An artificial intelligence-based text outline automatic generation method and system

Abstract

The present application relates to the technical field of data processing, and particularly relates to a text outline automatic generation method and system based on artificial intelligence, which comprises the following steps: calculating the weight of a temperature parameter in a PEGASUS model decoding stage, training the PEGASUS model to obtain a text outline generation model, and processing a document to be processed by using the text outline generation model to generate a text outline; the weight of the temperature parameter is calculated by the following method: converting preprocessed original corpus into a high-dimensional word vector sequence; extracting a hierarchical tree of a current generated candidate outline, calculating a hierarchical aggregation coefficient of each hierarchical node in the hierarchical tree; calculating a semantic fidelity index of the hierarchical node; calculating an iterative stability index based on the semantic fidelity index; and taking the normalized iterative stability index as the weight of the temperature parameter. The present application solves the problem of logical confusion of a traditional generative model when processing long professional documents.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for automatically generating text outlines based on artificial intelligence. Background Technology

[0002] With the explosive growth of digital information, the rapid extraction of core logic and generation of structured outlines from massive amounts of long texts has become an urgent need in scientific research, content creation, and academic research. The development of artificial intelligence technology, especially natural language processing, has made the automated generation of text outlines possible. Outline generation not only requires the system to accurately capture the central idea of ​​the text, but also demands that the generated content possess high logical coherence, hierarchy, and coverage of key information, thereby helping users grasp the article's structure in a short time. Among existing technologies, the PEGASUS model is a representative model in the field of generative text summarization. This algorithm, through a large-scale pre-training task involving the generation of missing sentences, demonstrates significant advantages in handling text structure and semantic extraction, enabling the generation of semantically coherent and highly generalized text fragments.

[0003] However, due to the unique characteristics of text outline generation tasks—strong logical hierarchy and a high concentration of domain knowledge—directly using the original PEGASUS model can lead to technical problems when dealing with long, professional documents, such as disordered logical hierarchy and an inability to accurately distinguish between primary and secondary points. This is because the original algorithm's loss function primarily focuses on overall semantic similarity, lacking a detailed characterization of structural features such as document chapter connections and indentation levels. Consequently, the generated outlines are difficult to directly use to guide actual document reconstruction or in-depth reading. Summary of the Invention

[0004] To address the technical problem that the outlines generated using the PEGASUS model suffer from disorganized logical hierarchy and an inability to accurately distinguish between primary and secondary points when dealing with long professional documents, this invention provides an artificial intelligence-based method and system for automatically generating text outlines.

[0005] In a first aspect, the present invention provides a method for automatically generating text outlines based on artificial intelligence, employing the following technical solution:

[0006] An AI-based method for automatically generating text outlines includes the following steps:

[0007] The weights of the temperature parameters in the decoding stage of the PEGASUS model are calculated, and the PEGASUS model is trained to obtain a text outline generation model. The text outline generation model is then used to process the document to be processed and generate a text outline.

[0008] The calculation method for the weight of the temperature parameter is as follows: The original corpus is acquired and preprocessed, transforming it into a high-dimensional word vector sequence; the hierarchical tree of the current candidate outline is extracted, and the hierarchical aggregation coefficient of each node in the hierarchical tree is calculated. The hierarchical aggregation coefficient is negatively correlated with the vector standard deviation of the set of sibling nodes to which the corresponding node belongs; the semantic fidelity index of the hierarchical nodes is calculated, and the semantic fidelity index is negatively correlated with the hierarchical aggregation coefficient; an iterative stability index is calculated based on the semantic fidelity index, and the iterative stability index is positively correlated with the semantic fidelity index; the iterative stability index is normalized, and the normalized iterative stability index is used as the weight of the temperature parameter.

[0009] By introducing hierarchical aggregation coefficients, semantic fidelity indices, and iterative stability indices to dynamically adjust the temperature parameter weights in the decoding stage of the PEGASUS model, the problem of logical hierarchy confusion and unclear distinction between primary and secondary points that easily occurs in existing PEGASUS models when directly processing long professional documents is effectively solved. This improves the logical coherence, hierarchical structure rationality, and ability to retain core semantics of the generated outline, ensuring that the generated outline not only conforms to the macro-context of the document but also accurately covers micro-details.

