An AI-based professional content intelligent generation method

Abstract

The application discloses an AI-based professional content intelligent generation method, relates to the technical field of professional content generation, and comprises the following steps: designing the structure of a knowledge base, collecting text data from a target data source, and preprocessing the text data; generating semantic representation of the text data based on a BERT model and updating the knowledge base; constructing a semantic generation formula based on the preprocessed text data and the BERT model, and obtaining a semantic correlation score. The dynamic updating and semantic compression of the knowledge base ensure that the generated content is synchronized with the latest field knowledge, the BERT model enhances the capture of deep-level semantics of the text, the semantic correlation score ensures the high correlation and logical consistency of the generated content, and the feedback mechanism further optimizes the logical order and semantic density of the generated content, so that the final output content meets the user demand and has high semantic density and coherence.

Description

Technical Field

[0001] This invention relates to the field of professional content generation technology, and in particular to an intelligent method for generating professional content based on AI writing. Background Technology

[0002] With the rapid development of artificial intelligence and natural language processing technologies, machine learning-based text generation technology has become an important tool in many fields. In particular, language generation systems implemented through deep learning models, such as BERT, can capture complex contextual semantic information, driving innovation in automated text generation technology. Traditional text generation technologies rely on rule-based or template-based methods, which often cannot handle professional content with complex semantic relationships, especially in fields such as law, scientific and technological literature, and medical reports, where the limitations of this technology are particularly obvious. In recent years, language model technologies based on deep learning, such as GPT and BERT, have been widely used in document writing, intelligent question answering, and automatic content generation, possessing cross-domain language understanding and generation capabilities.

[0003] Nevertheless, existing text generation technologies still face several technical bottlenecks that urgently need to be addressed. Traditional text generation models often struggle to dynamically update and expand their knowledge bases, and the accuracy of the generated content often depends on pre-defined knowledge base content. This results in the generated text lacking a connection to the latest domain knowledge. Existing generation models lack efficient feedback mechanisms when dealing with complex semantic logic, and the generated content may contain logical errors, semantic jumps, and other problems. In particular, when dealing with complex content involving multiple domains and interdisciplinary fields, text generation systems struggle to strike a balance between logical coherence and semantic accuracy. The large amount of data generated during the generation process lacks effective management and optimization, making it difficult to ensure that the generated content remains synchronized with dynamically changing domain knowledge. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned intelligent generation methods of professional content based on AI, this invention is proposed.

[0005] Therefore, the problem that this invention aims to solve is that existing AI-based text generation technologies mainly suffer from the following problems: lagging knowledge base updates, resulting in generated content that cannot reflect the latest professional knowledge in a timely manner; semantic analysis models are prone to errors when processing complex logic and long texts; generated content may contain logical contradictions or inconsistencies; and there is a lack of effective feedback and optimization mechanisms for the generated content, resulting in a lack of room for further improvement in the generated text.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent method for generating professional content based on AI, comprising: designing the structure of a knowledge base and collecting text data from a target data source for preprocessing; generating semantic representations of the text data based on a BERT model and updating the knowledge base; constructing a semantic generation formula based on the preprocessed text data and the BERT model to obtain a semantic relevance score; generating preliminary content based on the semantic relevance score and optimizing the logical order and semantic density by adjusting the semantic relevance score through feedback; and storing all data generated during the professional content generation process.

[0007] As a preferred embodiment of the AI-based intelligent generation method for professional content described in this invention, the structure of the knowledge base is designed by dividing it into a basic knowledge layer, a professional knowledge layer, and a case study layer. The basic knowledge layer stores common technical terms, knowledge definitions, and basic theories extracted from publicly available domain standards, technical manuals, and public corpora. The professional knowledge layer extracts professional knowledge from patent documents, technical reports, and academic papers in the target domain, and uses a semantic analysis-based clustering algorithm to cluster similar professional knowledge to form domain-specific semantic sets. The case study layer focuses on technical application cases and solutions in specific domains, and the data in this layer comes from verified successful and unsuccessful cases.

