parameters or optimize the model as needed.

[0022] Deploy the trained vertical pre-trained ancient book information processing model to corresponding application scenarios, such as ancient book digitization platforms, academic research tools, cultural heritage projects, etc.; use the model to process and analyze new ancient book information in real time to provide fast and accurate results; with the continuous addition of new data and the continuous development of technology, regularly update and retrain the model to maintain its adaptability and accuracy; collect user feedback on the model processing results, and regularly evaluate the model's performance, including accuracy, recall rate, F1 score and other indicators; iteratively optimize the model based on user feedback and performance evaluation results to continuously improve its processing capabilities and user experience.

[0023] S4: Introduce multi-form historical carriers, which include cultural relics, relics, and intangible cultural heritage; S5: Determine a first multi-modal perception set based on the multi-form historical carriers and the era and core issues in the classification indicators; determine a second multi-modal perception set based on the multi-form historical carriers and the original ideas and core issues in the classification indicators; determine a third multi-modal perception set based on the multi-form historical carriers and the era and original ideas in the classification indicators.

[0024] In the process of constructing the ancient book knowledge graph, introducing multiple forms of historical carriers (such as cultural relics, remains, and intangible cultural heritage) is an effective way to enrich and deepen knowledge representation. Introducing multiple forms of historical carriers not only provides the visual, physical, and cultural background of the ancient book text, but also enhances the understanding and interpretation of the content of the ancient books.

[0025] The first multimodal perception set focuses on the specific era in which the ancient books are located and their core issues. By combining physical evidence such as cultural relics and ruins with the era information and core issue descriptions in the ancient book texts, a comprehensive perception of the historical background and subject content of the ancient books is formed. Furthermore, archaeological discoveries, historical documents and other materials are used to determine the approximate age range of the ancient books, and cultural relics and ruins related to this era are selected as auxiliary materials; core issues in ancient books are extracted through text analysis, such as political systems, economic development, cultural changes, etc., and cultural relics, ruins or intangible cultural heritage related to these issues are found as examples; information from multiple modalities such as text descriptions, image displays (photos or restorations of cultural relics and ruins), and sound records (such as related commentaries, historical audio materials) are integrated to form a comprehensive perception of the era and core issues.

[0026] The second multimodal perception set focuses on the original ideas represented by ancient books and their embodiment in core issues. By deeply exploring the ideological connotations of ancient books and their contributions to specific issues, combined with relevant cultural relics, intangible cultural heritage, etc., a deep understanding of the original ideas and their influence is formed. Furthermore, through methods such as text analysis and intellectual history research, the original ideas in ancient books are refined, such as philosophical viewpoints, political propositions, cultural concepts, etc.; the extracted original ideas are correlated with the core issues in ancient books to clarify which issues the original ideas have played an important role in; physical evidence such as cultural relics, intangible cultural heritage, and related audio and video materials are used to demonstrate the inheritance and influence of the original ideas in history, and enhance the understanding of the original ideas and their core issues.

[0027] The third multimodal perception set combines the age of ancient books with the original ideas they represent, and reveals the ideological value and cultural significance of ancient books in a specific historical context through the display of multiple forms of historical carriers. Further, it deeply analyzes the historical context of the ancient books, including changes and developments in politics, economy, culture, etc.; combined with the background of the age, it explores the original ideas in the ancient books and understands their unique status and influence in the society at that time; it uses various forms of historical carriers such as cultural relics, ruins, intangible cultural heritage, as well as related text, images, audio and other multimedia materials to present the original ideas and cultural significance of ancient books in a specific era. At the same time, it can provide users with a more immersive perceptual experience through modern technical means such as virtual reality (VR) and augmented reality (AR).

[0028] After introducing multi-form historical carriers, the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set are determined to more comprehensively and deeply understand and display the historical value and cultural connotation of ancient books, providing a richer and more diverse source of information for the construction of the ancient book knowledge graph.

[0029] S6: Based on the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set, a streaming generation mode is adopted to configure a multi-layer branch guidance structure, and the M vertical pre-trained ancient book information processing models are embedded in the multi-layer branch guidance structure to establish a multivariate information network.

[0030] The streaming generation mode is a method of sequentially processing data, allowing data to be continuously transmitted and processed in the system in the form of a stream. In this scenario, the data in the multimodal perception set is regarded as an input stream, and the required output information is gradually generated through a series of processing steps (i.e., each layer in the multi-layer branch guidance structure).

