News image value evaluation method and system based on predicate logic and multi-modal fusion

Abstract

The application discloses a news image value evaluation method and system based on predicate logic and multi-modal fusion, and the method comprises the following steps: acquiring multi-modal data of a news image, and extracting key frames of audio-visual data by using an adaptive triggering mechanism; performing semantic analysis on the key frames by using a predicate logic analysis engine, generating subject-action-scene triplets in combination with text data, and performing double verification on the triplets by using spatial position and semantic association; calculating a news value score based on a three-dimensional dynamic evaluation model for the verified triplet data; the three-dimensional dynamic evaluation model fuses propagation intensity, information entropy compensation and ontology value; and the evaluation parameters and rules are visualized and optimized, so that the dynamic evaluation model is optimized. The application can efficiently process massive multi-modal data, greatly shortens the response time to emergencies, improves the overall review efficiency, and makes the evaluation result accurate, efficient, transparent and reliable, so that the user can understand and trust the model decision.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and media technology, and more specifically to a method and system for evaluating the value of news images based on predicate logic and multimodal fusion. It is applicable to the automated evaluation, screening, grading, and dynamic optimization management of massive amounts of news images and videos in the context of user-generated content (UGC). Background Technology

[0002] Currently, with the widespread adoption of social media and mobile devices, UGC news video data is experiencing explosive growth, posing significant challenges to content review and value assessment for media organizations. Existing news value assessment technologies have the following shortcomings: Modal fragmentation: Traditional methods typically process information from different modalities such as text, images, and audio separately, lacking unified semantic fusion, resulting in cross-modal semantic association errors exceeding 28% (according to data from the ICCV 2022 report). For example, a news article may contain text descriptions and video footage, but existing systems struggle to accurately correlate the meanings of the two, reducing evaluation accuracy.

[0003] Rigid Assessment: Most existing systems are based on pre-defined static models, resulting in slow responses to sudden events and changes in public opinion. When hot topics emerge, the assessment results often lag behind actual changes in public opinion because the models cannot dynamically adjust weights. Manual assessments cause even longer delays, measured in hours. This rigid assessment mechanism cannot meet the needs of real-time news value judgment.

[0004] Black-box decision-making: Many news value assessments rely on deep learning black-box models, lacking transparent decision-making paths, leading to results that are difficult to interpret. According to industry feedback, this lack of interpretability increases the workload of manual review by approximately 35%, reducing overall efficiency.

[0005] Therefore, with the development of the media environment, there is an urgent need for a news image value assessment method or system that can integrate multimodal data, respond promptly to changes in public opinion, and provide explainable decision-making, in order to overcome many subjective errors caused by workers' experience-based judgments. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for evaluating the value of news images based on predicate logic and multimodal fusion, aiming to solve the technical problems of insufficient cross-modal semantic fusion, poor dynamic adaptability and lack of interpretability in existing news image value evaluation technologies.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide a method for evaluating the value of news images based on predicate logic and multimodal fusion, comprising the following steps: S10. Acquire multimodal data of news images, including: associated audio, visual and text data, and extract key frames of audio and visual data using an adaptive triggering mechanism; S20. Semantic parsing of keyframes is performed through the predicate logic parsing engine, and subject-action-scene triples are generated by combining text data, and verified by both spatial location and semantic association. S30. Calculate the news value score based on the verified triplet data using a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation, and ontological value. S40. The dynamic evaluation model is optimized by visually tuning the evaluation parameters and rules.

[0008] Further, in step S10, the adaptive triggering mechanism extracts keyframes of the audiovisual data, including: For visual data feature change detection, sudden changes in the image are analyzed by the color histogram difference between adjacent frames, triggering keyframe extraction; For audio data feature change detection, short-time energy analysis is used to detect energy changes between adjacent audio frames, triggering keyframe extraction.

[0009] Further, step S20 includes: S201. Use a pre-trained target detection model and a non-maximum suppression strategy to filter redundant detection boxes and perform subject detection. S202. Extract the subject motion features using the image sequence optical flow method, feed the visual features into the trained action classification model, and integrate the spatiotemporal attention mechanism to capture the subject action. S203. Use a pre-trained CNN model to determine the semantic label of the entire frame; at the same time, perform instance segmentation on the background region in the frame, extract scene elements, and generate a brief semantic description of the scene. S204. Spatial location verification is performed by calculating whether the overlap between the subject and the scene in space exceeds the first threshold; and semantic association verification is performed by using cross-modal semantic analysis to calculate whether the CLIP similarity between the triplet text description and the keyframe image is greater than the second threshold.

[0010] Further, in step S30, the three-dimensional dynamic evaluation model is represented as follows:

[0011] In the formula, I(t), H(t), and S(t) represent propagation intensity, information entropy compensation, and ontological value, respectively. This indicates the trend of the event's importance changing over time; coefficients α, β, and γ are used to balance the contribution of each component to the total value score.

[0012] Furthermore, the propagation intensity I(t) is expressed by the following formula:

[0013] In the formula, D(t) represents the time decay function of the information's influence; when decaying exponentially... Or power-law decay δ represents the exponential decay coefficient of information's influence; ε represents the power-law decay coefficient of information's influence; t represents time. This represents the weighted sum of the distribution volume across different channels.