[0008] As a preferred embodiment of the AI-based intelligent generation method for professional content, as described in this invention, the preprocessing of text data collected from the target data source refers to the batch collection of text data at various levels through automated web crawlers; the collection of text data is decomposed into independent words using Natural Language Processing (NLP) technology, meaningless words and special symbols are removed, and different forms of words are normalized to basic forms by extracting stems and low-frequency words are filtered out.

[0009] As a preferred embodiment of the AI-based intelligent content generation method described in this invention, the step of generating semantic representations of text data and updating the knowledge base based on the BERT model refers to using a pre-trained embedding vector model BERT to generate semantic representations of text data and store them as embedding vectors. New knowledge is automatically added to the knowledge base through real-time semantic analysis, while old knowledge is gradually phased out; semantic compression of the content in the knowledge base is performed regularly; and the unsupervised clustering algorithm HDBSCAN is used regularly to reclassify the text data in the knowledge base.

[0010] As a preferred embodiment of the AI-based intelligent content generation method described in this invention, the method involves: constructing a semantic generation formula based on preprocessed text data and combining it with a BERT model to derive a semantic relevance score; and using a pre-trained embedding vector model BERT to convert user-inputted text data into high-dimensional vectors to obtain the text data embedding vectors. Semantic relevance is measured by calculating the cosine similarity between the input text data and the knowledge base topics; variational inference techniques are used to analyze the logical changes in the input text data and dynamically generate extended explanations that conform to the context; and normal distribution is used. Remove irrelevant noise information, among which, The mean of the noise. Let be the standard deviation of the noise; construct a complete semantic generation formula to calculate the semantic relevance score R(t) between the input text data and the topics in the knowledge base at time t.

[0011] As a preferred embodiment of the AI-based intelligent content generation method described in this invention, the following steps are included: generating preliminary content based on semantic relevance score refers to automatically selecting the context information of the most relevant topic from the knowledge base based on the calculated semantic relevance score; using hyponyms, synonyms, and related words to match the keywords of the current generated content to generate preliminary content; adjusting the level of detail of the generated content according to the semantic relevance score; and marking and correcting any inconsistencies after the preliminary content is generated.

[0012] As a preferred embodiment of the AI-based intelligent content generation method described in this invention, the following steps are included: Optimizing the logical order and semantic density by adjusting the semantic relevance score through feedback: After initial content generation, inserting a user feedback option allows users to evaluate the generated content, with evaluation indicators including "satisfied" and "unsatisfied"; collecting user evaluation indicators and adjusting the original semantic relevance score to obtain a new semantic relevance score; globally optimizing the generated content for the entire topic based on the new semantic relevance score; setting a scoring threshold based on the final semantic relevance score; maintaining normal operation if the final semantic relevance score is higher than the threshold, and optimizing the logical order and semantic density of the generated content if the final semantic relevance score is lower than the threshold; using a dependency parsing tree to extract the logical relationships between sentences in the topic and forming a logical dependency tree, where each node represents a sentence; judging the coherence of sentences based on the tree structure, marking places with logical jumps, and reordering sentences; evaluating the semantic density of each paragraph by calculating the proportion of information words in the sentences, comparing the semantic density of each paragraph, identifying paragraphs with low semantic density, and introducing additional descriptive information and deleting redundant descriptions for paragraphs with low semantic density.

[0013] As a preferred embodiment of the AI-based intelligent generation method for professional content, as described in this invention, the method of storing all data generated during the professional content generation process refers to storing all data generated during the generation process using a structured storage format, establishing an index for the content and related data generated in each round, establishing a data backup mechanism, regularly backing up the stored data, setting user permissions, and allowing administrators to access the stored data.

[0014] A computer device includes: a memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the above-described method.

[0015] A computer-readable storage medium having a computer program stored thereon, characterized in that: when the computer program is executed by a processor, it implements the steps of the above-described method.

[0016] The beneficial effects of this invention are as follows: dynamic updates of the knowledge base and semantic compression ensure that the generated content is synchronized with the latest domain knowledge; the BERT model enhances the capture of deep semantics of the text; semantic relevance scoring ensures high relevance and logical consistency of the generated content; and the feedback mechanism further optimizes the logical order and semantic density of the generated content, ensuring that the final output content not only meets user needs but also has high semantic density and coherence. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the intelligent generation method for AI-based professional content.