[0031] A multi-layer branch guidance structure is configured, including an input layer, which receives data from the first, second, and third multimodal perception sets, including information in various forms such as text, image, and audio; a preprocessing layer, which preprocesses the input data, such as text cleaning, image denoising, and audio conversion, to ensure the quality and consistency of the data; a feature extraction layer, which uses a special algorithm or model to extract features from the preprocessed data, which can be keywords in the text, visual features in the image, and spectral features in the audio; a branching layer, which distributes feature data to different branches according to the characteristics and processing requirements of the data, and each branch corresponds to one or more vertical pre-trained ancient book information processing models, which specialize in processing specific types of information or tasks; a model processing layer, in which each vertical pre-trained model receives corresponding feature data and performs its specific processing tasks, such as classification, clustering, and sentiment analysis; a fusion layer, which fuses the processing results of each branch model to form a comprehensive understanding and representation of ancient book information, and it is necessary to develop a special fusion algorithm or strategy to ensure the effective integration of different modalities and different types of information; an output layer, which outputs the fused results in an appropriate form, such as generating reports, visual displays, and providing decision support.

[0032] Based on the multi-layer branch guidance structure, in each branch layer, the corresponding vertical pre-trained ancient book information processing model is selected and embedded according to the data type and task requirements processed by the branch. The vertical pre-trained ancient book information processing model has been pre-trained on data in related fields, so it can efficiently process specific types of ancient book information. By embedding M vertical pre-trained ancient book information processing models into the multi-layer branch guidance structure, parallel processing and cross-modal fusion of multimodal information are realized.

[0033] With the establishment of a multi-layer branch-guided structure and the embedding of models, a multivariate information network is gradually formed. The multivariate information network can receive historical carrier data from different channels and forms, and generate a comprehensive and in-depth understanding and representation of ancient book information through a multi-level and multi-branch processing flow. At the same time, due to the embedding of multiple vertical pre-trained models, the multivariate information network also has a high degree of professionalism and flexibility, and can be customized and optimized according to different task requirements.

[0034] S7: Establish an ancient book information processing dialog box, receive the first round of input information, perform interactive connection in the multi-information network, output the first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context-joint guidance mechanism, and perform optimization and adjustment in combination with the minimum difference threshold and the maximum difference threshold.

[0035] Load the established multivariate information network, including a multi-layer branch-guided structure and an embedded vertical pre-trained ancient book information processing model; set the minimum difference threshold and the maximum difference threshold to control the similarity and difference of the response information; provide an intuitive user interface to allow users to input queries, questions or instructions about ancient books; parse the information input by the user to identify its type (such as text, image, etc.) and main content; map the parsed input information to the corresponding nodes or branches in the multivariate information network; use the embedded vertical pre-trained model to process the input information and generate preliminary response information; establish or update the context information library to store the context information of the current conversation, including historical input, response and user interaction behavior.

[0036] Generate the first round of output information based on the results of model processing and contextual information; feed back the output information to the user in an appropriate form (such as text, image, voice, etc.); when the user inputs the second round, third round and other subsequent information, identify the association between this information and the current context; use the context joint guidance mechanism to combine historical information and current input for joint reasoning to generate a more accurate response.

[0037] When the current cumulative difference index is less than the minimum difference threshold, the current cumulative difference index is ignored, which also indicates that continuing pre-training can enhance the model's ability to process domain data; when the current cumulative difference index is greater than the maximum difference threshold, roll back, which also indicates that when there is a large difference in language features from general data, only domain data cannot be used for training, otherwise catastrophic forgetting may occur.

[0038] For subsequent input information such as the second and third rounds, repeat the above steps to generate and optimize response information; maintain continuous interaction with the user until the user ends the conversation or reaches the preset interaction round limit; collect user feedback on system responses, including satisfaction, accuracy, etc.; based on user feedback and system performance, regularly adjust and optimize model parameters and structures in the multivariate information network; according to actual application conditions, adjust the minimum difference threshold and maximum difference threshold in a timely manner to further improve system performance; establish an efficient and intelligent ancient book information processing dialog box to provide users with convenient and accurate ancient book information query and interactive experience.

[0039] Furthermore, a machine learning model is used to determine M vertical pre-trained ancient book information processing models. The method of this application includes:

[0040] Perform fine-grained semantic analysis on the scanned text and add fine-grained annotation information, wherein the fine-grained annotation information includes a core vocabulary and a first internal logical relationship with a core topic, and a core phrase and a second internal logical relationship with the core topic; use a machine learning model to perform topic modeling, and continue to use M domain databases for pre-training to ensure that the pre-training data has fine-grained annotation information to support the training of the semantic analysis layer.

[0041] The scanned text is segmented according to language rules to form a sequence of words or phrases; each word is assigned a part-of-speech tag, such as noun, verb, adjective, etc.; the core vocabulary in the text is extracted using algorithms such as TF-IDF and TextRank; and core phrases are identified through methods such as N-gram model and dependency syntax analysis.