[0014] N represents the total number of information dissemination channels; This represents the weight of the i-th channel; This represents the propagation volume of the i-th channel at time t; It represents the propagation entropy, used to measure the uniformity of information distribution across channels;

[0015] in, This represents the proportion of propagation through the i-th channel at time t. , This represents the propagation volume of the j-th channel at time t.

[0016] Furthermore, the formula for the information entropy compensation H(t) is as follows: Reciprocal form: Or logarithmic form:

[0017] Indicates the current event E t Average similarity to historical events;

[0018] Sim(·) represents the event similarity calculation. This indicates the elimination of the zero constant.

[0019] Furthermore, the ontological value S(t) is expressed by the following formula:

[0020] In the formula, This represents the weight of the evaluation dimension for the k-th event; ; represents the standardized value of the k-th event evaluation dimension; M represents the total number of event evaluation dimensions; evaluation dimensions include: status of the person, economic impact, social attention, geographical scope of impact, and time type.

[0021] Furthermore, step S30 also includes: Obtain the real-time public opinion factor P(t) associated with the triplet data, and collect event attention indicators through the social media API interface; The model weight coefficients are dynamically adjusted based on the public opinion factor P(t): When P(t) > θ hot At that time, increase the propagation intensity weight α; When P(t) < θ cold At the same time, increase the information entropy compensation weight β; Where, θ hot For a high thermal threshold, θ cold It has a low heat threshold.

[0022] Further, step S40 includes: Construct a parameter sandbox to visualize the trajectories of α, β, and γ weights, showcasing the historical adjustment paths of the model weight coefficients and real-time changes in the score; Adjust any weight parameter based on natural language rules, compile it into machine constraints, and support reverse parsing; Record the logical basis and weight contribution ratio of each score adjustment to optimize the dynamic evaluation model.

[0023] Secondly, embodiments of the present invention also provide a news image value assessment system based on predicate logic and multimodal fusion, using the news image value assessment method based on predicate logic and multimodal fusion as described in any of the first aspects, the system comprising: The multimodal acquisition module is used to acquire multimodal data of news images, including associated audio, visual and text data, and uses an adaptive triggering mechanism to extract key frames of audio and visual data; The predicate logic parsing engine module is used to perform semantic parsing on keyframes through the predicate logic parsing engine, and generate subject-action-scene triples by combining text data, and verify them through both spatial location and semantic association; The news value scoring module is used to calculate the news value score based on the verified triplet data and a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation and ontological value. The human-machine collaborative optimization module is used to optimize the dynamic evaluation model by visually tuning the evaluation parameters and rules.

[0024] As can be seen from the above technical solution, compared with the prior art, the present invention has the following technical advantages: 1. Multimodal semantic fusion improves accuracy: By jointly analyzing multimodal data such as visual, text, and audio data and using predicate logic modeling to achieve deep semantic alignment, the error caused by inconsistency of information from different modalities is effectively reduced.

[0025] 2. Dynamic Adaptive Assessment: By introducing a dynamic weight network that incorporates public opinion factors and information entropy compensation, the system can automatically adjust assessment parameters as real-time events develop, enabling rapid response to hot events and significantly shortening the response delay to public opinion changes, which is significantly better than the traditional fixed model.

[0026] 3. The decision-making process is interpretable: Employing a hybrid reasoning architecture combining symbolic logic and deep learning, the system can provide a clear decision-making path (e.g., through visualization of triples and weight contributions), support rule-level intervention, and ensure the traceability of evaluation results. Compared to pure black-box models, this system reduces the workload of manual result verification and improves decision-making transparency.

[0027] 4. Improved assessment accuracy and efficiency: The introduction of human-machine collaborative optimization has improved the efficiency of manual review and greatly reduced labor costs.

[0028] 5. Wide range of applications: This invention can efficiently screen and classify massive amounts of UGC content, helping media organizations to discover high-value news leads in a timely manner. It can also be applied to scenarios such as content review on social media platforms and public opinion monitoring and early warning, and has broad application prospects and social value in the fields of news dissemination and content management. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0030] Figure 1 The flowchart of the news image value assessment method based on predicate logic and multimodal fusion provided by the present invention is shown.

[0031] Figure 2 The working principle diagram of the predicate extraction logic provided by this invention.

[0032] Figure 3 The framework diagram of the news image value assessment system based on predicate logic and multimodal fusion provided by this invention is shown. Detailed Implementation

[0033] 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 embodiments of the present invention, and not all embodiments. 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.

[0034] Example 1: This invention provides a news image value assessment method based on predicate logic and multimodal fusion. Through an innovative predicate logic parsing engine and dynamic weight evaluation model, it achieves intelligent evaluation of news photos and videos. (Refer to...) Figure 1 As shown, the specific steps include S10 to S40: S10. Acquire multimodal data of news images, including associated audio, visual and text data, and extract key frames of audio and visual data using an adaptive triggering mechanism.

