A video automatic clipping method based on semantic analysis

Abstract

The application relates to a video automatic clipping method based on semantic analysis and belongs to the technical field of video clipping. The method comprises the following steps: extracting anchor point information of text content of a streaming media file based on a knowledge graph, establishing a support relationship, combining the support relationship and the anchor point information to determine a main anchor point, a sub-anchor point and an argument anchor point; constructing a dynamic triggering mechanism, setting a verification direction of the dynamic triggering mechanism, determining a semantic extension direction of the anchor point information according to the verification direction, and verifying whether a prerequisite chain supporting the anchor point information exists; setting a credibility scoring strategy, determining the credibility of the anchor point information according to logical complexity, language confidence and text consistency, forming a sequence set of the anchor point information, and splicing to form a logically coherent video clipping text result. The method realizes automatic and intelligent video clipping based on semantic analysis, and significantly improves the clipping efficiency and quality.

Description

Technical Field

[0001] This invention belongs to the field of video editing technology and relates to an automatic video editing method based on semantic analysis. Background Technology

[0002] With the rapid development of multimedia technology and the explosive growth of video content, the demand for automatic video editing technology in short video creation, film and television production, and other fields is becoming increasingly urgent. Existing video editing methods are mainly divided into two categories: manual editing and automatic editing. Traditional manual editing relies on the editor's experience and subjective judgment. Due to differences in understanding of video content among different editors, problems such as logical inconsistencies and omissions of key information easily arise when splitting video footage. Furthermore, the editing efficiency is low, making it difficult to meet the demands of processing massive amounts of video. On the other hand, existing automatic editing methods are mostly based on superficial visual or auditory features for segment selection and splicing, lacking a deep understanding of the semantic level of video content. This may result in the awkward splicing of content from different themes, leading to logical breaks in the editing result and an inability to accurately convey core information. Therefore, both the subjective biases of manual editing and the insufficient semantic understanding of automatic editing make it difficult for existing editing technologies to produce logically coherent, high-quality videos. A new solution that can deeply analyze video semantics and improve the logical consistency of editing is urgently needed. Summary of the Invention

[0003] The purpose of this invention is to provide a semantic analysis-based automatic video editing method. By extracting anchor point information and establishing supporting relationships, a sequence set of anchor point information is formed. This solves the problems of logical breaks caused by subjective understanding biases in manual editing and the lack of logical construction ability in automatic editing in existing technologies, thus achieving automatic video editing with semantic logical coherence. To achieve the above objective, this invention is implemented using the following technical solution.

[0004] On one hand, this invention provides a method for automatic video editing based on semantic analysis, comprising: acquiring a streaming media file and extracting text content from the streaming media file; constructing a knowledge graph, extracting anchor information from the text content based on the knowledge graph, establishing supporting relationships, and determining main anchors, sub-anchors, and argument anchors by combining supporting relationships and anchor information; constructing a dynamic triggering mechanism, the dynamic triggering mechanism including forward prediction and backward verification, determining the semantic extension direction of the anchor information based on the forward prediction, and determining whether there is a precondition chain supporting the anchor information based on the backward verification; setting a credibility scoring strategy, the credibility scoring strategy including three-dimensional verification logic, the three-dimensional verification logic determining the credibility of the anchor information based on logical complexity, language confidence, and text consistency, forming an anchor information sequence set, and sequentially concatenating the anchor information sequence set to form the text result after editing the streaming media file.

[0005] Furthermore, the step of extracting anchor information from the text content based on the knowledge graph, establishing supporting relationships, and determining main anchors, sub-anchors, and argument anchors by combining the supporting relationships and anchor information includes: dividing the anchor information into main anchors, sub-anchors, and argument anchors according to the hierarchical structure of the knowledge graph, wherein the main anchor corresponds to the top-level core concept in the knowledge graph, the sub-anchors correspond to the middle-level branch concepts, and the argument anchors correspond to the bottom-level data instances; establishing supporting relationships between the anchor information, wherein each main anchor supports at least two sub-anchors, each sub-anchor is associated with at least one argument anchor, and the existence of the supporting relationships between the anchor information is verified by the knowledge graph rule engine through the relationship confidence between entities in the knowledge graph.

[0006] Further, determining the semantic extension direction of the anchor information based on the forward prediction includes: mapping the topic words and corresponding viewpoint directions in the text content to the entity associations in the knowledge graph based on the entity associations in the knowledge graph; converting the topic words and viewpoint directions into semantic vectors using the BERT model; calculating the cosine similarity between the two; and confirming successful mapping when the semantic similarity exceeds a preset threshold; generating a semantic extension direction based on the entity associations in the knowledge graph. The semantic extension direction is generated by successfully matching the vector similarity between the topic words and viewpoint directions, thus generating a path extending from the current anchor information to the associated entity.

[0007] Furthermore, the backward verification to determine whether there is a precondition chain supporting the anchor information includes: when the topic words in the text content are detected to generate candidate anchor information, a backtracking threshold is set, and the preceding sentences of the candidate anchor information are backtracked step by step according to the backtracking threshold to verify whether there is a precondition chain supporting the viewpoint.

[0008] Furthermore, the three-dimensional verification logic determines the credibility of anchor information based on logical complexity, linguistic confidence, and textual consistency, including: the first dimension, logical complexity, which calculates the logical depth value by analyzing the length of the precondition chain of anchor information; the second dimension, linguistic confidence, which assesses the certainty of the expression based on text sentiment analysis and calculates the professionalism score by combining the frequency of domain terms; and the third dimension, textual consistency, which calculates the semantic similarity score between the anchor information and the preceding statements.

[0009] Furthermore, forming the sequence set of anchor information includes: using a dynamic weight allocation algorithm to assign a weight coefficient to each anchor information. The weight coefficient is dynamically determined based on the importance score of the entity in the knowledge graph and the position of the anchor information in the supporting relationship. The logical complexity weight of the main anchor is higher than that of the argument anchor, and the text consistency weight of the argument anchor is higher than that of the main anchor.