[0010] Preferably, the hierarchical aggregation coefficient is calculated as follows: the cosine similarity between the hierarchical node vector and its direct parent hierarchical node vector is calculated, and the ratio of the cosine similarity to the standard deviation of the vector of the same level node set to which the corresponding hierarchical node belongs is used as the hierarchical aggregation coefficient.

[0011] By constructing hierarchical aggregation coefficients by calculating the semantic similarity between hierarchical nodes and their parent hierarchical nodes, as well as the degree of dispersion among sibling nodes, the structural rationality of the generated outline can be quantitatively evaluated. This effectively prevents logical misalignment problems that may occur during the outline generation process, ensures that sub-items closely adhere to the parent topic and that there are appropriate semantic differences between sibling sub-items to reduce redundancy, thereby improving the structured quality of the outline.

[0012] Preferably, before calculating the semantic fidelity index of hierarchical nodes, the method further includes: using an encoder to vectorize the core semantics of the whole text, calculating the modulus of each vectorized core semantic, and segmenting the vectorized core semantics to obtain a sequence of topic keywords.

[0013] By using an encoder to vectorize the core semantics of the entire text and extracting thematic keyword sequences based on the module length, the central idea and key information of the document can be accurately captured. This provides a reliable benchmark for subsequent semantic fidelity calculations, prevents the model from deviating from the main idea of ​​the document when generating deep outlines, and enhances the coverage of the generated outlines with the core content of the original text.

[0014] Preferably, the Otsu threshold segmentation method is used to segment the core semantics, and the sequence of core semantics that is greater than the segmentation threshold is taken as the subject keyword sequence of the document.

[0015] The Otsu threshold segmentation method is used to adaptively determine the segmentation threshold of the core semantics. Compared with fixed threshold truncation, this method can more objectively and dynamically select the most representative topic keywords, avoiding keyword omission or noise introduction caused by manually setting thresholds, and further improving the accuracy and robustness of key information extraction.

[0016] Preferably, the semantic fidelity index of hierarchical nodes is calculated as follows: the mutual information matching score between the outline item of the hierarchical node and the topic keyword sequence is calculated; the product of the logical depth of the hierarchical node and the hierarchical aggregation coefficient is calculated, and the product is used as the exponent of the exponential function; the ratio of the mutual information matching score to the exponential function is used as the semantic fidelity index.

[0017] By combining mutual information matching scores with hierarchical depth to construct a semantic fidelity index, it is possible to effectively monitor and suppress semantic drift in the deep outline generation process. This solves the problem that existing algorithms are prone to going off-topic or generating irrelevant details when generating multiple levels, ensuring that the outline remains highly consistent with the core theme of the document while delving into details, thus achieving the unity of global logic and local semantics.

[0018] Preferably, the calculation method for the iterative stability index is as follows: the sequence of semantic fidelity indices generated in the most recent multiple iterations is denoted as the outline index sequence, and the expression for the iterative stability index is:

[0019]

[0020] in, Indicates the first Iterative stability metrics for each iteration; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; The number of data points in the outline indicator sequence; For summation and counting variables; These are preset parameters.

[0021] By calculating iterative stability indices to quantify the changes in semantic fidelity during the generation process, the dynamic stability of the generation system can be captured in real time. This avoids drastic fluctuations in outline quality caused by parameter randomness, provides accurate quantitative feedback signals for dynamic tuning of model parameters, and ensures that the final output outline is a fully converged and stable result.

[0022] Preferably, the method for generating a text outline from a document using a text outline generation model includes: generating a first-level outline from the document using the text outline generation model; using the generated first-level outline as context; and generating deeper sub-items by combining temperature parameters adjusted by an iterative stability index; calculating the iterative stability index in real time after each iteration; stopping the iteration when the convergence requirement is met; and outputting the final text outline with clear hierarchy and semantic coherence.