[0035] This step is responsible for acquiring and preprocessing news video data, and can provide efficient hierarchical storage to support subsequent analysis. Data sources include user-uploaded video streams, images, and related text descriptions and audio. Data can be collected through multiple access interfaces: for example, providing an API interface that conforms to RESTful specifications, supporting the uploading of necessary metadata fields (such as user_id, timestamp, geo_location, etc.) in JSON / XML format; for real-time video streams, the RTMP protocol is used for access and H.264 encoding is used (ensuring a resolution of no less than 720p and a bitrate of approximately 2 Mbps) to ensure high-quality transmission.

[0036] For example, the captured video first undergoes dynamic keyframe extraction processing, which can employ a dual-threshold triggering mechanism, combining image HSV histogram differences and audio energy mutations to detect changes in video content. When the color histogram difference between adjacent frames exceeds a set threshold or is accompanied by an abnormal peak in the detected audio signal, it is determined that a scene change or important event has occurred, and that frame is extracted as a keyframe.

[0037] Example: Keyframe extraction based on HSV histogram differences and audio energy abrupt changes: Step (1): Preprocessing of video and audio: Video frame extraction: Decompose the video into a sequence of frames, typically at a rate of 25 frames per second (25fps).

[0038] Audio signal extraction: Extract audio signals from video and perform preprocessing such as noise reduction and normalization.

[0039] Step (2): Calculate the HSV histogram difference: HSV Histogram Calculation: For each frame of the image, calculate its HSV color space histogram.

[0040] Histogram difference calculation: Calculate the HSV histogram difference between adjacent frames. Common methods include Bhattacharyya distance or histogram cross-entropy.

[0041] Step (3): Calculate the audio energy change: Short-time energy calculation: Divide the audio signal into frames and calculate the short-time energy of each frame.

[0042] Energy mutation detection: Detects energy changes between adjacent audio frames and identifies energy mutation points.

[0043] Step (4): Dual threshold triggering mechanism Set threshold: Image histogram difference threshold (e.g., 0.5).

[0044] Audio energy change threshold (e.g., 10 dB).

[0045] Triggering conditions: When the difference between the HSV histograms of adjacent frames exceeds a set threshold, or the change in audio energy exceeds a set threshold, it is determined to be a scene switch or an important event, and the current frame is extracted as a key frame.

[0046] For example, in addition to histograms and energy thresholds, more advanced triggering conditions can be introduced, such as optical flow change detection and semantic segmentation differences: by calculating the optical flow field between the current frame and the previous keyframe, the motion amplitude can be evaluated; or semantic segmentation of the frame can be performed to compare changes in the object region before and after. When the motion vector amplitude or semantic region difference exceeds a predetermined threshold, keyframe extraction is also triggered. This multi-condition fusion dynamic frame extraction mechanism improves the sensitivity and accuracy of keyframe extraction, and can adapt to the frame selection needs in scenes of rapid motion and semantic abrupt changes.

[0047] Frame extraction strategies are applied to video data to extract keyframes, and further, a spatiotemporal correlation coding structure between frames can be generated. Specifically, this involves organizing the extracted keyframes into a graph structure in chronological order, ensuring that each frame retains image content features while recording temporal adjacency relationships between frames through graph nodes and edges. For example, using keyframes as nodes, directed edges are established between adjacent frame nodes in a temporal order; edge weights can be assigned based on inter-frame similarity (e.g., calculated using optical flow average displacement or structural similarity SSIM), thus forming a temporal graph. This graph structure can further integrate spatial relationships: if multiple objects are detected within a frame, these objects can be treated as child nodes, connected by additional edges to represent spatial co-occurrence relationships. The final output spatiotemporally coded data contains both visual content features and explicitly depicts the temporal progression of events.

[0048] While extracting keyframes from the video, information such as audio segments and text descriptions corresponding to the timestamp of each keyframe can also be obtained and appended to the frame node to form a multimodal joint representation. In this way, when analyzing a keyframe later, its related audio features (such as Mel-frequency cepstral coefficients (MFCC) and speech-to-text content) and text descriptions can be easily accessed, thus laying the foundation for multimodal fusion analysis.

[0049] Furthermore, in practical implementation, a tiered storage architecture can be adopted to improve efficiency, for example, for data of different values ​​and access frequencies: recently accessed high-frequency data can be stored in a high-speed cache layer (such as a Redis cluster, using an LRU strategy to manage memory for fast retrieval of the latest content), historical data can be stored in a distributed object storage (such as a MinIO cluster) and combined with search engine indexes (such as Elasticsearch) for retrieval of massive amounts of content; when necessary, structured metadata can also be stored in a graph database (such as Neo4j) to associate event relationships. Through the above collection and storage mechanisms, well-structured multimodal data with good spatiotemporal correlation can be provided to support subsequent value assessment.

[0050] S20. The keyframes are semantically parsed through the predicate logic parsing engine, and the subject-action-scene triples are generated by combining the text data, and then verified by both spatial location and semantic association.

[0051] In this step, semantic parsing is performed on the input image content to extract "subject-action-scene" predicate triples that can represent news events. (See reference...) Figure 2 The diagram shows the working process of the predicate extraction logic, which includes: 1. First, subject detection is performed, which involves using computer vision object detection algorithms to locate the main entities (such as people, vehicles, etc.) in the image. Preferably, a pre-trained object detection model (such as YOLO or ResNet-CNN model) is used to obtain preliminary detection results, and then the non-maximum suppression (NMS) algorithm is used to filter overlapping bounding boxes to ensure that only the most relevant subject targets are retained.