[0042] Internal logical relationship annotation: Further, the first internal logical relationship refers to analyzing the direct relationship between each core word and the core topic, such as causal relationship, parallel relationship, modification relationship, etc., and annotating it; the second internal logical relationship refers to analyzing the logical relationship between each core phrase and the core topic, and annotating it in more detail. This may involve complex relationships between the internal components of the phrase, such as subject-predicate relationship, verb-object relationship, etc.; design a set of annotation systems, such as XML, JSON and other formats, to store text, core words, core phrases and their internal logical relationships in a structured way.

[0043] Collect a large amount of text data related to ancient books from M domain databases; ensure that these pre-training data are also fine-grained semantically parsed and annotated, including core vocabulary, core phrases and their inherent logical relationship with core topics; select topic modeling models suitable for processing text data, such as LDA (latent Dirichlet allocation), NMF (non-negative matrix factorization), etc.; according to the needs of fine-grained annotation information, select appropriate deep learning models, such as BERT, GPT, etc., for training the semantic parsing layer.

[0044] The model is pre-trained using pre-training data with fine-grained annotations so that the model can learn the fine-grained semantic features in the text; based on the pre-training, the model is fine-tuned using task-specific datasets to improve the performance of the model on specific tasks; the results of topic modeling (such as topic distribution) are combined with the results of fine-grained semantic parsing to form a more comprehensive and in-depth understanding of the text; users are allowed to conduct interactive queries based on the results of topic modeling, and the system uses the semantic parsing layer to return more detailed and relevant text fragments or explanations.

[0045] The performance of the model in fine-grained semantic parsing and topic modeling is evaluated through cross-validation, A / B testing and other methods; based on the evaluation results and user feedback, the model structure and parameter settings are continuously optimized to improve the overall performance and user experience. Through the above steps, an ancient book information processing system capable of handling fine-grained semantic parsing and topic modeling is built to provide users with more accurate and in-depth information query and analysis services.

[0046] Furthermore, the scanned text is subjected to fine-grained semantic analysis and fine-grained annotation information is added. The method of the present application further includes:

[0047] Using NLP technology, define a set of ancient Chinese linguistics rules and a set of modern Chinese linguistics rules; in the ancient book information processing dialog box, generate a first task instruction and a first text to be processed through the ancient Chinese linguistics rule set and the modern Chinese linguistics rule set, wherein the first task instruction activates an NLP dynamic response mechanism; utilizing the NLP dynamic response mechanism, comparing with the text processing requirements corresponding to the first task instruction, selectively trigger any one of a core vocabulary extraction mode and a core phrase extraction mode.

[0048] Using NLP technology, we define a set of ancient Chinese linguistics rules and a set of modern Chinese linguistics rules. The ancient Chinese linguistics rule set includes: vocabulary rules, which refer to the unique vocabulary, homophones, variant characters, word usage, etc. in ancient Chinese; syntactic rules, which refer to the unique sentence structure of ancient Chinese, such as inverted sentences, omitted sentences, and judgment sentences; semantic rules, which refer to metaphors, allusions, cultural background knowledge, etc. in ancient Chinese, which are crucial for understanding the deep meaning of the text.

[0049] A collection of modern Chinese linguistic rules, including vocabulary rules, which refer to commonly used vocabulary, new words, foreign words, etc. in modern Chinese; syntactic rules, which refer to the grammatical structure of modern Chinese, including basic sentence patterns such as subject, predicate, object, attributive, adverbial, and complement; semantic rules refer to contextual understanding, polysemous word analysis, and sentiment analysis of modern Chinese.

[0050] In the ancient book information processing dialog box, according to user input or preset scenarios, the system first generates the first task instruction, which clearly points out the text processing task to be performed (such as translation, summary, information extraction, etc.). At the same time, the system also generates or receives the first text to be processed, that is, the ancient or modern text to be processed. The first task instruction is designed to be the key to activate the NLP dynamic response mechanism, which can dynamically select and configure the most suitable NLP processing flow according to the specific content of the task instruction.

[0051] Within the NLP dynamic response mechanism, the system will analyze and determine the core text processing mode to be adopted based on the text processing requirements corresponding to the first task instruction, including but not limited to the core vocabulary extraction mode: when the task requirements focus on identifying key words in the text, such as names of people, places, proper nouns, etc., the system will trigger this mode and use predefined vocabulary rules and context analysis technology to extract core vocabulary from the text; core phrase extraction mode: when the task needs to identify and extract key phrases or sentences in the text, such as parts that express core ideas and important information, the system will select this mode. Through syntactic analysis and semantic understanding technology, the system can identify and extract core phrases that meet the task requirements.