[0019] Figure 2 This is a schematic diagram of the intelligent content generation process based on semantic relevance scoring. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0023] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a method for intelligently generating professional content based on AI writing. The method for intelligently generating professional content based on AI writing includes the following steps:

[0024] S1. Design the structure of the knowledge base and collect text data from the target data source for preprocessing; specifically, designing the structure of the knowledge base means dividing the structure of the knowledge base into a basic knowledge layer, a professional knowledge layer, and a case layer.

[0025] The foundational knowledge layer stores common technical terms, knowledge definitions, and basic theories extracted from publicly available domain standards, technical manuals, and public corpora (such as WordNet and DBpedia) as themes.

[0026] The professional knowledge layer extracts professional knowledge as themes by collecting patent documents, technical reports and academic papers in the target field. For the collected text data, the BERT model is first used to convert the text content into high-dimensional semantic embedding vectors to retain its semantic information. Then, a density clustering algorithm based on semantic embedding vector distance (such as HDBSCAN) is used to measure the semantic proximity between professional knowledge texts according to the Euclidean distance or cosine similarity between vectors. The HDBSCAN algorithm can automatically identify high-density semantic clusters, classify highly similar professional knowledge into the same semantic cluster, and automatically remove low-density or outlier points, thereby forming a set of domain-specific professional knowledge semantics with high semantic consistency.

[0027] The target areas include engineering and manufacturing, computer and information technology, life sciences and medicine, environment and energy, and chemistry and materials science;

[0028] The case study layer focuses on technical application cases and solutions in specific fields. The data in this layer comes from verified success and failure cases, and the focus is on providing reference standards and logical frameworks.

[0029] Specific areas include information technology and software development, engineering design and manufacturing, biomedical and medical devices, energy and environmental engineering, and financial and economic analysis.

[0030] By dividing the knowledge base into a basic knowledge layer, a professional knowledge layer, and a case study layer, this structure achieves comprehensive coverage and effective organization of knowledge across a wide range of fields. The basic knowledge layer provides general technical terms and fundamental theories to ensure that the generated content has a solid knowledge foundation. The professional knowledge layer categorizes patents and academic resources in the target field through semantic clustering to form a set of professional knowledge in a specific field, ensuring the depth and accuracy of the generated content. The case study layer provides practical experience and logical support for the generated content by storing successful and unsuccessful technical application cases, enhancing the practicality and guiding value of the content. This structure not only improves the scalability of the knowledge base but also achieves dynamic updates and accurate retrieval capabilities through domain-specific data and clustering algorithms.

[0031] Furthermore, preprocessing text data collected from the target data source refers to using automated web crawlers to collect text data at various levels in batches. Automated web crawlers are a technology used to automatically extract data from the Internet. Natural Language Processing (NLP) technology is used to decompose the collected text data into independent words, remove meaningless words and special symbols, and normalize different forms of words into basic forms by extracting stems and filtering low-frequency words. NLP technology is used to analyze the text data collected by automated web crawlers, automatically extract terms and domain key points, and convert the text into a semantic representation.

[0032] By using automated web crawlers to collect text data in batches and combining it with natural language processing (NLP) technology for data preprocessing, this solution effectively improves the efficiency of collecting and processing large-scale text data. Automated web crawling technology ensures that text information at various levels can be quickly extracted from the Internet, avoiding the inefficiency of manual collection. The introduction of NLP technology, through steps such as word segmentation, removal of invalid information, stemming, and filtering of low-frequency words, achieves standardized processing of text data, ensuring the accuracy and consistency of subsequent semantic analysis. NLP technology can also automatically extract terms and domain key points and convert text into an efficient semantic representation, providing clear and standardized input data for downstream semantic analysis and generation tasks, thus optimizing the accuracy and efficiency of the entire data processing chain.