[0052] 2. Next, action recognition is performed: For the detected subject region, a video action recognition model is used to analyze its behavior category. During training, public datasets (such as COCO, ImageNet videos, etc.) and a self-built news event action sample library are combined. A deep convolutional neural network (such as ResNet-50) is used to extract spatial features, and a temporal attention mechanism (such as introducing a spatiotemporal attention module) is fused to improve the robustness of action recognition for consecutive frames, thereby accurately determining the subject's behavior (such as "running", "talking", "collision", etc.).

[0053] 3. Then, perform scene semantic analysis: extract environmental and scene information from the images through image classification or semantic segmentation techniques (such as "fire scene", "sports stadium" or "everyday street"), or combine the text description of the images to enrich the scene semantics.

[0054] 4. Finally, the above results are integrated to generate a predicate triple representation, namely, subject-action-scene. To ensure the validity and accuracy of this triple, this module introduces a dual spatial and semantic verification mechanism at the symbolic logic layer: on the one hand, it uses spatial positional relationships to verify the consistency between the subject and the scene (e.g., requiring the detected subject bounding box to have sufficient overlap within the scene region, with an IoU value of 0.6 or higher); on the other hand, it calculates semantic matching degree (e.g., using the CLIP multimodal model to calculate the similarity between image content and corresponding text description, with a threshold of no less than 0.7) to verify the reasonable matching of action and scene. If a candidate triple fails the above spatial or semantic threshold detection, it is eliminated or marked as uncertain. Only when both spatial relationship and semantic relevance meet the threshold conditions does the triple pass verification. This step ensures that the generated predicate triple is accurate and conforms to the actual context, avoiding errors in subject-action pairing. Through this predicate logic parsing engine, the structured semantic information contained in news images can be extracted, providing highly interpretable basic features for subsequent value assessment.

[0055] S30. Calculate the news value score based on the verified triplet data using a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation, and ontological value. The three-dimensional dynamic evaluation model comprehensively considers dimensions such as dissemination intensity, information entropy compensation, and intrinsic value to quantitatively evaluate the value of news images. The three-dimensional dynamic evaluation model is represented as follows: (1) In the formula, I(t), H(t), and S(t) represent propagation intensity, information entropy compensation, and ontological value, respectively. This indicates the trend of the event's importance changing over time; coefficients α, β, and γ are used to balance the contribution of each component to the total value score.

[0056] 1) I(t) represents the propagation intensity, reflecting the breadth and speed of news spread through various channels (which can be calculated from channel weight, propagation entropy and time decay function). The propagation intensity I(t) is commonly used to measure the breadth and speed of news or information dissemination across various channels. Its calculation can comprehensively consider factors such as channel weight, propagation entropy, and time decay function. The specific calculation methods for these factors are as follows: 1.1) Channel Weight: Different communication channels (such as social media, news websites, and television) have varying degrees of influence in information dissemination. To quantify this difference, a weight can be assigned to each channel. These weights reflect the importance of the message within the overall communication strategy. These weights can be determined based on methods such as historical data, audience reach, or expert evaluation.

[0057] The contribution of channel weight to dissemination intensity can be expressed as: (2) N represents the total number of information dissemination channels; This represents the weight of the i-th channel; This represents the amount of information disseminated through the i-th channel at time t (such as the number of reposts, reads, etc.). It represents the propagation entropy, which is used to measure the uniformity of information distribution across channels.

[0058] 1.2) Propagation Entropy: Propagation entropy measures the uniformity of information distribution across channels. A higher entropy value indicates a more even distribution of information across multiple channels and a wider propagation range. It can be calculated using Shannon entropy. (3) in: This represents the propagation proportion of the i-th channel at time t.

[0059] (4) This represents the propagation volume of the j-th channel at time t.

[0060] 1.3) Time Decay Function: The influence of information typically diminishes over time. To reflect this, a time decay function D(t) can be introduced. Commonly used time decay functions include exponential decay and power-law decay: a. Exponential decay: (5) in >0 is the attenuation coefficient, which controls the attenuation rate.

[0061] b. Power-law decay: (6) Where ε>0 is the decay exponent.

[0062] The choice of which decay function to use depends on the specific application scenario and data characteristics.

[0063] Taking all the above factors into account, the propagation intensity I(t) can be expressed as: (7) It represents the propagation entropy, which is used to measure the uniformity of information distribution across channels.

[0064] 2) H(t) represents information entropy compensation, which measures the rarity or uncertainty of event information (it can be represented by the inverse of event similarity; when events are too common, the entropy value is high and a deduction is made to compensate). In information propagation models, the information entropy compensation term H(t) measures the scarcity or uncertainty of event information. Its calculation is typically based on the reciprocal of event similarity, as follows: 2.1) Event Similarity Calculation First, define the current event E. t With the set of historical events {E1, E2, ..., E n The similarity between}. Similarity Sim(E t E i It can be calculated in various ways, for example: Semantic similarity: Cosine similarity between event descriptions is calculated using word vector models (such as Word2Vec and BERT).

[0065] Keyword matching: Calculate Jaccard similarity based on the set of keywords for an event.

[0066] Content features: Considering attributes such as time, location, and participants of the event, calculate the similarity of structured data.