[0010] Furthermore, the credibility scoring strategy also includes classifying anchor credibility levels based on the credibility score of the anchor information and the evidence support of conflicting anchors: A credibility score is calculated by analyzing the logical complexity, linguistic confidence, and textual consistency of the anchor information, and the difference between supporting and rebuttal evidence is calculated as the conflict support based on a knowledge graph rule engine; a first credibility threshold and a first conflict support threshold are set. When the anchor information simultaneously satisfies both the credibility score exceeding the first credibility threshold and the conflict support exceeding the first conflict support threshold, and there are at least two independent premise chains, it is classified as a high credibility level; a second credibility threshold and a second conflict support threshold are set below the first threshold. When the anchor information satisfies either the credibility score exceeding the second credibility threshold or the conflict support exceeding the second conflict support threshold, and there is one complete premise chain or multiple partially supporting evidence chains, it is classified as a medium credibility level; when the anchor information does not meet the above classification conditions for high or medium credibility levels, it is classified as a low credibility level.

[0011] Furthermore, the assessment of the evidentiary support for conflict anchors includes: identifying logically contradictory conflicts through knowledge graph attribute assertion comparison algorithms; verifying the completeness of preconditions using a knowledge graph rule engine to identify insufficient evidence conflicts; and calculating the semantic vector similarity between video footage and audio text content using the CLIP model to identify cross-modal inconsistencies. A three-layer analytical framework is constructed based on argument game theory. In this framework, the claim layer extracts the core viewpoints of both sides of the conflict using the BERT model; the rebuttal layer uses dependency parsing to identify causal negation or factual opposition between viewpoints; and the defense layer uses evidence theory to integrate the confidence levels of multiple sources of arguments, calculates the difference in evidentiary support between the supporting and rebuttal arguments, and triggers anchor information retention when the difference exceeds a preset threshold.

[0012] Furthermore, the step of sequentially concatenating the anchor information sequence set to form the edited text result of the streaming media file includes: setting a video segment retention strategy based on the anchor credibility level classification result; retaining segments corresponding to high credibility anchors as core content; compressing key information of segments corresponding to medium credibility anchors using the TextRank algorithm; and replacing segments corresponding to low credibility anchors with conflict markers or deleting them; using a knowledge graph rule engine to analyze the causal, temporal, and hierarchical relationships between anchors, and determining the timeline arrangement order of video segments based on the support relationships between anchors.

[0013] Furthermore, the construction of the knowledge graph includes: collecting data information, identifying entities in the data information through the BERT model, and extracting semantic relationships between entities using a remote supervised relation extraction algorithm. The semantic relationships include causal relationships, hierarchical relationships, temporal relationships, and attribute relationships. An entity alignment mechanism is established to resolve entity ambiguity issues in multi-source data by calculating the semantic similarity between entities. The identified entities are mapped to graph nodes, the extracted relationships are mapped to graph edges, and entity attributes are mapped to the attributes of nodes or edges.

[0014] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0015] (1) Extract anchor information through dynamic triggering mechanism, establish support relationship, identify main anchor, sub-anchor and argument anchor of text content, improve semantic understanding depth, and arrange the clips according to anchor sequence set;

[0016] (2) Credibility scores are given based on logical complexity, language confidence and text consistency. Combined with a dynamic weight allocation algorithm, the weights of each dimension are adjusted according to the position of anchor information in the supporting relationship, which improves the accuracy of anchor credibility assessment.

[0017] (3) Based on the argument game theory, the credibility of the conflict anchor point is dynamically adjusted, video frame numerical evidence and audio emotional evidence are integrated, and the anchor point is corrected or deleted by calculating the difference in evidence support, so as to avoid semantic contradictions in the edited segments and improve the consistency and reliability of the edited content.

[0018] (4) Based on the audio and video synchronization achieved by the dynamic time warping algorithm, the priority and order of segments are determined according to the credibility of anchor points and the support of evidence, and the transition effect is automatically generated, which improves the smoothness of the edited segments and enhances the logical coherence of the editing results. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the workflow of a semantic analysis-based automatic video editing method.

[0020] Figure 2 Flowchart for anchor point information extraction;

[0021] Figure 3 This is a schematic diagram of anchor point information;

[0022] Figure 4 This is a schematic diagram of splicing according to the anchor point sequence set. Detailed Implementation

[0023] 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.

[0024] 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.

[0025] Secondly, the term "one 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 in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0027] This embodiment introduces a video automatic editing method based on semantic analysis, such as... Figure 1 As shown, it includes:

[0028] S101. Obtain a streaming media file and extract the text content from the streaming media file;

[0029] S102. Construct a knowledge graph, extract anchor information from the text content based on the knowledge graph, establish supporting relationships, and determine the main anchor, sub-anchor, and argument anchor by combining the supporting relationships and anchor information.

[0030] S103. Construct a dynamic triggering mechanism, which includes forward prediction and backward verification. The semantic extension direction of the anchor information is determined based on the forward prediction, and the backward verification determines whether there is a precondition chain that supports the anchor information.

[0031] S104. Set a credibility scoring strategy, which includes three-dimensional verification logic. The three-dimensional verification logic determines the credibility of anchor information based on logical complexity, language confidence, and text consistency, forms a sequence set of anchor information, and splices the sequence set sequentially to form the text result of the edited streaming media file.

[0032] This embodiment acquires streaming media files and extracts their text content as a semantic carrier for analysis. Anchor information is extracted from the text content and categorized into main anchors, sub-anchors, and argument anchors. A credibility scoring strategy is used to generate a sequence set of anchor information, ensuring logical relationships in subsequent editing. A dynamic triggering mechanism is constructed during anchor information extraction. A two-way verification method of forward prediction and backward verification ensures the rationality of the semantic extension direction of the anchor information and the sufficiency of its preconditions, avoiding semantic deviation. Furthermore, the reliability of the anchor information is quantified from three dimensions: logical complexity, linguistic confidence, and textual consistency. The splicing order of audio and video segments in the streaming media file is determined based on the anchor credibility level.

