Method and system for determining news transmission effect

Abstract

The invention discloses a method and system for determining a news transmission effect, and relates to the technical field of computers, and the method comprises the following steps: carrying out the multi-dimensional distortion degree detection of secondary creation content, and generating a distortion degree vector containing a semantic distortion degree, an emotional distortion degree and a logic distortion degree; according to the method, a dynamic cross-time-period propagation effect evaluation system is constructed through technologies such as multi-dimensional distortion degree detection, cognitive deviation quantification driven by causal inference, a dynamic knowledge graph and a space-time heterogeneous graph neural network, and the limitation that a traditional method only depends on propagation range statistics and short-term data is broken through; according to the method, cognitive influence quantification of content deformation and dynamic mining of long-tail propagation value are realized, news propagation effect evaluation is upgraded from flow statistics to quality-value drive, secondary creation content evaluation accuracy and long-tail propagation recognition sensitivity are remarkably improved, an objective and scientific evaluation tool is provided for news providers, and the news propagation effect evaluation method has good application prospects. And the propagation strategy is assisted to be optimized and healthy propagation ecology is guided.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for determining news dissemination effects. Background Art

[0002] In today's highly fragmented information landscape, evaluating news dissemination effectiveness is crucial for measuring news value. Traditional methods for evaluating news dissemination effectiveness rely primarily on formal statistics of dissemination reach (such as the number of users reached, number of reposts, and short-term exposure), as well as content deformation analysis based on text similarity (such as TF-IDF and BERT cosine distance). However, these methods only reflect the breadth of dissemination and superficial changes in content, failing to fully understand the actual impact of content deformation on audience perception during the dissemination process. They also struggle to capture the potential value of dissemination across time periods, resulting in significant technical gaps in practical application.

[0003] 1. Technical Blind Spots in Content Deformation Assessment: Existing technologies for evaluating secondary news creation (such as content truncation, adaptation, and commentary during forwarding) rely solely on formal statistics, ignoring multi-dimensional nonlinear distortion and its profound impact on audience cognition: The one-sidedness of distortion deconstruction: Traditional methods measure content distortion solely through text similarity, failing to effectively identify complex distortion types such as semantic attenuation (e.g., the removal of key entity relationships), sentiment shift (e.g., neutral content being recast as satire), and logical discontinuities (e.g., reposting content out of context). For example, if a news story about the "economic impact of a policy" is truncated to include "the policy causing price increases," existing technology cannot quantify the impact of missing key entity relationships, such as "policy implementation scope" and "price statistics period," on audience perception, resulting in evaluation results that deviate from the actual communication quality.

[0004] Lack of cognitive impact assessment: Traditional assessments assume that "dissemination equals effective reach" and fail to establish a closed loop of "content distortion → cognitive bias → behavioral feedback." For example, while joke-style adaptations may expand reach, they can reduce audience retention of core information and accuracy in factual judgment. Existing technologies, lacking in-depth behavioral data analysis covering user knowledge verification, opinion expression, and decision-making, are unable to quantify these cognitive biases, leading to misjudgment of the "high-repost, low-value" communication phenomenon.

[0005] 2. Technical gaps in mining the value of long-tail communication Traditional short-term statistics (such as dissemination data within 7 days) only weaken historical data through simple time decay factors (such as exponential decay functions), and cannot capture the time compounding effect of dissemination effectiveness: Ignoring the delayed trigger mechanism: Long-tail communication is essentially triggered by the semantic correlation between subsequent events (such as the introduction of policies and social hot spots) and historical news, but existing technologies are unable to identify the dynamic correlation between "events-news", resulting in historical news with potential value being underestimated due to the simple time factor, and its enabling effect on current communication cannot be quantified.

[0006] Static processing of cross-time communication networks: There are essential differences in the user groups and communication paths of initial communication and long-tail communication (for example, initial communication relies on the media matrix, while long-tail communication relies on user-generated content (UGC) secondary creation). Traditional static network models are unable to describe this cross-time structural evolution, resulting in the inability to accurately evaluate the communication effectiveness of phenomena such as "old news becoming popular again."

[0007] III. Essential Defects of Existing Technologies The core limitation of existing technologies is that they focus only on the "quantity" of communication (such as scope and speed), while ignoring the "quality" of communication (such as the impact of content distortion on cognition) and "time-varying characteristics" (such as the delayed value of long-tail communication). Specifically, Non-dynamic evaluation: Lack of multi-dimensional real-time deconstruction of content deformation during the communication process (semantic, emotional, and logical distortion), as well as quantification of cognitive impact based on causal inference, makes it impossible to achieve an accurate evaluation of "communication effect = scope × quality".