[0067] After obtaining all similarities, calculate the average similarity: (8) Sim(·) represents the calculation of event similarity.

[0068] 2.2) Information Entropy Compensation Calculation The information entropy compensation term H(t) reflects the rarity or uncertainty of an event. When an event has a high similarity to historical events, it indicates that the event is relatively common, and the entropy value should be reduced; conversely, the entropy value should be increased. It can be calculated in the following way: Reciprocal form: (9) in, This indicates that it is a small constant to prevent the denominator from being zero.

[0069] Logarithmic form: (10) This form emphasizes that low similarity (i.e., rare events) corresponds to higher entropy values.

[0070] 2.3) Example Calculation Assume the average similarity between the current event and historical events is . =0.2, let =0.01: Reciprocal form:

[0071] Logarithmic form:

[0072] This indicates that the event is relatively rare and has a high information entropy compensation value.

[0073] Using the above method, the information entropy compensation term H(t) can effectively reflect the rarity or uncertainty of an event, thereby adjusting the propagation intensity of the event in the propagation model and improving the accuracy and robustness of the model.

[0074] 3) S(t) represents the intrinsic value or static importance of an event (calculated by combining multiple dimensions of characteristics, such as the status of the people involved, economic impact, and social attention). The intrinsic value or static importance of an event, S(t), is an indicator that measures the inherent influence and importance of an event before it is disseminated. To quantify S(t), it can be comprehensively evaluated from multiple dimensions, including the status of the people involved, economic impact, and social attention.

[0075] The following is the specific calculation process: 3.1) Definition of Dimensional Indicators S(t) is decomposed into several key dimensions, and quantifiable metrics are set for each dimension:

[0076] 3.2) Indicator Quantification and Standardization Each indicator is quantified and standardized to ensure its value falls within a uniform scale (e.g., 0 to 1). Standardization methods: Use min-max normalization or Z-score normalization.

[0077]

[0078] Example: If there are 500 news reports related to a certain event, and the historical data ranges from a minimum of 100 to a maximum of 1000, then the standardized value is:

[0079] 3.3) Weighting and Overall Scoring Assign weights to each dimension This reflects its importance in the overall evaluation. Weights can be determined using methods such as expert scoring and the Analytic Hierarchy Process (AHP).

[0080] For example, based on historical data and expert knowledge, basic weights are assigned to elements of news events, and the news value score of the video content is calculated by combining real-time factors. First, a basic weight table is constructed: initial weights are determined jointly by domain experts and large-scale historical news data statistics for different categories of events. For instance, high-impact actions with a significant impact on news value (such as "explosion," "collapse," "collision," etc.) are assigned higher weights (e.g., 0.8–0.9), social event actions (such as "assemblies," "strikes," "contract signings," etc.) are assigned medium weights (approximately 0.5–0.7), and ordinary daily actions (such as "walking," "talking," "standing," etc.) are assigned lower weights (approximately 0.3). These weights reflect the differences in the amount of information (self-information) and social attention contained in different actions. Through statistical analysis of historical news events, the above weight table can be optimized to better align with the objective laws of news value.

[0081] The formula for calculating the overall score S(t) is: (11) In the formula, This represents the weight of the evaluation dimension for the k-th event; ; denoted by , represents the standardized value of the k-th event evaluation dimension; M represents the total number of event evaluation dimensions; the evaluation dimensions are abstracted and analyzed from the mainstream academic definition of news value, including but not limited to: the status of the person, economic impact, social attention, geographical scope of influence, and time type, etc.

[0082] 4) Example Calculation Suppose the standardized metric values ​​and weights for a certain event are as follows:

[0083] The overall score is: S(t)=(0.3×0.8)+(0.25×0.6)+(0.2×0.7)+(0.15×0.5)+(0.1×0.9)=0.695 Therefore, the ontological value score of this event is 0.695.

[0084] The above methods can be used to systematically assess the intrinsic value of an event, providing a quantitative basis for news value assessment and resource allocation.

[0085] By using the above model or similar mathematical expressions (such as other differential equations or neural network approximations), the propagation / entropy information from the basic weight table (corresponding to the S dimension) and the real-time calculated propagation / entropy information (corresponding to the I and H dimensions) are fused together to calculate the comprehensive news value score of the current video content.

[0086] Furthermore, step S30 also includes: obtaining the real-time public opinion factor P(t) associated with the triplet data, and collecting event attention indicators through the social media API interface; The model weight coefficients are dynamically adjusted based on the public opinion factor P(t): When P(t) > θ hot At that time, increase the propagation intensity weight α; When P(t) < θ cold At the same time, increase the information entropy compensation weight β; Where, θ hot For a high thermal threshold, θ cold It has a low heat threshold.