[0033] In this embodiment, the streaming media file is first acquired and the text content is extracted for preprocessing. FFmpeg is used to separate the audio and video tracks. On the audio track, WebRTC is used for noise reduction, and the audio is converted to text using the Whisper model. Speech segments are segmented using Mel-spectrum feature differences, and audio and video timestamps are aligned using a dynamic time warping algorithm. Terminology consistency is corrected based on a knowledge graph. Furthermore, a knowledge graph is constructed by collecting multi-source data such as academic papers and industry reports. Entities are identified using the BERT-BiLSTM-CRF model, and causal and temporal relationships are extracted using a graph neural network. After aligning multi-source entities using cosine similarity, a knowledge graph containing entities and their relationships is formed. Further, forward prediction converts the keywords in the text content into BERT vectors, calculates the cosine similarity with entity vectors in the knowledge graph, and determines whether the keywords in the anchor information are successfully mapped to entities in the knowledge graph. Backward verification generates candidate anchor information by detecting keywords in the anchor information, backtracks to construct a premise chain, and verifies the evidence support in the premise chain using a knowledge graph rule engine. This embodiment provides an implementable rule engine, Drools. Drools supports rule definition, reasoning, and execution, and can be combined with knowledge graphs for complex logical reasoning, such as calculating the support of conflicting evidence. Furthermore, a credibility scoring strategy is set up, evaluating from three dimensions: logical complexity, linguistic confidence, and textual consistency. Based on the scoring results and evidence support, anchor point credibility levels are divided, segments are arranged chronologically according to supporting relationships, and an editing result is generated, automatically inserting transitions.

[0034] Furthermore, the step of extracting anchor information from the text content based on the knowledge graph, establishing supporting relationships, and determining the main anchor, sub-anchor, and argument anchor by combining the supporting relationships and anchor information includes:

[0035] S201. Based on the hierarchical structure of the knowledge graph, the anchor information is divided into main anchors, sub-anchors and argument anchors, wherein the main anchor corresponds to the top-level core concept in the knowledge graph, the sub-anchors correspond to the middle-level branch concepts, and the argument anchors correspond to the bottom-level data instances.

[0036] S202. Establish the support relationship between the anchor points, where each main anchor point supports at least two sub-anchor points, and each sub-anchor point is associated with at least one argument anchor point. The support relationship between the anchor points is verified by the knowledge graph rule engine through the relationship confidence between entities in the knowledge graph.

[0037] In this embodiment, during the construction of the knowledge graph, multi-source domain data is first collected and preprocessed. The BERT-BiLSTM-CRF model is then used to identify entities in the text content and map them to graph nodes. Top-level core concept entities serve as main anchors corresponding to main nodes, mid-level branch concept entities serve as sub-anchors corresponding to branch nodes, and bottom-level data instance entities serve as argument anchors corresponding to leaf nodes. The top-level core represents the highest-level entity node in the knowledge graph, representing the core theme or key category of domain knowledge, and is the logical starting point and overarching concept of the entire semantic system. Mid-level branch concepts represent the intermediate-level entity nodes between the top-level core concepts and the bottom-level data, representing sub-domains or branch arguments of the core concepts. Bottom-level data instances represent the lowest-level entity nodes in the knowledge graph, representing verifiable factual information such as specific experimental data, cases, and phenomena. Figure 3 As shown. Further, graph neural networks are used to extract causal, temporal, and hierarchical relationships between entities, setting these relationships as edges, and calculating the cosine similarity of entity vectors to unify synonyms in multi-source data, achieving node alignment. For example, the semantic similarity between "fast charging" and "fast charging" is calculated using BERT vectors, merging entities with the same name but different meanings. Going further, the identified entity attributes are mapped to the attributes of nodes or edges, constructing a knowledge graph containing entity nodes and relational edges. The rationality of the edges is verified using a knowledge graph rule engine, and the PageRank algorithm is used to evaluate entity importance scores, dynamically adjusting the graph hierarchy to ensure that core concepts are identified first.

[0038] In the anchor point information extraction stage, firstly, based on the knowledge graph, the BERT-BiLSTM-CRF model is used to classify entity types and calculate the PageRank score of each entity, with higher scores indicating greater coreity. Therefore, the top 10% of core concept entities by score are used as primary anchor points, the middle-level branch concept entities by score (10%-30%) are used as sub-anchor points, and the bottom-level data instance entities are used as argument anchor points, with the importance threshold dynamically adjusted to 1.2 times the average score. Further, based on semantic similarity calculation, at least two related sub-anchor points are found for each primary anchor point to determine the sufficiency of the logical branches; using the BM25 algorithm and semantic retrieval algorithm, at least one argument anchor point is matched for each sub-anchor point to ensure the necessity of supporting evidence.

[0039] Furthermore, the anchor point support relationship is verified. The knowledge graph rule engine calculates the evidence support of the premise chain by propagating the relationship confidence. If the evidence support is greater than or equal to the set evidence support threshold, it indicates that the reasoning conclusion of the premise chain in the current knowledge graph has sufficient reliability and can be determined as a supporting relationship that meets the reasoning requirements. The formula for relationship confidence is:

[0040]

[0041] Where Confidence(h,r,t) represents the relation confidence, h represents the head entity, r represents the relation between entities, t represents the tail entity, count(h,r,t) represents the number of times (h,r,t) appears, and count(h,r,*) represents the total number of times the head entity h is connected to any tail entity through relation r.

[0042] The formula for the degree of support of evidence is:

[0043]

[0044] Among them, Support(h1,r n ,t n ) represents the degree of evidence support, n represents the length of the premise chain, i represents the level depth, counted sequentially from 1 to n, h i r i t i These correspond to the head entity, relation, and tail entity of the i-th stage, respectively.