[0023] By using iterative stability indicators to adjust the generated temperature parameters in real time, randomness is automatically suppressed when the generation process is unstable, and appropriate creativity is allowed when it is stable. This dynamic feedback mechanism ensures that the generated deep sub-items are both rich in detail and do not get out of control. Through multiple rounds of iterative refinement, a high-quality text outline with both clear hierarchical structure and semantic coherence is finally output.

[0024] Preferred methods for preprocessing the original corpus include: using regular expression matching to remove webpage noise; using a segmentation function to structurally split the long text, identifying and retaining the original chapter numbers and hierarchical headings; and calling NLP word segmentation tools to perform Chinese word segmentation, part-of-speech tagging, and stop word filtering.

[0025] Preferably, the method for obtaining the original corpus is as follows: collect the long text data to be processed through the enterprise's internal knowledge base API, and obtain the original corpus through keyword retrieval and URL depth traversal.

[0026] By combining the enterprise's internal knowledge base API with keyword retrieval and deep traversal technology, efficient and automated collection of long text data is achieved. It can quickly build a raw corpus containing rich format information, meeting the data input requirements of large-scale document outline generation tasks.

[0027] Secondly, this invention provides an artificial intelligence-based automatic text outline generation system, employing the following technical solution:

[0028] An AI-based automatic text outline generation system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the system implements the aforementioned AI-based automatic text outline generation method.

[0029] The aforementioned method for automatically generating text outlines based on artificial intelligence is used to generate a computer program, which is then stored in a memory for loading and execution by a processor. This allows for the creation of a system based on the memory and processor, making it convenient to use.

[0030] The present invention has the following technical effects:

[0031] By constructing hierarchical aggregation coefficients, semantic fidelity indicators, and iterative stability indicators, the structural logic, core semantic retention, and stability of the generation process of the outline are quantitatively evaluated. The iterative stability indicators are used to adaptively adjust the temperature parameters in the model decoding stage, which effectively solves the problems of logical confusion, semantic drift, and redundancy of details that occur in traditional generative models when processing long professional documents. This achieves high-quality, hierarchical, and semantically coherent automated outline generation. Attached Figure Description

[0032] Figure 1 This is a flowchart of an artificial intelligence-based method for automatically generating text outlines according to the present invention.

[0033] Figure 2 This is a schematic diagram comparing the semantic fidelity of the present invention with the hierarchical depth.

[0034] Figure 3 This is a schematic diagram comparing the convergence stability of the iterative process of this invention.

[0035] Figure 4 This is a schematic diagram illustrating the comprehensive evaluation of the quality of the outline generated by this invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] This invention discloses an artificial intelligence-based method for automatically generating text outlines, referring to... Figure 1 This includes the following steps:

[0038] S1: Data Acquisition and Preprocessing.

[0039] Long text data to be processed was collected through the enterprise's internal knowledge base API. The collection equipment included a cloud server cluster with high-performance processing capabilities and a distributed storage array. During the collection process, Python's Requests library and Scrapy framework were used as software tools to obtain raw text data containing HTML tags or in PDF format through keyword search and deep URL traversal.

[0040] The original corpus undergoes refined preprocessing. First, regular expression matching is used to remove webpage noise, such as advertisements, illegal characters, and redundant blank lines. Then, a segmentation function is used to structurally split the long text, identifying and retaining the original chapter numbers and hierarchical headings. Next, NLP word segmentation tools are called to perform Chinese word segmentation, part-of-speech tagging, and stop word filtering. Finally, the BERT encoder is used to transform the preprocessed original corpus into a high-dimensional word vector sequence.

[0041] S2: Construct hierarchical aggregation coefficients.

[0042] In the process of generating text outlines, due to the tree-like topology of documents, semantic overlap and cross-level logical dependencies exist. According to the information increment theory, outline items at the same level should maintain a moderate degree of dispersion in the semantic space to reduce redundancy, while sub-items and parent items should exhibit a high degree of implicit semantic relevance. This characteristic leads to a clear clustered distribution and centripetal force shift in the data in the vector space. Directly using the original algorithm often ignores these inter-level constraints, resulting in semantically correct outline items but misaligned hierarchical affiliations. Therefore, it is necessary to construct hierarchical aggregation coefficients. To reflect the first The hierarchical logical attributes of each generated node.