[0033] S2. Generate semantic representations of text data based on the BERT model and update the knowledge base;

[0034] Specifically, generating semantic representations of text data and updating the knowledge base based on the BERT model refers to using the pre-trained BERT embedding vector model to generate semantic representations of text data and storing them as embedding vectors. BERT is a deep learning model based on the Transformer architecture, used to process local semantic relationships, and is good at capturing the context and local similarity of words;

[0035] Through real-time semantic analysis, new knowledge is automatically added to the knowledge base, while old knowledge is gradually phased out.

[0036] Regularly perform semantic compression on the content in the knowledge base. Semantic compression includes calculating the redundancy between different knowledge points and removing duplicate information.

[0037] The unsupervised clustering algorithm HDBSCAN is used periodically to reclassify the text data in the knowledge base to ensure semantic clustering accuracy. The algorithm can dynamically adapt to changes and expansions in domain knowledge through periodic reconstruction.

[0038] By generating semantic representations of text data based on the BERT model and updating the knowledge base in real time, this method can ensure dynamic adaptation to semantic changes, automatically adding new knowledge and discarding old knowledge, maintaining the real-time accuracy of the knowledge base. Regular semantic compression reduces redundant data and optimizes storage and retrieval efficiency. At the same time, the unsupervised clustering algorithm HDBSCAN is used to reclassify the text in the knowledge base, ensuring the accuracy and consistency of semantic clustering. This method can not only effectively capture local semantic relationships in the context, but also dynamically adapt to the expansion and changes of domain knowledge by continuously reconstructing the knowledge base, thereby improving the intelligence and adaptability of the system.

[0039] S3. Based on the preprocessed text data, construct a semantic generation formula using the BERT model to obtain a semantic relevance score;

[0040] Specifically, based on the preprocessed text data, a semantic generation formula is constructed using the BERT model to derive the semantic relevance score. This formula uses the pre-trained embedding vector model BERT to convert the user-input text data into a high-dimensional vector, obtaining the text data's embedding vector. The formula is as follows:

[0041] ;

[0042] in, This represents the embedding vector of the input text data, and this represents the semantic representation vector after processing by the BERT model. This represents the embedding vector generated by the BERT model, indicating that the BERT model receives input text data and returns a vector representation of that text data.

[0043] Semantic relevance is measured by calculating the cosine similarity between the input text data and the topics in the knowledge base. Cosine similarity quantifies the angle between the two elements in a multidimensional space; the smaller the angle, the higher the semantic similarity. The formula is:

[0044] ;

[0045] in, This represents the embedding vector of the k-th topic in the knowledge base structure. and They are respectively and In the model, it means and The Euclidean norm in a multidimensional vector space is obtained by summing the squares of the vector components, taking the square root, and then calculating the Euclidean norm. represent and The angle between them is used to measure the semantic similarity between two vectors. The cosine of the angle between two vectors is used to measure their similarity in direction.

[0046] when hour, =1 indicates that the two vectors are completely identical and semantically similar.

[0047] when hour, =0 indicates that the two vectors are orthogonal to each other and have no semantic correlation;

[0048] when hour, =-1 indicates that the two vectors are completely opposite and semantically contradictory;

[0049] Using embedding vectors in combination with cosine similarity can better capture the implicit semantic connections in text, going beyond surface word similarity to capture the deeper meaning in the context.

[0050] Variational inference techniques are used to analyze the logical changes in the input text data and dynamically generate extended explanations that conform to the context. The formula is as follows:

[0051] ;

[0052] Where L(t) is the logical derivation result at time t, representing a vector of generated content changing over time. This represents the total number of topics in the knowledge base related to the input text data, and signifies the logical derivation of all possible semantic content. This represents the Tanh nonlinear transformation function applied to the time partial derivative results of the logical derivation, indicating the effect of... and The semantic relationships are adjusted non-linearly. It is the partial derivative of the logical derivation function over time, representing the change of the logical derivation with time t, and is used to capture the dynamic changes in semantic logic. The sign of the partial derivative. Let be the integration variable, indicating that the integration is performed with respect to the embedding vector. Performed;

[0053] The expression for the logical derivation function is:

[0054] ;