[0045] Because the mean support of the premise chain in a large number of historical data experiments is 0.65 and the standard deviation is 0.05, based on the statistical characteristics of similar premise chains in the knowledge graph, 0.7 is set as the minimum standard for the credibility of the premise chain, denoted as the evidence support threshold. Evidence support is a measure of the overall reliability of the reasoning conclusion obtained by the knowledge graph rule engine based on the relationship confidence in the premise chain through reasoning and calculation methods. If the calculated evidence support is ≥0.7, the chain relationship is considered to have basic reliability in the knowledge graph logic and can be used as valid supporting evidence for anchor information; if the evidence support is <0.7, the premise chain is considered to have loopholes, such as insufficient relationship confidence, and supplementary evidence or anchor point correction is required. For example, when verifying "sulfide electrolyte → increased energy density", if there is a chain relationship with a relationship confidence of 0.8 for "sulfide electrolyte → increased ionic conductivity" and a relationship confidence of 0.9 for "increased ionic conductivity → increased energy density", then the evidence support is 0.8 × 0.9 = 0.72 ≥ 0.7.

[0046] Further, determining the semantic extension direction of the anchor information based on the forward prediction includes:

[0047] S301. Based on the entity association relationships in the knowledge graph, the topic words and corresponding viewpoint directions in the text content are mapped to the entity association relationships in the knowledge graph. The topic words and viewpoint directions are converted into semantic vectors through the BERT model, and the semantic similarity between the two is calculated. When the semantic similarity exceeds a preset threshold, the mapping is confirmed to be successful.

[0048] S302. Generate semantic extension directions based on entity relationships in the knowledge graph. The semantic extension directions are generated by successfully matching the vector similarity between the topic word and the viewpoint direction, and then generating a path extending from the current anchor information to the associated entity.

[0049] In this embodiment, the forward prediction is represented as a knowledge graph-based semantic navigator. It calculates the semantic extension direction of anchor information within the knowledge graph using predefined entity relationships and BERT semantic vectors. The purpose of determining the semantic extension direction based on forward prediction is to organize illogical text content according to the logical system of domain knowledge, avoiding semantic confusion caused by editing fragments. Figure 2As shown. First, a viewpoint direction vector library is constructed. The core purpose is to provide clear and logically consistent guidance for the semantic extension of anchor information, addressing the problems of fragmented text semantic organization and insufficient coherence. Firstly, the knowledge graph is broken down into core directions such as technical principles, application scenarios, and development trends according to domain knowledge logic. Each direction corresponds to a set of associated entities, such as the technical principles direction being associated with "sulfide electrolytes, ionic conductivity," etc. The TransE algorithm is used to convert entities into low-dimensional semantic vectors, quantifying entity relationships. The viewpoint direction vector is synthesized by weighting the associated entity vectors of each direction according to their PageRank scores. Core entities with higher PageRank scores have higher weights, ensuring the vectors contain both entity semantics and reflect the logic of knowledge importance. Further, entity mapping is performed on topic words. Topic words in the audio text are converted into vectors using BERT. The semantically most similar entities are retrieved from the knowledge graph, and their cosine similarity is calculated. Further, direction prediction and path generation are performed. The similarity between the topic word vector and each viewpoint direction vector is calculated. The direction with the highest similarity is selected. When the similarity exceeds a preset threshold, the semantic mapping is confirmed as successful, guiding the semantic extension direction of the anchor information. Based on the entity association types in the knowledge graph, an extension path is generated starting from the matching entity. Furthermore, dynamic threshold adjustment is performed. The similarity threshold is automatically adjusted according to knowledge complexity to ensure mapping accuracy. For example, based on predefined entity associations in the knowledge graph, a vector library of viewpoints such as technical principles, application scenarios, and development trends is constructed. Each direction vector is generated by weighted averaging of associated entity vectors according to their PageRank scores. When a keyword such as "low-temperature performance" is detected in the text content, it is converted into a BERT vector, and its cosine similarity is calculated with each viewpoint direction vector. An average similarity threshold is set for comparison. If the cosine similarity between "low-temperature performance" and "technical challenges" is greater than the average similarity threshold, the mapping is confirmed as successful, and a path is generated through causal edges in the knowledge graph as "low-temperature performance → influencing factors → electrolyte characteristics → material improvement". This mechanism, based on the knowledge graph, reduces semantic deviation by predicting the semantic extension direction of the text content forward, effectively ensuring the semantic coherence of the video content and providing guidance for the logical extension of anchor information.

[0050] Furthermore, the backward verification to determine whether a precondition chain supporting the anchor point information exists includes:

[0051] When candidate anchor information is generated from topic words in the text content, a backtracking threshold is set, and the preceding statements of the candidate anchor information are backtracked step by step according to the backtracking threshold to verify whether there is a precondition chain that supports the viewpoint.

[0052] This embodiment achieves deep verification of the logical rationality of anchor point information by constructing a backward verification module with a dynamic triggering mechanism. For example... Figure 2As shown, when keywords, including conclusion words, sequence words, and explanatory words, are detected in the text content, candidate anchor point generation is automatically triggered. An N-gram sliding window is used to backtrack the preceding statements of the candidate anchor point information step by step according to the backtracking threshold, extracting the premise chain. The existence of premises in the preceding text is checked, and the causal relationship between these premises is verified using a knowledge graph. If the evidence support is greater than the evidence support threshold, the logic is deemed reasonable. During this process, BERT is used to calculate the semantic similarity between the candidate anchor point and the premises, and the completeness of the premise chain is verified using a knowledge graph rule engine to ensure the semantic relevance between the premises and the conclusion.

[0053] Furthermore, the three-dimensional verification logic determines the credibility of anchor information based on logical complexity, language confidence, and text consistency, including:

[0054] S401, First dimension of logical complexity: Calculate the logical depth value by analyzing the length of the precondition chain of the logical identification anchor information.