[0008] Short-term perspective: It relies on a static time decay model, fails to build an event-driven temporal attention mechanism and spatiotemporal heterogeneous graph neural network, and is unable to capture the correlation between historical news and current events and the dynamic changes of the cross-time communication network.

[0009] In view of this, a method and system for determining news dissemination effect are provided to overcome the above problems. Summary of the Invention

[0010] The purpose of the present invention is to provide a method and system for determining the news dissemination effect to solve the problems raised in the above background technology.

[0011] To solve the above technical problems, the present invention provides a method for determining news dissemination effect, comprising the following steps: Multi-dimensional distortion detection of secondary creative content is performed to generate a distortion vector including semantic distortion, emotional distortion, and logical distortion. Analyze the correlation between the distortion vector and the user's in-depth behavior data through a causal inference model to calculate the cognitive bias index; Build a dynamic knowledge graph to identify the entity association between historical news and current events, and dynamically adjust the weight of historical communication data based on the event correlation; The original dissemination range, distortion vector, cognitive bias index and long-tail dissemination weight are integrated to output a comprehensive score of news dissemination effect.

[0012] Furthermore, multi-dimensional distortion detection includes: Entity relationship extraction technology based on knowledge graphs detects semantic distortion and identifies missing entities or distorted relationships in secondary creations; A multimodal model that integrates BERT text sentiment analysis, emoji recognition, and the sentiment intensity of forwarded messages to detect sentiment distortion; Combine the logic rule engine with the deep learning model to detect logical distortions and identify logical fallacies such as taking things out of context or replacing concepts.

[0013] Furthermore, user in-depth behavior data includes: Knowledge verification behavior data, such as querying records of key entities through search engines; Opinion expression behavior data, such as semantic similarity analysis results for repeated or distorted core information in user comments; Decision-related behavior data, such as browsing logs of visits to relevant professional content pages.

[0014] Furthermore, the dynamic knowledge graph contains news entities, event entities, and their subject associations, time associations, and entity co-occurrence relationships, which are stored and updated in real time through a graph database.

[0015] A system for determining news dissemination effect, comprising: Distortion Analysis Module: This module is used to detect semantic, emotional, and logical distortion in secondary creative content, generate distortion vectors, and calculate the cognitive bias index through causal inference. Long-tail communication modeling module: used to build a dynamic knowledge graph to identify the relationship between historical news and current events, and calculate the cross-period communication effectiveness through spatiotemporal heterogeneous graph neural networks; Comprehensive evaluation module: used to integrate the original dissemination range, distortion vector, cognitive bias index and long-tail dissemination weight to output a comprehensive score of news dissemination effect.

[0016] Furthermore, the distortion analysis module includes: Multi-dimensional distortion detection submodule: This module includes a semantic distortion detection unit based on knowledge graph, a multimodal sentiment transfer model, and a logical distortion detection unit that integrates logic rules and deep learning. Cognitive bias calculation submodule: Contains a user behavior database and a causal inference engine, and analyzes the causal effect of distortion types on user behavior based on propensity score matching.

[0017] Furthermore, the long-tail propagation modeling module includes: Dynamic knowledge graph construction submodule: used to extract events in real time and update the knowledge graph containing news, events, entities and their relationships; Spatiotemporal heterogeneous graph neural network submodule: used to model the propagation process as a heterogeneous graph containing time slices, and analyze the structural differences of the propagation network in different time periods through graph neural networks.

[0018] Furthermore, it also includes a data collection module, which is used to obtain heterogeneous propagation data through multi-platform API interfaces, log collection systems and third-party data platforms, and unify the format, deduplicate and time align the data.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. Multi-dimensional distortion assessment improves transmission quality and accuracy: Comprehensive capture of nonlinear distortion: Using knowledge graph entity relationship extraction, multimodal sentiment analysis, and logic rule-deep learning fusion technology, it achieves the first multi-dimensional detection of semantic distortion (such as entity relationship separation), sentiment distortion (such as polarity reversal), and logical distortion (such as taking words out of context), solving the problem that traditional text similarity methods cannot identify complex distortion forms.