[0052] After selecting the appropriate text processing mode, the system will process the first text to be processed accordingly and generate processing results, including translated text, summary, keyword list, core phrase list, etc., depending on the requirements of the task instruction; the processing results will be fed back to the user to complete the entire processing flow. Through the above steps, the ancient book information processing dialog system can use NLP technology to efficiently process ancient and modern texts to meet the diverse information needs of users.

[0053] Furthermore, the present application method also includes:

[0054] With a multi-level attention mechanism, a long short-term memory network is used, and a gated recurrent unit is added, wherein the gated recurrent unit is used to enhance the recognition of multiple text sequences of different lengths and types; in the ancient book information processing dialog box, a second task instruction and a second text to be processed are generated through the gated recurrent unit, wherein the second task instruction activates a grammatical analysis mechanism; utilizing the grammatical analysis mechanism, in accordance with the text processing requirements corresponding to the second task instruction, any one of a part-of-speech tag marking mode and a punctuation tag marking mode is selectively triggered.

[0055] A multi-level attention mechanism is used to enhance the ability of LSTM to better capture key information in the text. The multi-level attention mechanism allows the model to focus on different parts of the text at different levels, thereby improving the ability to understand complex text structures. LSTM is responsible for processing long-term dependencies in text sequences, retaining important information and forgetting irrelevant information through its internal "gate" structure (forget gate, input gate, output gate).

[0056] In the gated recurrent unit (GRU), by simplifying the structure of LSTM (such as merging the forget gate and input gate into an update gate), the computational efficiency and training speed of the model are improved. In ancient book information processing, the flexibility of GRU enables it to effectively process text sequences of various lengths, including short poems and long ancient texts. Through the update gate and reset gate of GRU, the model can dynamically adjust its internal state to adapt to the characteristics of different text sequences.

[0057] In the ancient book information processing dialog box, when the user raises a new query or instruction, the system processes the input text through GRU to generate a second task instruction and a second text to be processed. At the same time, the attention mechanism of GRU will help the system identify the key parts of the user's intention, thereby generating more accurate task instructions. The second task instruction is designed to be a key signal that can activate the grammatical analysis mechanism; the grammatical analysis mechanism is a core module in the system, responsible for parsing the grammatical structure of the text sequence. When the second task instruction is received, the grammatical analysis mechanism is activated and prepares to further process the second text to be processed.

[0058] According to the text processing requirements corresponding to the second task instruction, the grammatical analysis mechanism will selectively trigger any one of the part-of-speech tag marking mode or the punctuation tag marking mode, where the part-of-speech tag marking mode: when the task needs to identify the part-of-speech information in the text (such as nouns, verbs, adjectives, etc.), the system will start the part-of-speech tag marking mode. Using the pre-trained part-of-speech tagging model or rule set, the system can assign the correct part-of-speech tag to each word in the text. Punctuation tag marking mode: when the task focuses on punctuation marks or sentence boundaries in the text, the system will trigger the punctuation tag marking mode, including identifying sentence terminators in the text (such as periods, question marks, exclamation marks, etc.), and processing the impact of other punctuation marks (such as commas, semicolons, quotation marks, etc.) on the text structure.

[0059] After selecting the appropriate tagging mode, the system will process the second text to be processed accordingly and generate processing results, including text with part-of-speech tags, text structure with punctuation tags, etc., depending on the requirements of the task instructions; the processing results will be fed back to the user to complete the entire processing flow. Through the above steps, combined with multi-level attention mechanism, LSTM, GRU and grammatical analysis mechanism, the ancient book information processing dialog system can efficiently process text sequences of various lengths and types to meet the diverse needs of users.

[0060] Further, the context joint guidance mechanism is used to respond to the second round of input information, the third round of input information, ..., the Nth round of input information, and optimize and adjust in combination with the minimum difference threshold and the maximum difference threshold. The method of the present application includes:

[0061] A minimum difference threshold and a maximum difference threshold are set, and the first round output information is determined through the first round input information; the second round output information is determined through the first round input information, the first round output information, and the second round input information using the context joint guidance mechanism; the third round output information is determined through the first round input information, the first round output information, the second round input information, the second round output information, and the third round input information using the context joint guidance mechanism; the context joint guidance mechanism is repeated N-1 times, and a difference evaluation is performed simultaneously, and when the current cumulative difference index meets the minimum difference threshold but does not meet the maximum difference threshold, the M vertical pre-trained ancient book information processing models are synchronously enhanced with the highest similarity with the pre-training stage as the learning direction.