[0055] S402, the second dimension, language confidence, assesses the degree of certainty of the expression based on text sentiment analysis, and calculates the professionalism score by combining the frequency of occurrence of domain terms;

[0056] S403, Third Dimension: Textual Consistency. This involves calculating the semantic similarity score between anchor information and preceding statements.

[0057] This embodiment achieves multi-dimensional quantitative evaluation of anchor information by setting a credibility scoring strategy and constructing a three-dimensional verification logic. The first dimension is logical complexity. The knowledge graph rule engine parses the precondition chain of the anchor information, calculates the length of the precondition chain and the average number of association levels. The average number of association levels is used to measure the hierarchical depth of relationships between entities in the knowledge graph. If the chain length exceeds the average number of association levels, an exponential decay algorithm is triggered to adjust the score, avoiding overly complex precondition chains that could lead to semantic understanding difficulties. Decay coefficient = e -0.2×(链长-平均层级数) For example, regarding the anchor point "sulfide electrolyte improves the energy density of solid-state batteries," the formula is "sulfide electrolyte → improved ionic conductivity → improved energy density," and the chain length is calculated to be 2 layers. When the chain length is 3 layers, substituting into the formula yields the attenuation coefficient e. -0.2×1.2 ≈0.78. At this point, the logical complexity score is adjusted from the base score of 80 to 80 × 0.78 = 62.4, in order to avoid the semantic understanding difficulty caused by overly complex precondition chains.

[0058] Furthermore, the second dimension is language confidence, which integrates sentiment analysis and terminology recognition through natural language processing technology. The implementation logic for this dimension is as follows: First, sentiment analysis models such as VADER are used to analyze deterministic and ambiguous words in the text, assigning different initial confidence weights to each. Simultaneously, the BERT-BiLSTM-CRF model is used to identify terms, setting a frequency threshold for each term occurrence: at least two occurrences per 100 words. An increment is set, and for each occurrence exceeding this threshold, the professionalism score iterates and increments by the increment. The language confidence score is calculated using the formula: Language Confidence Score = Initial Weight × (1 + Increment). Taking "sulfide electrolytes necessarily increase ionic conductivity" as an example, the initial confidence weight for this statement is 0.9. The terms "sulfide electrolytes" and "ionic conductivity" appear twice, with an increment of 10%, so 0.9 × 1.1 = 0.99, resulting in a final score of 99.

[0059] Furthermore, the third dimension is text consistency. A sliding window mechanism is used to extract preceding statements, and the anchor information and preceding statements are converted into semantic vectors using the BERT model. The cosine similarity between the two is then calculated. Since a similarity threshold that is too high will miss semantically weak but reasonably extended content, and a similarity threshold that is too low will tolerate too much semantic deviation, this embodiment sets the similarity threshold to 0.7. If the calculated cosine similarity is higher than this threshold, it indicates a strong causal relationship; if the calculated cosine similarity is lower than this threshold, a semantic conflict warning is triggered and the anchor credibility is reduced, indicating a weak causal relationship. Taking the anchor information "solid-state batteries have a low risk of thermal runaway" as an example, by backtracking the preceding text "solid-state batteries have no liquid electrolyte and sulfide electrolytes have high thermal stability," the cosine similarity between the two vectors is calculated using the BERT model to be 0.83. Since 0.83 > 0.7, the text consistency score is rated as 83 points. If the semantic similarity between the anchor information and the preceding text is less than 0.7, it indicates that there is a deviation in the topic between the two, and the semantic conflict warning mechanism is automatically triggered to adjust the credibility of the anchor information accordingly.

[0060] Furthermore, the scores based on the three-dimensional verification logic are weighted and fused to determine the total credibility score of the anchor information. In this embodiment, the weights for logical complexity, linguistic confidence, and textual consistency are preferably set at 40%, 30%, and 30%, respectively. This ratio aligns with the priority requirement of video editing for content that prioritizes "logical rigor > accurate expression = semantic coherence," ensuring that the selected anchor information can serve as a reliable basis for editing. Taking the scores of each dimension as an example, the total credibility score is calculated as 62.4 × 40% + 99 × 30% + 83 × 30% = 79.56 points. When the total score is ≥ 75 points, the anchor information is considered highly credible and can be used as the core basis for video editing; when 60 points ≤ total score < 75 points, further verification of the preconditions or optimization of the expression is required; when the total score < 60 points, it is determined to be of low credibility.

[0061] Further, forming the sequence set of the anchor point information includes:

[0062] A dynamic weight allocation algorithm is adopted to assign a weight coefficient to each anchor point. The weight coefficient is dynamically determined based on the importance score of the entity in the knowledge graph and the position of the anchor point in the supporting relationship. The logical complexity weight of the main anchor point is higher than that of the argument anchor point, and the text consistency weight of the argument anchor point is higher than that of the main anchor point.

[0063] This embodiment uses a dynamic weighting algorithm to form a sequence set of anchor information. First, the importance score of the anchor information is determined based on the PageRank scores of entities in the knowledge graph. Since the main anchor corresponds to the top-level core concept, its PageRank score is typically in the top 10%, while the supporting anchor corresponds to mid-level branch concepts, with scores in the top 10%-30%. Based on this, the weight coefficients are dynamically adjusted according to the position of the anchor information in the supporting relationship: the main anchor, as the core of the premise chain, has its logical complexity weight increased to 45%, higher than the supporting anchor's 35%, to ensure the logical rigor of the core viewpoint; the supporting anchor, as evidence supporting the main anchor, has its textual consistency weight increased to 35%, higher than the main anchor's 30%, to ensure the semantic coherence of the evidence with the preceding discussion. For example, the main anchor "the energy density advantage of solid-state batteries" has a logical complexity weight of 45%, a linguistic confidence weight of 30%, and a textual consistency weight of 25%; its supporting anchor "the improved ion conductivity of sulfide electrolytes" has a logical complexity weight of 35%, a linguistic confidence weight of 30%, and a textual consistency weight of 35%. This dynamic weight allocation mechanism forms a hierarchical set of anchor points, enabling differentiated optimization of core viewpoints and supporting arguments in terms of logical depth and semantic coherence, ultimately enhancing the logical hierarchy of the video content.