[0055] in, This represents the target embedding vector, which is the target embedding vector of the content the model expects to generate. This is the inverse of the covariance matrix. The covariance matrix represents the correlation structure between the input embedding vector and the logical derivation parameters, while the inverse matrix is ​​used to measure the distance and correlation between two variables. The difference between the embedded vector and the target embedded vector is represented and expressed in standard form by transpose, where T is the entire time interval;

[0056] The input text data is logically divided into multiple dimensions, including topic, context and sentence structure. The time dimension in the logical deduction process affects the progression of the text data content in each dimension. In order to ensure that the generated content is logically coherent, the logical deduction must not only capture the semantic state at the current moment, but also update and deduce based on the previous content. This progression is manifested in the adjustment of each semantic dimension at different points in time, and the vector form can well represent the changes in these dimensions.

[0057] By using the normal distribution To remove irrelevant noise, the expression is:

[0058] ;

[0059] in, The mean of the noise. The standard deviation of the noise;

[0060] Using the normal distribution function to filter noise is more accurate than simple rule filtering. It can dynamically adjust the noise level according to the context, ensuring that the generated content is not affected by irrelevant information.

[0061] Construct a complete semantic generation formula to calculate the semantic relevance score R(t) between the input text data and the topics in the knowledge base at time t. The larger the value, the closer the semantics of the two are. The formula is as follows:

[0062] ;

[0063] in, The semantic weight coefficient represents the i-th topic and is used to adjust the importance of different semantic topics. The semantic weight coefficient is set based on the importance of domain knowledge. is the normalization factor for the i-th topic, used to normalize the similarity scores of different topics to ensure that the similarity results are comparable, T is the entire time interval, and dt represents a small time increment.

[0064] By calculating the cosine similarity between the input text and topics in the knowledge base, the semantic relevance is quantified. This accurately measures the semantic angle between the text and the knowledge base topics in a multi-dimensional space, enabling the method to automatically select the most relevant topics for content generation based on similarity. This ensures high relevance and relevance of the generated content and avoids unnecessary content redundancy. The introduction of variational inference technology analyzes the logical changes in the input text data, making the generated content not only semantically coherent but also progressive in the temporal dimension. Through the temporal partial derivatives in the logical derivation process, the system can capture the semantic changes of the text at different times, ensuring that the progressive generation of content conforms to the context requirements and enhancing the logical rigor and coherence of content generation. Using a normal distribution function to remove noise information effectively avoids the errors caused by simple rule filtering, making content generation more accurate. Dynamically adjusting the noise level ensures that the generated content is not interfered with by irrelevant information, further improving the quality of text generation. By constructing a complete semantic generation formula and combining the semantic weight coefficients and normalization factors of topics, the system can adjust for different topic importance, ensuring the comparability of similarity scores and providing the system with a flexible topic selection mechanism to better meet user needs.

[0065] S4. Generate preliminary content based on semantic relevance scores and adjust the logical order and semantic density by adjusting the semantic relevance scores based on feedback; specifically, generating preliminary content based on semantic relevance scores means automatically selecting the context information of the most relevant topic from the knowledge base based on the calculated semantic relevance scores.

[0066] The system uses hyponyms, synonyms, and related words to match keywords in the current content to generate preliminary content. The level of detail in the generated content is adjusted based on the semantic relevance score. The higher the relevance, the more details are generated, and the lower the relevance, the simpler the generated content.

[0067] After the initial content generation is completed, cosine similarity is used to calculate the semantic consistency between the current generated content and the original context of the referenced knowledge base. For sentences and segments with similarity below the preset threshold (set in experimental tuning), they are marked as potentially inconsistent content and corrected through semantic completion or context rewriting, thereby ensuring that the new content maintains semantic consistency with the current topic context.

[0068] By generating preliminary content based on semantic relevance scores, this method intelligently selects the most relevant topics and contextual information from the knowledge base and dynamically adjusts the level of detail in the generated content. This mechanism ensures that highly relevant content has more detail, while low-relevance content is simplified, thereby improving the accuracy and efficiency of the generated text. By introducing hyponyms, synonyms, and conjunctions, the semantic hierarchy of the generated content is further enriched, ensuring semantic depth and diversity. The cosine similarity correction mechanism ensures that the newly introduced context remains consistent with the current topic, effectively avoiding the problem of semantic incoherence.