[0087] As support for dynamic weight calculation, it provides real-time public opinion factor acquisition and event background knowledge. First, a news event database (historical news value map) is constructed: a large amount of historical news event data is collected, and event clustering analysis is performed using natural language processing and clustering algorithms. For example, the pre-trained language model BERT is used to extract news text features, and combined with the OPTICS density clustering algorithm, similar news is clustered into one category, forming event clusters. For each event cluster, its information entropy is calculated: if a certain type of event contains too much redundant or common information, the entropy value is high; conversely, rare events with a large amount of new information have a low entropy value (indicating high unique value). This information entropy can serve as a compensation coefficient for news value—when the event involved in a video content belongs to a high-entropy (common) category, its score is appropriately reduced; conversely, for low-entropy (rare) events, the score can be increased, encouraging the discovery of unique news. Second, real-time public opinion factors are acquired: attention indicators related to the current event are obtained through public data interfaces of social media and online platforms (such as Weibo hot search index, Baidu Index API, etc.). These real-time public opinion data are transformed into quantitative factors, such as "public opinion heat value". By combining public opinion heat with basic weights and information entropy, the weight network is dynamically adjusted: when an event involved in a news video generates strong attention in real-time public opinion, the corresponding evaluation weight is increased; for events with low real-time heat but high basic value, the influence of both is balanced to avoid underestimating important but obscure news. Through the combination of an event database and public opinion factors, the evaluation model of this invention can achieve sensitivity to time evolution: it both refers to historical patterns and responds promptly to current dynamics, making news value scoring more objective and real-time.

[0088] S40. The dynamic evaluation model is optimized by visually tuning the evaluation parameters and rules.

[0089] In practical implementation, interactive interfaces and mechanisms can be provided to integrate human experience and feedback into the model, continuously optimizing the evaluation results.

[0090] First, a parameter visualization sandbox is designed: the effects of key parameters in the model (such as the weights of each indicator α, β, γ, the basic weight of actions, the public opinion weighting coefficient, etc.) are displayed as trajectory graphs or heatmaps through a visual interface, intuitively showing the trajectory of the impact of parameter adjustments on the scoring results. Users can drag sliders in the sandbox interface to adjust parameter values ​​and observe the changing trends of news value scores, thereby intuitively understanding the basis of the model's decisions. This parameter sandbox mechanism ensures that the parameter adjustment process is reversible and transparent: each adjustment is recorded as a parameter evolution trajectory, which can be rolled back or repeated at any time.

[0091] Secondly, the system implements reversible rule compilation: it allows users to write evaluation rules in natural language (e.g., "If the event category is disaster, increase the weight of related actions to above 0.8"), dynamically compiling these rules into machine-executable constraints for application to the model; conversely, it can also extract the implicit decision logic within the current model into readable rule descriptions for user review and modification. This bidirectional programmability ensures a deep integration of human experience and machine models. Through this human-machine collaboration in step S40, the news image value assessment method forms a closed-loop optimization: human feedback continuously corrects model biases, and the model's interpretable output guides human decision-making, achieving a virtuous cycle of automatic evaluation and human supervision, thus improving the system's robustness and credibility.

[0092] After the human-computer interaction is completed, the actual model update and effect tracking can be performed. First, based on the parameter adjustments confirmed by the user, the corresponding values ​​are sent to the value assessment module to dynamically update α, β, γ, or relevant thresholds; simultaneously, the model's decision logic is updated according to the newly added rules. Since the model is continuously running, feedback control needs to ensure a smooth update process: a gradual update can be adopted, for example, the new parameters linearly replace the old parameters over several minutes, rather than an instantaneous jump, and the triggering of new rules is paused during this period, and the rules are activated again after the parameters are in place, to avoid uncertainty caused by both taking effect simultaneously. Next, the model output is monitored, and the changes before and after the adjustment are compared. If there is a significant deviation or anomaly, a warning will be recorded or even the changes will be automatically rolled back (for example, if a new rule causes all scores to drop to 0, the rule will be immediately deactivated and an alarm will be triggered). In this step, the results of manual adjustments are also permanently saved, for example, by writing them to a log database, including the adjustment content, the adjuster, the time, etc., for auditing and subsequent analysis.

[0093] Example 2: Based on the same inventive concept, this invention also provides a news image value assessment system based on predicate logic and multimodal fusion. Since the principle of the problem solved by this system is similar to the aforementioned news image value assessment method based on predicate logic and multimodal fusion, the implementation of this system can refer to the implementation of the aforementioned method, and the repeated parts will not be described again.

[0094] Reference Figure 3 As shown, the overall system architecture includes: The multimodal acquisition module is used to acquire multimodal data of news images, including associated audio, visual and text data, and uses an adaptive triggering mechanism to extract key frames of audio and visual data; The predicate logic parsing engine module is used to perform semantic parsing on keyframes through the predicate logic parsing engine, and generate subject-action-scene triples by combining text data, and verify them through both spatial location and semantic association; The news value scoring module is used to calculate the news value score based on the verified triplet data and a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation and ontological value. The human-machine collaborative optimization module is used to optimize the dynamic evaluation model by visually tuning the evaluation parameters and rules.

[0095] The various modules of the system work together to perform semantic analysis, value calculation, and feedback optimization on the input image content.