[0062] Setting the minimum difference threshold and the maximum difference threshold is an important step to ensure that the model can be optimized gradually and smoothly during the learning process while avoiding overfitting. At the same time, the context joint guidance mechanism is used to continuously generate multiple rounds of output information, and guide the synchronous enhancement of the model based on the differences in this information. Specifically, setting the difference threshold includes: the minimum difference threshold: defines a minimum difference standard. When the output change of the model is less than this threshold, the model is considered to be close to convergence; the maximum difference threshold: defines a maximum difference upper limit to prevent the model from fluctuating too much during the learning process, which helps to avoid the model from falling into an unstable state or overfitting.

[0063] The first round of input information is the initial query or text provided by the user; the first round of output information is the preliminary result generated based on the first round of input information using the current state of M vertical pre-trained ancient book information processing models; the context joint guidance mechanism uses the previous input and output information, as well as the current input information, to jointly guide the generation of the next round of output information, including: combining the first round of input information, the first round of output information and the second round of input information, and generating the second round of output information through the context joint guidance mechanism; repeating the above process, using more context information in turn to generate the third round, the fourth round... until the Nth round of output information.

[0064] After each round of output information is generated, the difference between the output information of the current round and the previous round is calculated and accumulated to form the current cumulative difference index; when the current cumulative difference index meets the minimum difference threshold (indicating that the model is learning stably) but does not reach the maximum difference threshold (to avoid over-adjustment), the model enhancement process is triggered.

[0065] The learning direction with the highest similarity to the pre-training stage is selected, and the M vertical pre-trained ancient book information processing models are synchronously enhanced. The learning direction with the highest similarity evaluation is used to ensure that the model can be optimized gradually and smoothly during the learning process; based on the performance of the model on the validation set or a specific test set, or based on some measurement of the internal parameters or output of the model; the above process (from the second round to the Nth round) will be repeated N-1 times to gradually optimize the output of the model; with the generation of output information and the progress of difference evaluation in each round, the model will continue to adjust in a better direction until the stopping condition is met (such as reaching the preset number of iterations, performance improvement is no longer significant, etc.), which can gradually improve the processing ability of ancient book information while ensuring stability, and provide users with more accurate and useful output information.

[0066] Furthermore, while performing a difference assessment, the present application method also includes:

[0067] Configure the single-round difference measurement formula: Where, ΔT round The difference index used to characterize a round of dialogue, t i,input It is used to represent the weight of the i-th topic in the input information, p i,output Used to characterize the weight of the i-th topic in the output information; configure the multi-round difference measurement formula: ΔT total It is used to characterize the current cumulative difference index corresponding to multiple rounds of dialogue, where R is the number of rounds.

[0068] Through the single-round difference measurement formula: Calculate the sum of the absolute differences in the weights of all topics in the input and output information in a round of dialogue, which reflects the degree of information change in this round of dialogue; configure the multi-round difference measurement formula: Obtaining the degree of cumulative information change throughout the conversation helps evaluate the coherence of the entire conversation process and the effectiveness of information processing.

[0069] In the dialog box for intelligent processing of ancient book information, when the user asks a question or inputs information, the difference evaluation formula calculates the difference index of each round of dialogue and the cumulative difference index of the entire dialogue; if the current cumulative difference index meets the minimum difference threshold but does not reach the maximum difference threshold, the information processing process is considered to be in a good learning state. Through difference evaluation, the intelligent processing process of ancient book information can be monitored and adjusted in real time to ensure that the system can respond to user needs accurately and coherently, thereby improving the quality and efficiency of information processing.

[0070] Furthermore, the application method includes:

[0071] Connect to the user terminal device, obtain the user location information, and match it with the location tag. After the match is successful, establish an interactive interface using Vue.js technology, and the interactive interface includes an output reading plug-in; request access to the audio output permission of the user terminal device, and after the user accepts, the output reading plug-in provides an audio guide service.

[0072] Front-end part (user terminal device): Use HTML5 Geolocation API to obtain the user's current location information (latitude and longitude); considering user privacy, when requesting location information, the user should be clearly informed and their consent should be obtained.

[0073] Backend part: receives the location information sent by the frontend; matches the location information with the preset location tag (which can be the location ID and the corresponding longitude and latitude range in the database); and returns the matching result to the frontend.

[0074] Front-end part (Vue.js): Use Vue.js framework to build user interface; receive the location matching results returned by the back-end and display relevant content accordingly; introduce output reading plug-in (such as using HTML5 <audio>tag or third-party libraries such as howler.js).

[0075] Before trying to play audio, make sure that you have obtained the user's audio output permission; before playing audio, request access to the audio output permission of the user's terminal device (corresponding to audio modules such as speakers), which needs to be requested when the application starts or before trying to play; considering user experience and privacy, be sure to clearly inform the user how the location information will be used, and test compatibility on various devices and browsers.