[0043] First, extract the hierarchical tree of the currently generated candidate outlines, vectorize each level node, and calculate the cosine similarity between each level node vector and its direct parent level node vector. The expression for the hierarchical aggregation coefficient is as follows:

[0044]

[0045] in, Indicates the first The hierarchical aggregation coefficient of each level node; Indicates the first Semantic feature vectors of each level node outline item; Indicates the first The semantic feature vector of the outline item of the direct parent node of each level node; This represents the vector dot product operation, used to calculate semantic relevance. Indicates the first The standard deviation of the vector of the set of sibling nodes to which a node belongs at each level is used to measure the semantic dispersion of sibling nodes. These are preset parameters, using fixed reference values. This is to avoid a denominator of zero. and It is obtained through a pre-trained deep learning model.

[0046] The cosine similarity between the semantic vector of a current-level node and the semantic vector of its direct parent-level node is used to characterize the degree of inheritance of the parent node's theme by the child node. The standard deviation of the semantic vectors of the set of sibling nodes is used to measure the level of semantic dispersion among different nodes within the same level. When the semantic similarity of sibling nodes is high, their vector standard deviation is small, indicating that there is a risk of information redundancy at that level.

[0047] Therefore, the hierarchical aggregation coefficient, constructed by normalizing and combining semantic relevance and peer dispersion, can comprehensively reflect the degree of clustering of the current node in the hierarchical structure. When the hierarchical aggregation coefficient is large, it indicates that the node has a high degree of semantic overlap with its parent node and low distinguishability from its peer nodes, posing a risk of ambiguous hierarchical affiliation or structural redundancy. When the hierarchical aggregation coefficient is small, it indicates that the node at this level maintains reasonable distinguishability while inheriting the topic from its superior, resulting in a more reasonable hierarchical structure.

[0048] S3: Construct semantic fidelity metrics.

[0049] Building upon hierarchical features, semantic drift occurs due to the multiple iterations involved in outline generation. This means that as the hierarchy deepens, sub-items tend to deviate from the core theme. According to the attention distribution dilution principle, the model's focus on initial global features decreases when processing deeper details, leading to irrelevant details appearing at the end of the outline. To reflect this feature, a feature reflecting the hierarchical features needs to be constructed. Semantic fidelity index of the degree of semantic preservation of individual nodes This indicator utilizes hierarchical aggregation coefficients. As a weight gain, it assesses the degree to which the deep outline retains the core semantics.

[0050] Before constructing the semantic fidelity index, the BERT encoder is used to vectorize the core semantics of the entire text. The modulus of each core semantic vector is calculated. The modulus of all core semantics is used as input to the Otsu threshold segmentation method to obtain the segmentation threshold of the core semantic modulus. The sequence of core semantics with a modulus greater than the segmentation threshold is taken as the document's topic keyword sequence. The matching score between the current node and the core sequence is calculated. The expression for the semantic fidelity index is:

[0051]

[0052] in, Indicates the first Semantic fidelity metrics for each level of node; Indicates the first The hierarchical aggregation coefficient of each level node; Indicates the first The mutual information matching score between each level node outline item and the topic keyword sequence, with a value range of [value range missing]. ; Indicates the first The logical depth of each level node (e.g., first-level headings) ); exp() represents an exponential function with the natural constant as the base; the method for calculating the mutual information matching score is existing technology, and the specific calculation steps will not be described here.

[0053] The higher the semantic fidelity metric, the better the outline items can accurately anchor the topic at a deeper level and maintain clear logic. This metric addresses the issue of existing algorithms easily going off-topic during multi-level generation, ensuring the consistency of the outline both globally and locally.

[0054] S4: Construct iterative stability metrics.

[0055] During iterative refinement, gradient oscillations and semantic collapse can occur. Since outline generation is a discrete search process, parameter variations can cause drastic fluctuations in the generated outline across different iterations. According to dynamic equilibrium theory, a high-quality outline generation process should tend towards stability. Therefore, it is necessary to construct a system that reflects the... Iterative stability index of round iteration stability This metric uses semantic fidelity as an indicator. The changes in the data reflect the robustness of the generation system.