[0096] Its overall architecture can be divided into three layers: the data layer, the computation layer, and the decision layer. The data layer is responsible for the acquisition and hierarchical storage of multimodal image data, providing structured spatiotemporal encoded data; the computation layer includes historical news value maps and dynamic weight evaluation models, realizing the core calculation of image value; the decision layer uses a human-machine collaboration module to perform interpretable tuning and feedback optimization of the model, forming a closed-loop adaptive improvement mechanism. The collaborative relationship between the modules at each layer is shown below: The multimodal acquisition module preprocesses the real-time acquired raw video and audio streams into spatiotemporally encoded data and stores them in a hierarchical storage medium; the historical news value graph module uses a large amount of historical event data to calculate the benchmark reference for news value assessment (including background data in dimensions such as dissemination intensity, information entropy, and ontological value); the predicate logic parsing engine module and the news value scoring module extract semantic triples from real-time image frames and combine them with historical graphs and real-time public opinion to calculate value scores; the human-machine collaborative optimization module obtains human feedback (adjusting weights, editing rules, etc.) to generate optimized model parameters, which in turn affect the three-dimensional dynamic evaluation model, thereby enabling the system to continuously evolve and adapt to environmental changes.

[0097] Among them, the tiered storage architecture adopts a hot / warm / cold tiered storage strategy to improve read and write efficiency. Hot data layer: Deploys a high-speed caching database (such as a Redis in-memory database cluster) to temporarily store keyframe data and metadata from the most recent period. The hot layer employs eviction policies such as LRU to ensure that the latest data is quickly available.

[0098] Warm data layer: Deploy a distributed NoSQL database (such as Cassandra or MongoDB) to store data in the medium term and support efficient querying by event or time index.

[0099] Cold data layer: Deploy massive object storage (such as MinIO clusters or HDFS) combined with offline indexes (Elasticsearch, etc.) to store historical data from long ago and large video clips, image files, etc. Cold data is archived through periodic batch processing to provide a basis for future reference.

[0100] Historical News Value Graph Construction: A large amount of historical news images and related event data are collected and organized into a graph database using knowledge graph technology. Nodes in the graph represent news events or elements (people, locations, event themes, etc.), and edges represent relationships between events (e.g., similar themes, chronological causality, dissemination paths, etc.). To obtain the embedding representation and clustering features of the graph, the TransR algorithm is used to vectorize event relationships; simultaneously, the OPTICS clustering algorithm is used to discover high-density clusters of similar events in the embedding space. From this, the rarity and aggregation of events can be calculated: if an event falls into a dense cluster, it indicates that the event type is common; conversely, if it is in a sparse region, it is a rare event. The graph records the historical dissemination intensity (e.g., the dissemination range and popularity indicators on various media platforms at the time), information entropy (e.g., the frequency and uncertainty of event occurrence), and ontological value (the importance of the event itself, such as casualties, social impact level, etc.) of each type of event. These statistics will serve as a reference benchmark for subsequent models.

[0101] The system of this invention consists of a complete automatic evaluation and adaptive improvement cycle, from multimodal data acquisition and preprocessing to dynamic model calculation, then to semantic parsing and scoring, and finally continuous optimization through human-machine collaboration.

[0102] Specifically, this invention employs predicate logic parsing based on a "subject-action-scene" triple to perform deep semantic analysis of news video content, extracting core elements from news events. It combines a constructed historical news value map and information entropy model, introducing public opinion factors to dynamically weight the evaluation process. Furthermore, it optimizes model parameters through human-machine collaboration to ensure the accuracy and interpretability of the evaluation results. This system can automatically identify key figures (subjects), behaviors (actions), and environmental backgrounds (scenes) in news videos, and adjust evaluation strategies based on real-time changes in public opinion, thereby providing a timely and reasonable assessment of the news value of the video content.

[0103] Through the coordinated operation of the above modules, the system of this invention has achieved significant technical results in the evaluation of the value of news images. The system can efficiently process massive amounts of multimodal data, greatly shortening the response time to emergencies. Due to the introduction of interpretable logical parsing and human-computer interaction optimization mechanisms, the number of manual review interventions has decreased, improving overall review efficiency. At the same time, the evaluation results are more transparent and reliable; the system can output key triples and weight information supporting its score, making it easier for users to understand and trust the model's decision-making.

[0104] This invention has broad application scenarios and practical value. Firstly, for news media organizations, the system can automatically filter user-uploaded news photos and videos, quickly identifying high-news-value clues from massive amounts of material, improving editing efficiency and reducing the risk of missing important news. Secondly, on social media and content community platforms, this invention can serve as a content recommendation or review tool, classifying UGC video content according to news value, thereby prioritizing the promotion of significant events and suppressing the spread of low-value information that may disrupt public opinion. Thirdly, in the fields of government supervision and public opinion monitoring, the system can help relevant departments promptly capture online public opinion hotspots, assisting in judging the severity and potential impact of events through value assessment of citizen-uploaded event videos, providing data support for emergency response and decision-making. Finally, this invention can also be extended to media big data analysis, intelligent archive management, and other areas, empowering more diverse media application scenarios through the quantification of the value of video materials. This invention not only improves the accuracy and efficiency of assessment but also provides an important tool for the intelligent management and dissemination of media content, and is expected to generate significant social and economic benefits.

[0105] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0106] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A news image value assessment method based on predicate logic and multimodal fusion, characterized in that, Includes the following steps: S10. Acquire multimodal data of news images, including: associated audio, visual and text data, and extract key frames of audio and visual data using an adaptive triggering mechanism; S20. Semantic parsing of keyframes is performed through the predicate logic parsing engine, and subject-action-scene triples are generated by combining text data, and verified by both spatial location and semantic association. S30. Calculate the news value score based on the verified triplet data using a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation, and ontological value. S40. The dynamic evaluation model is optimized by visually tuning the evaluation parameters and rules.