[0069] Furthermore, by adjusting the semantic relevance score through feedback to optimize the logical order and semantic density, a user feedback option is inserted after the initial content is generated, allowing users to evaluate the generated initial content. Evaluation indicators include "satisfied" and "unsatisfied". The user evaluation indicators are collected and the original semantic relevance score is adjusted to obtain a new semantic relevance score.

[0070] The generated content for the entire topic is globally optimized based on the new semantic relevance score, using the following formula:

[0071] ;

[0072] in, For the new semantic relevance score, The semantic relevance score of the final generated content is used to evaluate the overall semantic accuracy and user satisfaction of the generated content, and to adjust future generation strategies. dt represents a small time increment.

[0073] A scoring threshold is set based on the final semantic relevance score. If the final semantic relevance score is higher than the scoring threshold, the system continues to operate normally. If the final semantic relevance score is lower than the scoring threshold, the logical order and semantic density of the generated content are optimized.

[0074] Dependency parse trees are used to extract the logical relationships between sentences in a topic and form a logical dependency tree. Each node represents a sentence. The coherence of sentences is judged based on the tree structure, places with logical jumps are marked, and sentences are reordered. Dependency parse trees are a technique used in natural language processing. By identifying the dependencies between words in a sentence, a tree-like structure is constructed to show the grammatical and logical relationships of each word in the sentence.

[0075] The semantic density of each paragraph is assessed by calculating the proportion of information words in the sentence. Information words include nouns, verbs, adjectives, and adverbs. By comparing the semantic density of each paragraph, paragraphs with low semantic density are identified. For paragraphs with low semantic density, additional descriptive information is used and redundant descriptions are removed. The additional descriptive information includes case analysis in the case layer related to the corresponding paragraph and detailed explanations of nouns in the information words involved in the paragraph.

[0076] By adjusting semantic relevance scores based on feedback and optimizing the logical order and semantic density of generated content, the system can dynamically update semantic relevance scores based on user evaluations, making the generated content more in line with user needs. A parse tree is used to adjust the logical coherence between sentences, ensuring a reasonable logical order of the generated content. The semantic density of paragraphs is assessed by calculating the proportion of information words; this method can effectively identify and optimize paragraphs with low semantic density. Additional descriptive information is used to make the content more compact and information-rich. By combining user feedback with automated semantic optimization, the system significantly improves the coherence, information density, and semantic accuracy of the generated content, ensuring that the final output meets high-quality standards.

[0077] S5. Store all data generated during the professional content generation process; specifically, storing all data generated during the professional content generation process means using structured storage to store all data generated during the generation process, creating an index for the content and related data generated in each round, establishing a data backup mechanism, regularly backing up the stored data, setting user permissions, and allowing administrators to access the stored data.

[0078] By structuring, indexing, backing up, and managing permissions for all data generated during the professional content creation process, the system achieves efficient data management and security. The establishment of indexes ensures that relevant data can be quickly retrieved during subsequent content creation, improving response speed. Regular backup and data recovery mechanisms guarantee data security and availability, preventing information loss due to accidental loss. Permission management further controls data access permissions, ensuring that sensitive information is properly protected and giving administrators control over the data, making the entire generation system more flexible, secure, and reliable.

[0079] Example 2: This is the second embodiment of the present invention, which differs from the previous embodiment in that:

[0080] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0082] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0083] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