[0064] Furthermore, the credibility scoring strategy also includes classifying anchor credibility levels based on the credibility score of the anchor information and the evidentiary support of conflicting anchors, including:

[0065] S501. Calculate the credibility score by analyzing the logical complexity, linguistic confidence and text consistency of the anchor information, and calculate the difference between supporting evidence and rebuttal evidence as the conflict support score based on the knowledge graph rule engine.

[0066] S502. Set a first credibility threshold and a first conflict support threshold. When the anchor information simultaneously satisfies the conditions that the credibility score exceeds the first credibility threshold and the conflict support exceeds the first conflict support threshold, and there are at least two independent precondition chains, it is classified as a high credibility level.

[0067] S503. Set a second credibility threshold and a second conflict support threshold that are lower than the first threshold. When the anchor information satisfies the credibility score exceeding the second credibility threshold or the conflict support exceeding the second conflict support threshold, and there is a complete precondition chain or multiple partially supporting evidence chains, it is classified as a medium credibility level.

[0068] S504. When the anchor information does not meet the above-mentioned criteria for high or medium confidence, it is classified as low confidence level.

[0069] This embodiment constructs an anchor sequence set by integrating anchor credibility scores and conflicting evidence support scores. The difference in evidence support scores between conflicting anchors, i.e., the difference in confidence between supporting and refuting evidence, is evaluated using a knowledge graph rule engine. The knowledge graph organizes the evidence supporting and refuting anchor information into structured information. After scoring this evidence using the knowledge graph rule engine, the difference in scores reflects the relative strengths of the supporting and refuting evidence. When calculating the evidence support scores of conflicting anchors, based on the knowledge graph rule engine and N-gram sliding window technology, all evidence related to the anchor information is extracted from the text content. The semantic relevance between the evidence and the anchor information is calculated using the BERT model, and invalid evidence with a relevance score below a preset similarity threshold is deleted. Simultaneously, the confidence score of valid evidence is evaluated: on the one hand, an initial confidence score is assigned based on the entity relationships in the knowledge graph, such as 0.8 for strong causal relationships and 0.4 for weak relationships; on the other hand, the confidence score is weighted and adjusted by combining the linguistic confidence score of the anchor information. The formula for the confidence level of a single piece of evidence is: Confidence Level of Single Evidence = Initial Confidence Level × Linguistic Confidence Level. Furthermore, when multiple pieces of similar evidence exist, the Dempster-Shafer evidence theory is used for fusion calculation to improve the accuracy of the evidence confidence level. The fused confidence level is the result of integrating the confidence levels of multiple pieces of similar evidence using the Dempster-Shafer evidence theory. When there are two or more pieces of similar evidence, the correlation and conflict between the propositions of each piece of evidence are comprehensively considered. The confidence levels of multiple pieces of evidence are merged, weakening contradictions and strengthening consistent parts. This allows the fused confidence level to more accurately reflect the overall support or refutation strength of this type of evidence for the anchor point, improving the accuracy of the assessment and providing a more reliable basis for subsequent calculations of the support of conflicting evidence and judgment of the logical relationship of the anchor point.

[0070] Furthermore, anchor points with a total credibility score of 75 or higher and a conflict support score of 0.5 or higher are classified as high-credibility anchor points. These anchor points are used to construct a complete chain of preconditions, including the main argument, sub-arguments, and corresponding evidence. Medium-credibility anchor points are defined as those with a total credibility score between 60 and 75, or a conflict support score between 0.2 and 0.5. For these anchor points, a content compression mechanism is implemented, using the TextRank algorithm to extract core elements of the argument and key evidence fragments. Anchor points with a total credibility score below 60, or a conflict support score below 0.2, are classified as low-credibility anchor points and replaced with conflict-marked warning boxes. Simultaneously, multimodal evidence, including experimental data charts and comparative literature texts, is automatically retrieved to generate visual comparison charts. For example, regarding the contradictory viewpoints of high cost and increased energy density in solid-state batteries, the evidence comparison is displayed as a curve, helping users quickly identify logical conflicts. This mechanism ensures the logical integrity of core viewpoints through a credibility level classification strategy, provides visual conflict warnings for low-credibility content, and achieves intelligent layering and logical optimization of video content.

[0071] Furthermore, the evidentiary support for assessing the conflict anchor points includes:

[0072] S601. Logical contradiction type conflicts are identified by using the knowledge graph attribute assertion comparison algorithm, insufficient evidence type conflicts are identified by using the knowledge graph rule engine to verify the completeness of the preconditions, and cross-modal inconsistency type conflicts are identified by using the CLIP model to calculate the semantic vector similarity between video images and audio text content.

[0073] S602. Based on the argumentation game theory, a three-layer analysis framework is constructed. The claim layer extracts the core viewpoints of both sides of the conflict through the BERT model. The rebuttal layer uses dependency parsing to identify the causal negation or factual opposition between viewpoints. The defense layer uses evidence theory to integrate the confidence of multiple sources of arguments and calculates the difference in the support of supporting and rebuttal evidence. When the difference exceeds a preset threshold, the anchor information is retained.

[0074] This embodiment achieves a systematic evaluation of the evidentiary support of conflict anchors by defining conflict anchor types and combining them with argumentation game theory. First, conflict anchors are identified based on knowledge graphs and multimodal semantic analysis. For logically contradictory conflicts, a knowledge graph attribute assertion comparison algorithm is used to detect mutually exclusive attributes of the same entity. For example, simultaneously asserting "solid-state battery energy density 300Wh / kg" and "less than 200Wh / kg". Conflicts with insufficient evidence are verified for the completeness of preconditions using a knowledge graph rule engine. For example, the anchor "sulfide electrolyte improves battery performance" lacks the pre-set premise "improved ion conductivity". Cross-modal inconsistency conflicts are calculated using the CLIP model to calculate the cosine similarity between video image features and audio text semantic vectors. A multimodal threshold is set; when the similarity is below the multimodal threshold, it is determined to be a modal conflict.