[0076] In summary, the beneficial effects of the embodiments of the present application are:

[0077] 1. Through OCR recognition and NLP technology, accelerate the digitization process of ancient books, improve the speed of information processing, use semantic analysis and topic modeling, and enhance the ability to understand and interpret the content of ancient books.

[0078] 2. Through fine-grained annotation and topic modeling, the ancient book knowledge graph is automatically constructed to facilitate knowledge query and association analysis, and combined with historical carriers such as cultural relics, ruins and other information to form a multimodal perception set and enrich the dimension of ancient book information.

[0079] 3. Establish an ancient book information processing dialog box to realize natural language interaction with users, provide more personalized information services, and continuously optimize the information processing model through context joint guidance mechanism and difference evaluation algorithm to improve response quality and accuracy.

[0080] 4. Due to the adoption of setting the minimum difference threshold and the maximum difference threshold, and determining the first round of output information through the first round of input information; determining the second round of output information through the first round of input information, the first round of output information, and the second round of input information, through the context joint guidance mechanism; determining the third round of output information through the first round of input information, the first round of output information, the second round of input information, the second round of output information, and the third round of input information, through the context joint guidance mechanism; repeating the context joint guidance mechanism N-1 times, while performing difference evaluation, when the current cumulative difference index meets the minimum difference threshold and does not meet the maximum difference threshold, the learning direction with the highest similarity with the pre-training stage is used to synchronously enhance the M vertical pre-trained ancient book information processing models. It can gradually improve the processing ability of ancient book information while ensuring stability, and provide users with more accurate and useful output information.

[0081] Embodiment 2

[0082] Based on the same inventive concept as the ancient book information intelligent processing method based on artificial intelligence in the aforementioned embodiment, Figure 2 As shown, the embodiment of the present application provides an intelligent processing device for ancient book information based on artificial intelligence, wherein the device includes:

[0083] The ancient book scanning module M100 is used to scan ancient books and obtain scanned text, which is in an editable text format;

[0084] The knowledge graph establishment module M200 is used to establish an ancient book knowledge graph based on the scanned text, wherein the ancient book knowledge graph sets triples based on text punctuation as constraints, and the triples are used for text editing based on segmentation and part of speech;

[0085] The database classification module M300 is used to classify M domain databases through the ancient book knowledge graph, and use a machine learning model to determine M vertical pre-trained ancient book information processing models. The classification indicators corresponding to the M domain databases include age, core topics, and representative ideas of the original works;

[0086] The historical carrier introduction module M400 is used to introduce various forms of historical carriers, including cultural relics, relics, and intangible cultural heritage;

[0087] The multimodal perception module M500 is used to determine a first multimodal perception set based on the multi-form historical carriers, the era and the core topics in the classification indicators; determine a second multimodal perception set based on the multi-form historical carriers, the original ideas and the core topics in the classification indicators; determine a third multimodal perception set based on the multi-form historical carriers, the era and the original ideas in the classification indicators;

[0088] A multivariate information network establishment module M600 is used to configure a multi-layer branch guidance structure based on the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set in a streaming generation mode, and embed the M vertical pre-trained ancient book information processing models into the multi-layer branch guidance structure to establish a multivariate information network;

[0089] The optimization and adjustment module M700 is used to establish an ancient book information processing dialog box, receive the first round of input information, perform interactive connection in the multi-information network, output the first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context-joint guidance mechanism, and perform optimization and adjustment in combination with the minimum difference threshold and the maximum difference threshold.

[0090] Furthermore, the database classification module M300 is used to execute the following method:

[0091] Performing fine-grained semantic analysis on the scanned text and adding fine-grained annotation information, wherein the fine-grained annotation information includes a core word and a first internal logical relationship with a core topic, and a core phrase and a second internal logical relationship with the core topic;

[0092] Use machine learning models for topic modeling, and continue to use M domain databases for pre-training to ensure that the pre-training data has fine-grained annotation information to support the training of the semantic parsing layer.

[0093] Furthermore, the database classification module M300 is also used to execute the following method:

[0094] Use NLP technology to define ancient Chinese linguistics rule sets and modern Chinese linguistics rule sets;

[0095] In the ancient book information processing dialog box, a first task instruction and a first text to be processed are generated by using the ancient Chinese linguistics rule set and the modern Chinese linguistics rule set, wherein the first task instruction activates the NLP dynamic response mechanism;

[0096] By utilizing the NLP dynamic response mechanism and comparing the text processing requirements corresponding to the first task instruction, any one of the core vocabulary extraction mode and the core phrase extraction mode is selectively triggered.