2. The news image value assessment method based on predicate logic and multimodal fusion according to claim 1, characterized in that, In step S10, the adaptive triggering mechanism extracts key frames of audiovisual data, including: For visual data feature change detection, sudden changes in the image are analyzed by the color histogram difference between adjacent frames, triggering keyframe extraction; For audio data feature change detection, short-time energy analysis is used to detect energy changes between adjacent audio frames, triggering keyframe extraction.

3. The news image value assessment method based on predicate logic and multimodal fusion according to claim 1, characterized in that, Step S20 includes: S201. Use a pre-trained target detection model and a non-maximum suppression strategy to filter redundant detection boxes and perform subject detection. S202. Extract the subject motion features using the image sequence optical flow method, feed the visual features into the trained action classification model, and integrate the spatiotemporal attention mechanism to capture the subject action. S203. Use a pre-trained CNN model to determine the semantic label of the entire frame; at the same time, perform instance segmentation on the background region in the frame, extract scene elements, and generate a brief semantic description of the scene. S204. Spatial location verification is performed by calculating whether the overlap between the subject and the scene in space exceeds the first threshold; and semantic association verification is performed by using cross-modal semantic analysis to calculate whether the CLIP similarity between the triplet text description and the keyframe image is greater than the second threshold.

4. The news image value assessment method based on predicate logic and multimodal fusion according to claim 1, characterized in that, In step S30, the three-dimensional dynamic evaluation model is represented as follows: In the formula, I(t), H(t), and S(t) represent propagation intensity, information entropy compensation, and ontological value, respectively. This indicates the trend of the event's importance changing over time; coefficients α, β, and γ are used to balance the contribution of each component to the total value score.

5. The news image value assessment method based on predicate logic and multimodal fusion according to claim 4, characterized in that, The propagation intensity I(t) is expressed by the following formula: In the formula, D(t) represents the time decay function of the information's influence; when decaying exponentially... Or power-law decay δ represents the information's influence decay coefficient. ε represents the power-law decay coefficient of the information's influence; t represents time. This represents the weighted sum of the distribution volume across different channels. N represents the total number of information dissemination channels; This represents the weight of the i-th channel; This represents the propagation volume of the i-th channel at time t; It represents the propagation entropy, used to measure the uniformity of information distribution across channels; in, This represents the proportion of propagation through the i-th channel at time t. , This represents the propagation volume of the j-th channel at time t.

6. The news image value assessment method based on predicate logic and multimodal fusion according to claim 4, characterized in that, The formula for the information entropy compensation H(t) is as follows: Reciprocal form: Or logarithmic form: Indicates the current event E t Average similarity to historical events; Sim(·) represents the event similarity calculation. This indicates the elimination of the zero constant.

7. The news image value assessment method based on predicate logic and multimodal fusion according to claim 4, characterized in that, The ontological value S(t) is expressed by the following formula: In the formula, This represents the weight of the evaluation dimension for the k-th event; ; This represents the standardized value of the evaluation dimension of the k-th event; M represents the total number of dimensions for evaluating the event; the evaluation dimensions include: the status of the person, economic impact, social attention, geographical scope of impact, and time type.

8. The news image value assessment method based on predicate logic and multimodal fusion according to claim 4, characterized in that, Step S30 further includes: Obtain the real-time public opinion factor P(t) associated with the triplet data, and collect event attention indicators through the social media API interface; The model weight coefficients are dynamically adjusted based on the public opinion factor P(t): When P(t) > θ hot At that time, increase the propagation intensity weight α; When P(t) < θ cold At the same time, increase the information entropy compensation weight β; Where, θ hot For a high thermal threshold, θ cold It has a low heat threshold.

9. The news image value assessment method based on predicate logic and multimodal fusion according to claim 4, characterized in that, Step S40 includes: Construct a parameter sandbox to visualize the trajectories of α, β, and γ weights, showcasing the historical adjustment paths of the model weight coefficients and real-time changes in the score; Adjust any weight parameter based on natural language rules, compile it into machine constraints, and support reverse parsing; Record the logical basis and weight contribution ratio of each score adjustment to optimize the dynamic evaluation model.

10. A news image value assessment system based on predicate logic and multimodal fusion, characterized in that, Using the news image value assessment method based on predicate logic and multimodal fusion as described in any one of claims 1-9, the system comprises: The multimodal acquisition module is used to acquire multimodal data of news images, including associated audio, visual and text data, and uses an adaptive triggering mechanism to extract key frames of audio and visual data; The predicate logic parsing engine module is used to perform semantic parsing on keyframes through the predicate logic parsing engine, and generate subject-action-scene triples by combining text data, and verify them through both spatial location and semantic association; The news value scoring module is used to calculate the news value score based on the verified triplet data and a three-dimensional dynamic evaluation model; the three-dimensional dynamic evaluation model integrates dissemination intensity, information entropy compensation and ontological value. The human-machine collaborative optimization module is used to optimize the dynamic evaluation model by visually tuning the evaluation parameters and rules.

Images