[0056] First, the most recent The sequence of semantic fidelity indices generated in each iteration is denoted as the outline index sequence. The semantic fidelity indices are then calculated. The expression for the iterative stability index, which is the absolute value of change at each time step, is:

[0057]

[0058] in, Indicates the first Iterative stability metrics for each iteration; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; For the first The average semantic fidelity index of all nodes at all levels in the round generation result; The number of data points in the outline indicator sequence is determined by a preset reference value. ; This is an accumulation operation within the window; These are preset parameters, using fixed reference values. This is to avoid situations where the denominator is zero.

[0059] As the iterative process gradually stabilizes and the quality continues to improve, the denominator decreases, and the iterative stability index increases. Significantly increased. This indicator can capture abnormal jumps in the generation process in real time, providing accurate feedback for parameter tuning.

[0060] S5: Improve the PEGASUS model and generate a text outline.

[0061] A text outline generation model is obtained by training the PEGASUS model. During the training process, the sigmoid function is used to stabilize the iterative index. Normalization is performed, and the normalized iterative stability index is then used. Temperature parameter during the model decoding stage The weights of the normalized iterative stability index When the temperature falls below the preset stability threshold, it indicates that the current generation process is unstable, and the system automatically reduces the temperature parameter. To suppress randomness, in this embodiment, the stability threshold is set to 0.7.

[0062] The text outline generation model is used to process the document to generate a first-level outline, followed by an iterative refinement loop: the text outline generation model uses the generated outline as context, combined with iterative stability metrics. The adjusted temperature parameters generate deeper subdivisions. The hierarchical aggregation coefficient is calculated in real-time after each generation. Semantic fidelity index and iterative stability index When the convergence requirement is met, the iteration stops and the final text outline with clear hierarchy and semantic coherence is output. The convergence requirement is that the number of iterations reaches a preset number, which is 50 times in this embodiment.

[0063] The technical effects of this invention can also be illustrated in conjunction with the accompanying drawings. Figure 2 This diagram illustrates the comparison between the semantic fidelity of the present invention and existing methods at different outline logical depths. The horizontal axis represents the logical depth level of the outline, and the vertical axis represents the semantic fidelity index value. The dashed curve represents the semantic fidelity trend of the existing method, while the solid curve represents the semantic fidelity trend of the present invention.

[0064] As can be seen from the figure, the semantic fidelity index of existing methods shows a significant downward trend as the logical depth of the outline increases. However, the method of this invention can still maintain a high and stable semantic fidelity level under a multi-level structure. This indicates that by introducing a hierarchical aggregation coefficient and a semantic fidelity constraint mechanism, this invention effectively suppresses the semantic drift phenomenon in the multi-level generation process and improves the consistency between the deep outline content and the core theme.

[0065] Figure 3 This diagram illustrates a comparison of the convergence stability of the method of this invention and existing methods during the iterative generation process. Figure 3 It can be seen that in the initial stage of iteration, the iterative stability index of this invention rises rapidly and tends to stabilize within a few rounds, while existing methods exhibit significant fluctuations during multiple iterations, resulting in slow stability improvement. This indicates that by constructing an iterative stability index based on semantic fidelity changes and introducing it into the temperature parameter adjustment process during the decoding stage, the generative model can adaptively suppress random perturbations during training and inference, thereby improving the convergence speed and robustness of the iterative process.

[0066] Figure 4 This is a radar chart illustrating the comprehensive comparison between the method of this invention and existing methods under multiple evaluation indicators for outline generation quality. The chart compares the overall performance of the method of this invention and existing methods across multiple evaluation dimensions, including hierarchical rigor, semantic fidelity, generation stability, key information coverage, and domain accuracy. Figure 4 As can be seen, the index values ​​of the method of the present invention are significantly higher than those of the existing methods in all evaluation dimensions. In particular, the advantages are more significant in terms of the rationality of hierarchical structure, semantic consistency and generation stability. This shows that the present invention achieves comprehensive optimization of the quality of text outline generation through multi-index joint constraints and dynamic adjustment mechanism, and improves the overall reliability and practical value of the generation results.