[0097] Furthermore, the database classification module M300 is also used to execute the following method:

[0098] A multi-level attention mechanism is used, a long short-term memory network is used, and a gated recurrent unit is added to enhance the recognition of multiple text sequences of different lengths and types;

[0099] In the ancient book information processing dialog box, a second task instruction and a second text to be processed are generated through a gated loop unit, wherein the second task instruction activates a syntax analysis mechanism;

[0100] By utilizing the grammatical analysis mechanism and comparing with the text processing requirement corresponding to the second task instruction, any one of the part-of-speech tag marking mode and the punctuation tag marking mode is selectively triggered.

[0101] Furthermore, the optimization and adjustment module M700 is used to perform the following method:

[0102] Set the minimum difference threshold and the maximum difference threshold, and determine the first round of output information through the first round of input information;

[0103] Determine the second round output information by using the first round input information, the first round output information, and the second round input information by using the context joint guidance mechanism;

[0104] Determine the third round output information by using the first round input information and the first round output information, the second round input information and the second round output information, and the third round input information by using the context joint guidance mechanism;

[0105] The context-joint guidance mechanism is repeated N-1 times, and difference evaluation is performed at the same time. When the current cumulative difference index meets the minimum difference threshold but does not meet the maximum difference threshold, the M vertical pre-trained ancient book information processing models are synchronously enhanced with the highest similarity with the pre-training stage as the learning direction.

[0106] Furthermore, the optimization and adjustment module M700 is also used to execute the following method:

[0107] Configure the single-round difference measurement formula:

[0108] Where, ΔT round The difference index used to characterize a round of dialogue, t i,input It is used to represent the weight of the i-th topic in the input information, p i,output Used to represent the weight of the i-th topic in the output information;

[0109] Configure the multi-round difference measurement formula: ΔT total It is used to characterize the current cumulative difference index corresponding to multiple rounds of dialogue, where R is the number of rounds.

[0110] Furthermore, the optimization and adjustment module M700 is also used to execute the following method:

[0111] Connect to the user terminal device, obtain the user location information, and match it with the location tag. After the match is successful, establish an interactive interface using Vue.js technology, and the interactive interface includes an output reading plug-in;

[0112] A request is made to access the audio output permission of the user terminal device. After the user accepts, the output reading plug-in provides an audio guide service.

[0113] In summary, any step can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor, and no unnecessary restrictions are made here.

[0114] Furthermore, the above-mentioned technical scheme only reflects the preferred technical scheme of the technical scheme of the embodiment of the present application. Some changes that may be made to some parts thereof by technicians in this technical field all reflect the novel principles of the embodiment of the present application. Obviously, technicians in this field can make various changes and modifications to the present application without departing from the scope of the present application.< / audio>

Claims

1. An intelligent processing method for ancient book information based on artificial intelligence, characterized in that: The method comprises: Scanning ancient books to obtain scanned text, wherein the scanned text is in an editable text format; Based on the scanned text, an ancient book knowledge graph is established, wherein the ancient book knowledge graph sets triples based on text punctuation as constraints, and the triples are used for text editing based on segmentation and part of speech; Through the ancient book knowledge graph, M domain databases are classified, and a machine learning model is used to determine M vertical pre-trained ancient book information processing models. The classification indicators corresponding to the M domain databases include age, core topics, and representative ideas of the original works; Introducing multiple forms of historical carriers, including cultural relics, relics, and intangible cultural heritage; Based on the multi-form historical carriers, the era and core topics in the classification indicators, a first multi-modal perception set is determined; based on the multi-form historical carriers, the original ideas and core topics in the classification indicators, a second multi-modal perception set is determined; based on the multi-form historical carriers, the era and the original ideas in the classification indicators, a third multi-modal perception set is determined; Based on the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set, a streaming generation mode is adopted to configure a multi-layer branch guidance structure, and the M vertical pre-trained ancient book information processing models are embedded in the multi-layer branch guidance structure to establish a multivariate information network; Establish an ancient book information processing dialog box, receive the first round of input information, make interactive connections in the multi-information network, output the first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context-joint guidance mechanism, and perform optimization and adjustment in combination with the minimum difference threshold and the maximum difference threshold.

2. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 1 is characterized in that: Using a machine learning model, M vertical pre-trained ancient book information processing models are determined, and the method includes: Performing fine-grained semantic analysis on the scanned text and adding fine-grained annotation information, wherein the fine-grained annotation information includes a core word and a first internal logical relationship with a core topic, and a core phrase and a second internal logical relationship with the core topic; Use machine learning models for topic modeling, and continue to use M domain databases for pre-training to ensure that the pre-training data has fine-grained annotation information to support the training of the semantic parsing layer.

3. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 2 is characterized in that: Performing fine-grained semantic analysis on the scanned text and adding fine-grained annotation information, the method further includes: Use NLP technology to define ancient Chinese linguistics rule sets and modern Chinese linguistics rule sets; In the ancient book information processing dialog box, a first task instruction and a first text to be processed are generated by using the ancient Chinese linguistics rule set and the modern Chinese linguistics rule set, wherein the first task instruction activates the NLP dynamic response mechanism; By utilizing the NLP dynamic response mechanism and comparing the text processing requirements corresponding to the first task instruction, any one of the core vocabulary extraction mode and the core phrase extraction mode is selectively triggered.

4. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 3 is characterized in that: The method further comprises: With a multi-level attention mechanism, a long short-term memory network is used to add a gated recurrent unit to enhance the recognition of multiple text sequences of different lengths and types; In the ancient book information processing dialog box, a second task instruction and a second text to be processed are generated through a gated loop unit, wherein the second task instruction is used to activate a grammar analysis mechanism; By utilizing the grammatical analysis mechanism and comparing with the text processing requirement corresponding to the second task instruction, any one of the part-of-speech tag marking mode and the punctuation tag marking mode is selectively triggered.

5. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 4 is characterized in that: Using a context joint guidance mechanism, responding to the second round of input information, the third round of input information, ..., the Nth round of input information, and optimizing and adjusting in combination with the minimum difference threshold and the maximum difference threshold, the method includes: Set the minimum difference threshold and the maximum difference threshold, and determine the first round of output information through the first round of input information; Determine the second round output information by using the first round input information, the first round output information, and the second round input information and the context joint guidance mechanism; Determine the third round output information by using the first round input information and the first round output information, the second round input information and the second round output information, and the third round input information by using the context joint guidance mechanism; The context-joint guidance mechanism is repeated N-1 times, and difference evaluation is performed at the same time. When the current cumulative difference index meets the minimum difference threshold but does not meet the maximum difference threshold, the M vertical pre-trained ancient book information processing models are synchronously enhanced with the highest similarity with the pre-training stage as the learning direction.

6. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 5 is characterized in that: While performing a difference assessment, the method further comprises: Configure the single-round difference measurement formula: Where, ΔT round The difference index used to characterize a round of dialogue, t i,input It is used to represent the weight of the i-th topic in the input information, p i,output Used to represent the weight of the i-th topic in the output information; Configure the multi-round difference measurement formula: ΔT total It is used to characterize the current cumulative difference index corresponding to multiple rounds of dialogue, where R is the number of rounds.

7. The ancient book information intelligent processing method based on artificial intelligence as claimed in claim 6 is characterized in that: The method comprises: Connect to the user terminal device, obtain the user location information, and match it with the location tag. After the match is successful, establish an interactive interface using Vue.js technology, and the interactive interface includes an output reading plug-in; A request is made to access the audio output permission of the user terminal device. After the user accepts, the output reading plug-in provides an audio guide service.

8. An intelligent processing device for ancient book information based on artificial intelligence, characterized in that: The device for implementing the ancient book information intelligent processing method based on artificial intelligence according to any one of claims 1 to 7 comprises: The ancient book scanning module is used to scan ancient books and obtain scanned text, which is in an editable text format; A knowledge graph building module, used to build an ancient book knowledge graph based on the scanned text, wherein the ancient book knowledge graph sets triples based on text punctuation as constraints, and the triples are used for text editing based on segmentation and part of speech; A database classification module is used to classify M domain databases through the ancient book knowledge graph, and use a machine learning model to determine M vertical pre-trained ancient book information processing models. The classification indicators corresponding to the M domain databases include age, core topics, and representative ideas of the original works; A historical carrier introduction module is used to introduce various forms of historical carriers, including cultural relics, relics, and intangible cultural heritage; The multimodal perception module is used to determine a first multimodal perception set based on the multi-form historical carriers, the era and the core topics in the classification indicators; determine a second multimodal perception set based on the multi-form historical carriers, the original ideas and the core topics in the classification indicators; determine a third multimodal perception set based on the multi-form historical carriers, the era and the original ideas in the classification indicators; A multivariate information network establishment module, configured based on the first multimodal perception set, the second multimodal perception set, and the third multimodal perception set, using a streaming generation mode, configuring a multi-layer branch guidance structure, and embedding the M vertical pre-trained ancient book information processing models into the multi-layer branch guidance structure to establish a multivariate information network; The optimization and adjustment module is used to establish an ancient book information processing dialog box, receive the first round of input information, perform interactive connection in the multi-information network, output the first round of response information, and respond to the second round of input information, the third round of input information, ..., the Nth round of input information with a context joint guidance mechanism, and perform optimization and adjustment in combination with the minimum difference threshold and the maximum difference threshold.

Images