[0020] Causal quantification of cognitive bias: Propensity score matching (PSM) and causal inference models are introduced to establish a closed loop of "content distortion → cognitive bias → behavioral feedback". Through in-depth behavioral data such as user knowledge verification, opinion expression, and decision-making associations, the actual impact of distortion types on audience cognition is quantified.

[0021] 2. Dynamic mining and quantification of long-tail communication value: Event-driven delayed trigger identification: Build a dynamic knowledge graph and a BERT-based event classifier to capture the entity association between historical news and current events in real time (for example, the old news about "remote working technology" was triggered by the "epidemic rebound" event). Dynamically adjust the weight of historical data through event correlation scores to solve the problem of underestimation of the value of long-tail communication by traditional time decay models.

[0022] Dynamic modeling of cross-period communication networks: Using spatiotemporal heterogeneous graph neural networks and time slicing technology, initial communication (media-led) and long-tail communication (UGC-led) are modeled as dynamically evolving heterogeneous graphs to capture the structural differences of communication networks in different time periods (such as changes in node types and edge relationships).

[0023] 3. Technological progress brought about by systemic innovation: Objective quantification and dynamic evaluation system: The original communication range, distortion vector, cognitive bias index and long-tail communication weight are integrated into the communication quality score formula to form a three-dimensional evaluation system of "range × quality × time-varying value", breaking through the subjectivity of the traditional "self-evaluation" model and providing news providers with an objective and scientific communication effect measurement tool. Dynamic weight parameter and Through reinforcement learning optimization, it adapts to different communication scenarios (such as social media and professional media), making the evaluation model more suitable for practical applications.

[0024] Synergistic effect of technology integration: The combination of knowledge graph and causal inference realizes the quantification of the causal chain from content distortion to cognitive influence; the integration of spatiotemporal GNN and dynamic knowledge graph builds a dynamic model of cross-time communication network, generating a synergistic effect of "1+1>2".

[0025] 4. Industry application value: Scientific measurement of news value: This addresses the limitation of traditional methods that only focus on "dissemination scope". By evaluating distortion and cognitive bias, it accurately identifies "effective dissemination" and "ineffective dissemination", helping news providers optimize content strategies and increase the dissemination priority of high-value news.

[0026] Healthy guidance of the communication ecology: restrain low-quality communication such as "clickbait" and "quoting out of context", encourage in-depth communication based on facts, promote the transformation of news communication from "traffic-oriented" to "quality-oriented", and help build a healthier information communication ecology.

[0027] In summary, the present invention realizes the dynamic, multi-dimensional, and cross-time period accurate evaluation of news dissemination effects, filling the gaps in the existing technology in the cognitive impact of content deformation and the mining of long-tail dissemination value. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of a method and system for determining news dissemination effects according to the present invention. DETAILED DESCRIPTION

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0030] See also Figure 1 , the present invention provides a technical solution: See Figure 1 As shown, an embodiment of a method and system for determining news dissemination effect: 1. Method embodiment: 1. Multidimensional distortion modeling and cognitive bias assessment: Step 1: Multi-dimensional deconstruction of content deformation characteristics: Semantic Distortion Detection: This embodiment uses knowledge graph-based entity relationship extraction technology to construct a semantic network model of news content. Specifically, natural language processing (NLP) technology is used to identify entities in original news and secondary creative content, extracting entity information such as time, location, person, policy name, and economic data. Graph databases (such as Neo4j) are then used to construct relationships between entities, including causal, subordinate, and influential relationships. For example, in the case of a news item titled "The Economic Impact of a Certain Policy" being rewritten as "The Policy Leads to Price Rise," the system compares the completeness of entities and the accuracy of relationships in the knowledge graph to identify missing key entity relationships such as "Policy Implementation Scope" and "Price Statistics Period." The system then calculates a semantic distortion score and generates a semantic distortion vector, achieving a quantitative assessment of semantic attenuation.

[0031] Sentiment distortion detection: A multimodal sentiment transfer model is constructed, integrating BERT-based text sentiment analysis, emoji and modal particle recognition modules, and a forwarded message sentiment intensity analysis algorithm. In sentiment analysis of neutral technology news, the model extracts text sentiment features using BERT. This is combined with the sentiment polarity of emojis and the sentiment intensity of modal particles (for example, "even" enhances sentiment intensity), as well as the sentiment tendency of the forwarded message. This allows for accurate identification of sentiment distortion and improves the assessment of sentiment dimensions.