Claims

1. An AI-based professional content intelligent generation method based on writing, characterized by: Comprising, designing the structure of the knowledge base and collecting text data from target data sources for preprocessing; generating semantic representations of the text data based on the BERT model and updating the knowledge base; constructing a semantic generation formula based on the preprocessed text data and the BERT model to obtain a semantic relevance score; generating preliminary content based on the semantic relevance score and adjusting the semantic relevance score through feedback to optimize the logical order and semantic density; storing all data generated during the generation of professional content; The semantic correlation score is obtained by constructing a semantic generation formula according to the preprocessed text data and combining a BERT model s ; measuring semantic relevance by calculating the cosine similarity between input text data and knowledge base topics; analyzing the logical changes in input text data using variational inference techniques and dynamically generating contextually appropriate extended explanations; By using a normal distribution removing irrelevant noise information, wherein, is the mean of the noise, is the standard deviation of the noise; constructing a complete semantic generation formula to calculate the semantic relevance score R(t) of input text data and topics in the knowledge base at time t; the feedback adjustment of the semantic relevance score to optimize the logical order and semantic density refers to inserting user feedback options after the generation of preliminary content, allowing users to evaluate the generated preliminary content, with evaluation indicators including "satisfied" and "dissatisfied"; collecting user evaluation indicators and adjusting the original semantic relevance score to obtain a new semantic relevance score; globally optimizing the generated content of the entire topic based on the new semantic relevance score; setting a score threshold based on the final semantic relevance score, if the final semantic relevance score is higher than the score threshold, the system remains in normal operation, if the final semantic relevance score is lower than the score threshold, the logical order and semantic density of the generated content are optimized; extracting the logical relationship between sentences in the topic using dependency parsing trees and forming a logical dependency tree, with each node representing a sentence, judging the coherence of the sentences based on the tree structure, marking places with logical jumps and reordering the sentences; evaluating the semantic density of each paragraph by calculating the proportion of information words in the sentence, comparing the semantic density of each paragraph to identify paragraphs with low semantic density, and introducing additional descriptive information and deleting redundant descriptions for paragraphs with low semantic density. 2.The AI-based professional content intelligent generation method of claim 1, wherein: The structure of the knowledge base refers to dividing the structure of the knowledge base into a basic knowledge layer, a professional knowledge layer, and a case layer; the basic knowledge layer stores general technical terms, knowledge definitions, and basic theories for the topic by extracting them from public domain standards, technical manuals, and public corpora; the professional knowledge layer extracts professional knowledge for the topic by collecting patent documents, technical reports, and academic papers in the target field, and uses a clustering algorithm based on semantic analysis to cluster similar professional knowledge to form a domain-specific semantic set; the case layer takes technical application cases and solutions in specific fields as the topic, and this layer of data comes from verified successful cases and failed cases. 3.The AI-based professional content intelligent generation method of claim 2, wherein: The preprocessing of the collected text data refers to batch collecting text data at different levels through automated crawlers; use natural language processing (NLP) techniques to decompose the collected text data into independent words, remove meaningless words and special symbols, and normalize different forms of words to their basic forms by extracting word stems and filtering low-frequency words. 4.The AI-based professional content intelligent generation method of claim 3, wherein: The BERT model-based generating semantic representation of text data and updating knowledge base refers to using a pre-trained embedding vector model BERT to generate semantic representation of text data and store as embedding vector V k ; By real-time semantic analysis, new knowledge is automatically added to the knowledge base, and old knowledge is gradually eliminated; Periodically, the content in the knowledge base is semantically compressed; Periodically, the text data in the knowledge base is reclassified using the unsupervised clustering algorithm HDBSCAN. 5.The AI-based professional content intelligent generation method of claim 4, wherein: The preliminary content generation process according to the semantic correlation score generates a semantic correlation score, and automatically selects the context information of the highest correlation topic from the knowledge base; The keywords of the generated content are matched with the keywords of the generated content using the upper and lower words, synonyms and related words to generate the preliminary content, and the generated preliminary content adjusts the detail level of the generated content according to the high and low of the semantic correlation score; After the preliminary content generation is completed, the inconsistent parts are marked and corrected. 6.The AI-based professional content intelligent generation method of claim 5, wherein: The storage of all data generated in the professional content generation process refers to the use of structured storage to store all data generated in the generation process, index the content generated in each round and related data, establish a data backup mechanism, regularly back up the stored data, set user permissions, and allow managers to call stored data.

7. A computer device comprising: Memory and processor; The memory stores a computer program, and the processor executes the computer program to realize the steps of the AI-based professional content intelligent generation method according to any one of claims 1-6.

8. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the AI-based professional content intelligent generation method according to any one of claims 1-6.

Images