[0067] This invention also discloses an artificial intelligence-based automatic text outline generation system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an artificial intelligence-based automatic text outline generation method according to this invention is implemented.

[0068] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0069] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for automatically generating text outlines based on artificial intelligence, characterized in that, Including the following steps: The weights of the temperature parameters in the decoding stage of the PEGASUS model are calculated, and the PEGASUS model is trained to obtain a text outline generation model. The text outline generation model is then used to process the document to be processed and generate a text outline. The calculation method for the weight of the temperature parameter is as follows: The original corpus is acquired and preprocessed, and then converted into a high-dimensional word vector sequence; the hierarchical tree of the current candidate outline is extracted, and the hierarchical aggregation coefficient of each level node in the hierarchical tree is calculated. The hierarchical aggregation coefficient is negatively correlated with the vector standard deviation of the set of sibling nodes to which the corresponding level node belongs; the semantic fidelity index of the level nodes is calculated; the iterative stability index is calculated based on the semantic fidelity index; the iterative stability index is normalized, and the normalized iterative stability index is used as the weight of the temperature parameter. The semantic fidelity index of hierarchical nodes is calculated as follows: calculate the mutual information matching score between the outline item of the hierarchical node and the topic keyword sequence; calculate the product of the logical depth of the hierarchical node and the hierarchical aggregation coefficient, and use the product as the exponent of the exponential function; use the ratio of the mutual information matching score to the exponential function as the semantic fidelity index. The calculation method for the iterative stability index is as follows: the sequence of semantic fidelity indices generated in the most recent iterations is denoted as the outline index sequence, and the expression for the iterative stability index is: , Indicates the first The iterative stability index of round iteration, For the first The mean semantic fidelity index of all hierarchical nodes in the round generation result. For the first The mean semantic fidelity index of all hierarchical nodes in the round generation result. For the first The mean semantic fidelity index of all hierarchical nodes in the round generation result. The number of data points in the outline indicator sequence. For summation and counting variables, These are preset parameters.

2. The method for automatically generating text outlines based on artificial intelligence according to claim 1, characterized in that, The hierarchical aggregation coefficient is calculated as follows: the cosine similarity between the hierarchical node vector and its direct parent hierarchical node vector is calculated, and the ratio of the cosine similarity to the standard deviation of the vector of the same level node set to which the corresponding hierarchical node belongs is used as the hierarchical aggregation coefficient.

3. The method for automatically generating text outlines based on artificial intelligence according to claim 1, characterized in that, Before calculating the semantic fidelity index of hierarchical nodes, the following steps are also taken: using an encoder to vectorize the core semantics of the whole text, calculating the modulus of each core semantic vector, and segmenting the modulus of the core semantic vector to obtain the topic keyword sequence.

4. The method for automatically generating text outlines based on artificial intelligence according to claim 3, characterized in that, The Otsu threshold segmentation method is used to segment the core semantics, and the sequence of core semantics that are greater than the segmentation threshold is taken as the subject keyword sequence of the document.

5. The method for automatically generating text outlines based on artificial intelligence according to claim 1, characterized in that, The method of generating text outlines from documents using a text outline generation model includes: generating a first-level outline from the document using the text outline generation model; using the generated first-level outline as context and combining it with temperature parameters adjusted by an iterative stability index to generate deeper sub-items; calculating the iterative stability index in real time after each iteration; stopping the iteration when the convergence requirement is met and outputting the final text outline with clear hierarchy and semantic coherence.

6. The method for automatically generating text outlines based on artificial intelligence according to claim 1, characterized in that, The preprocessing method for the original corpus is as follows: regular expression matching is used to remove webpage noise; a segmentation function is used to structurally split the long text, identify and retain the original chapter numbers and hierarchical headings; and NLP word segmentation tools are called to perform Chinese word segmentation, part-of-speech tagging and stop word filtering.

7. The method for automatically generating text outlines based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the raw corpus is as follows: collect the long text data to be processed through the enterprise's internal knowledge base API, and obtain the raw corpus through keyword retrieval and URL depth traversal.

8. An artificial intelligence-based automatic text outline generation system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an artificial intelligence-based method for automatically generating text outlines according to any one of claims 1-7.

Images