[0075] Furthermore, a three-layer analytical framework is constructed based on argumentation game theory. At the assertion layer, BERTClassifier extracts the core viewpoints of both sides in the conflict, such as "solid-state batteries are expensive" and "mass production can reduce costs." The rebuttal layer uses dependency parsing to identify contradictory relationships between viewpoints, including causal denial and factual opposition. The defense layer employs Dempster-Shafer evidence theory to integrate the confidence levels of multiple sources of evidence. The difference in evidence support is calculated using the formula: Evidence Support Difference = Total Confidence of Supporting Evidence - Total Confidence of Rebuttal Evidence. When the evidence support difference exceeds a threshold, anchor information retention is triggered; when it falls below the threshold, an anchor information correction process is triggered or the information is marked for deletion, effectively filtering out low-quality anchor information.

[0076] Further, the step of sequentially concatenating the anchor point information sequence set to form the edited text result of the streaming media file includes:

[0077] S701. Based on the anchor point confidence level classification results, set the video segment retention strategy. The segments corresponding to high confidence anchor points are retained as core content. The segments corresponding to medium confidence anchor points use the TextRank algorithm to compress key information. The segments corresponding to low confidence anchor points are replaced with conflict markers or deleted.

[0078] S702. Use the knowledge graph rule engine to analyze the causal, temporal, and hierarchical relationships between anchor points, and determine the timeline arrangement order of video clips based on the support relationships between anchor points.

[0079] This embodiment uses anchor point confidence level and support relationship analysis to achieve intelligent editing and text result generation of streaming media files, and completes the timeline arrangement of video segments based on the support relationship between anchor points. For example... Figure 4As shown. First, a video segment retention priority system is constructed based on the anchor point's credibility level. For high-credibility anchor points, i.e., anchor points with a total credibility score of not less than 75 points and a conflict support of 0.5 or higher, their corresponding video segments are given the highest retention priority and marked as core content. For example, in the processing of segments explaining the technical principles of the commercialization prospects of solid-state batteries, such anchor point segments will be retained in their entirety. For medium-credibility anchor points, those meeting the conditions of a total credibility score between 60 and 75 points, or a conflict support between 0.2 and 0.5, their corresponding segments are set as secondary content, and a compression mechanism is activated to retain key information, such as simplifying segments demonstrating experimental data on the improvement of low-temperature performance of solid-state batteries. For low-credibility anchor points, i.e., anchor points with a total credibility score below 60 points or a conflict support of less than 0.2, their corresponding segments are marked as content to be verified, and the system replaces them with conflict annotation prompts or deletes them directly, such as processing segments involving contradictory viewpoints on the cost and energy density of solid-state batteries.

[0080] Furthermore, the construction of the knowledge graph includes:

[0081] S801. Collect data information, identify entities in the data information through the BERT model, and extract semantic relationships between entities by applying the remote supervised relation extraction algorithm. The semantic relationships include causal relationships, hierarchical relationships, temporal relationships, and attribute relationships.

[0082] S802. Establish an entity alignment mechanism to solve entity ambiguity problems in multi-source data by calculating the semantic similarity between entities.

[0083] S803. Map the identified entities to graph nodes, the extracted relations to graph edges, and the entity attributes to the attributes of nodes or edges.

[0084] This embodiment implements semantic parsing based on deep learning and weakly supervised learning principles. First, utilizing the bidirectional Transformer architecture of the BERT model, a deep semantic representation of the text is learned through a masked language model and a next-sentence prediction task, mapping the input data into semantic vectors. Then, entity boundaries are decoded through a CRF layer to achieve end-to-end entity recognition. For relation extraction, a multi-instance learning framework using a remotely supervised relation extraction algorithm is employed. Entity pairs from knowledge bases such as Freebase are used as seeds, such as "sulfide electrolyte—improvement—energy density." Heuristic rules are used to match text content containing this entity pair. CNN convolutional layers extract local sentence features, and an attention mechanism is used to weight sentences within the multi-instance package, filtering irrelevant statements such as "sulfide electrolyte is discussed in a laboratory environment," while retaining valid evidence such as "sulfide electrolyte improves energy density by reducing interface impedance." Finally, a softmax layer classifies the relation type. Furthermore, an entity alignment mechanism is established based on knowledge graph embedding technology to address entity ambiguity in multi-source data. The TransE algorithm is used to map entities from different data sources to a unified vector space, so that entity pairs that satisfy relational constraints, such as "solid-state battery" and "all-solid-state battery", are closer in distance in the vector space. The semantic distance of entity vectors is calculated by cosine similarity, and a threshold is set as the alignment standard.

[0085] In summary, this invention extracts anchor information through forward prediction and backward verification, establishes support relationships between the anchor information, scores the credibility of the anchor information, and solves the problem of audio-visual content mismatch caused by conflicting anchors. This improves the logical coherence and natural fluency of video editing, providing an efficient and intelligent automated editing solution for short video creation and other fields.

[0086] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for automatic video editing based on semantic analysis, characterized in that, The method comprises the following steps: acquiring a streaming media file and extracting text content in the streaming media file; constructing a knowledge graph, extracting anchor information in the text content based on the knowledge graph, establishing support relationships, and determining main anchors, sub-anchors and argument anchors in combination with the support relationships and the anchor information; The method of extracting anchor information in the text content based on the knowledge graph, establishing support relationships, and determining main anchors, sub-anchors and argument anchors in combination with the support relationships and the anchor information comprises the following steps: According to the hierarchical structure of the knowledge graph, the anchor information is divided into main anchors, sub-anchors and argument anchors, wherein the main anchors correspond to the top-level core concepts in the knowledge graph, the sub-anchors correspond to the middle-level branch concepts, and the argument anchors correspond to the bottom-level data instances; establishing support relationships among the anchor information, wherein each main anchor supports at least two sub-anchors, each sub-anchor is associated with at least one argument anchor, and the existence of the support relationships among the anchor information is verified by a knowledge graph rule engine through the relationship confidence between entities in the knowledge graph; constructing a dynamic triggering mechanism, wherein the dynamic triggering mechanism comprises forward prediction and backward verification, the semantic extension direction of the anchor information is determined according to the forward prediction, and it is judged whether there is a prerequisite chain supporting the anchor information according to the backward verification; setting a credibility scoring strategy, wherein the credibility scoring strategy comprises three-dimensional verification logic, the three-dimensional verification logic determines the credibility of the anchor information according to logical complexity, language confidence and text consistency, forms a sequence set of the anchor information, and sequentially splices a text result after editing the streaming media file according to the sequence set of the anchor information. 2.The method of claim 1, wherein, The method of determining the semantic extension direction of the anchor information according to the forward prediction comprises the following steps: mapping the topic words and corresponding viewpoint directions in the text content to the entity association relationship in the knowledge graph based on the entity association relationship in the knowledge graph, converting the topic words and the viewpoint directions into semantic vectors through a BERT model, calculating the semantic similarity of the two, and confirming the mapping success when the semantic similarity exceeds a preset threshold; generating a semantic extension direction based on the entity association relationship in the knowledge graph, wherein the semantic extension direction is generated from the current anchor information to the associated entity after the vector similarity matching of the topic words and the viewpoint directions is successful. 3.The method of claim 2, wherein, The method of judging whether there is a prerequisite chain supporting the anchor information according to the backward verification comprises the following steps: When detecting the topic words in the text content to generate candidate anchor information, setting a backtracking threshold, backtracking the previous sentences of the candidate anchor information according to the backtracking threshold, and verifying whether there is a prerequisite chain supporting the viewpoint.

4. The method of claim 3, wherein, The method of determining the credibility of the anchor information according to the logical complexity, the language confidence and the text consistency comprises the following steps: The first dimension logical complexity identifies the prerequisite chain length of the anchor information through logical analysis and calculates the logical depth value; The second dimension language confidence evaluates the certainty degree of the expression based on the text sentiment analysis and calculates the professional degree score in combination with the frequency of the domain terminology; The third dimension text consistency calculates the semantic similarity score of the anchor information and the previous sentences.

5. The method of claim 4, wherein, The method of forming the sequence set of the anchor information comprises the following steps: The dynamic weight distribution algorithm is used to assign a weight coefficient to each anchor point information, and the weight coefficient is dynamically determined based on the importance score of the entity in the knowledge graph and the position of the anchor point information in the supporting relationship. The logical complexity weight of the main anchor point is higher than that of the argument anchor point, and the text consistency weight of the argument anchor point is higher than that of the main anchor point.

6. The method of claim 5, wherein, The credibility scoring strategy also includes dividing anchor point credibility levels according to the credibility scores of the anchor point information and the evidence support of the conflicting anchor points: By analyzing the logical complexity, language confidence and text consistency of the anchor point information, the credibility score is calculated, and the difference between the supporting evidence and the refutation evidence is calculated based on the knowledge graph rule engine as the conflict support; Set the first credibility threshold and the first conflict support threshold. When the anchor point information meets the conditions of credibility score exceeding the first credibility threshold and conflict support exceeding the first conflict support threshold, and there are at least two independent premise condition chains, it is divided into a high credibility level; Set the second credibility threshold lower than the first threshold and the second conflict support threshold. When the anchor point information meets the conditions of credibility score exceeding the second credibility threshold or conflict support exceeding the second conflict support threshold, and there is one complete premise condition chain or multiple partially supported evidence chains, it is divided into a medium credibility level; When the anchor point information does not meet the division conditions of the high credibility or medium credibility level, it is divided into a low credibility level.

7. The method of claim 6, wherein, The evidence support of the conflict anchor point includes: Determine the logical contradiction type conflict by the knowledge graph attribute assertion comparison algorithm, verify the completeness of the premise condition by the knowledge graph rule engine to determine the evidence deficiency type conflict, and use the CLIP model to calculate the semantic vector similarity of the video picture and the audio text content to determine the cross-modal inconsistency type conflict; Based on the argument game theory, a three-layer analysis framework is constructed. The main argument layer extracts the core viewpoints of the conflicting parties through the BERT model, the refutation layer identifies the causal negation or factual opposition relationship between the viewpoints by dependency syntax analysis, and the defense layer fuses the confidence of multiple source arguments using evidence theory to calculate the difference between the support and refutation evidence support. When the difference exceeds the preset threshold, the anchor point information is retained.

8. The method of claim 7, wherein, The text result after editing the streaming media file is formed by sequentially splicing the anchor point information sequence set, which includes: Set the video segment retention strategy according to the anchor point credibility level division result. The high credibility anchor point corresponding segment is completely retained as the core content, the medium credibility anchor point corresponding segment is compressed by the TextRank algorithm to extract key information, and the low credibility anchor point corresponding segment is replaced by conflict annotation or deleted; Use the knowledge graph rule engine to analyze the cause-and-effect, time sequence and hierarchical relationship between anchor points, and determine the time axis arrangement order of video segments based on the support relationship between anchor points. 9.The method of claim 1, wherein, The knowledge graph is constructed, which includes: Collect data information, identify entities in the data information through the BERT model, and extract semantic relationships between entities by applying remote supervision relationship extraction algorithm. The semantic relationships include causal relationship, hierarchical relationship, time sequence relationship and attribute relationship; Establish an entity alignment mechanism to solve the entity ambiguity problem in multiple source data by calculating the semantic similarity between entities; The identified entities are mapped as graph nodes, the extracted relations are mapped as graph edges, and the entity attributes are mapped as attributes of nodes or edges.

Images