[0032] Logical distortion detection: This approach combines a logic rule engine with deep learning. Pre-set causal logic rules (e.g., A leading to B requires temporal and correlation conditions) and argumentation logic rules (e.g., arguments must be supported by evidence) are used. Furthermore, deep learning models (e.g., LSTM and Transformer) are used to analyze the logical structure of secondary creative content. When detecting logical fallacies such as quoting out of context (e.g., only extracting controversial points while ignoring the complete argumentation) or misinterpreting concepts (e.g., misinterpreting "industry regulation" as "government intervention"), the system classifies and quantifies the type of logical distortion based on pre-set rules and model analysis results, completing a logical distortion assessment.

[0033] Step 2: Causal inference-driven quantification of cognitive impact: User behavior data collection: This embodiment uses multiple channels to collect in-depth behavioral data on users after they encounter adapted content. Specifically, this includes: utilizing web crawler technology and app tracking to record user knowledge verification behaviors, such as whether they search for key entities within the content through search engines (e.g., Baidu Encyclopedia); using natural language processing to analyze semantic similarity in user comments to determine whether there is repetition or distortion of core information; and identifying decision-related behaviors based on user browsing logs, such as whether users visit relevant professional content pages, such as financial reports, after adapting financial news. This behavioral data provides objective evidence for subsequent causal analysis.

[0034] Propensity Score Matching (PSM): Using propensity score matching, based on user characteristics (such as age, gender, and browsing history), propensity scores are calculated for users who have been exposed to distorted content and those who have not, and then matched and grouped. Causal inference models (such as counterfactual causal models) are used to analyze the causal effects of different distortion types (semantic, emotional, and logical) on user behavior. For example, this analysis found that semantic distortion led to a 30% increase in user knowledge search activity but a 25% increase in opinion distortion. Based on this information, a cognitive bias index is generated, enabling quantitative analysis of the impact from dissemination data to audience perception.

[0035] Step 3: Weighting of the quality dimension of communication effect: Establish a communication effect evaluation formula: ; in: : It is a dynamic weight parameter, which is iteratively optimized based on historical propagation data and effect evaluation results through reinforcement learning algorithms such as deep Q network (DQN). Control the contribution of the original spread range to the final score, The degree of distortion and cognitive bias will determine the adjustment strength of the score. During the training process, the algorithm will dynamically adjust the score according to the difference between the actual communication effect and the expected target. , making the evaluation model more in line with the actual communication scenario.

[0036] Original Dissemination Scope: This is characterized by quantifiable dissemination scale indicators, including but not limited to the number of users reached, number of forwardings, and exposure. These indicators directly reflect the breadth of news coverage during the dissemination process and are core elements of traditional dissemination effectiveness evaluation.

[0037] (distortion vector): is a three-dimensional vector consisting of semantic distortion , emotional distortion ( ), logical distortion ( ) constitutes, that is Among them, the semantic distortion is calculated by comparing the entity relationships in the knowledge graph, the emotional distortion relies on the analysis of the multimodal emotional transfer model, and the logical distortion is determined by combining the logical rule engine with deep learning.

[0038] : represents the distortion vector The Euclidean norm of is calculated as: This value is used to measure the overall distortion degree after the secondary creation of news, and the value range is The closer the value is to 0, the less the content is distorted; the larger the value is, the more severe the distortion is.

[0039] Cognitive Bias Correction Coefficient: This coefficient is dynamically adjusted based on the Cognitive Bias Index. The Cognitive Bias Index uses Propensity Score Matching (PSM) and causal inference models to quantify the impact of content distortion on user behavior (such as knowledge search, opinion expression, and decision-making). The Cognitive Bias Correction Coefficient is positively correlated with the Cognitive Bias Index. When users have greater cognitive bias, the coefficient increases, thereby reducing the communication quality score.

[0040] This formula combines the traditional evaluation of the spread of news with the new evaluation of distortion and cognitive bias to form a multi-dimensional evaluation system of the spread of news. When the spread of news is expanded, but there is a high degree of distortion and significant cognitive bias, the formula The cognitive bias correction coefficient will decrease accordingly, lowering the communication quality score, effectively avoiding the misjudgment of "high communication volume and low communication value", and achieving an accurate quantitative evaluation of the news communication effect.

[0041] 2. Event-driven long-tail communication value calculation Step 1: Modeling cross-period event-news correlations Knowledge Graph Construction: Build a dynamic knowledge graph that integrates information on news (titles, entities, release dates, keywords), events (hot topics, policy releases, social events), and entities (people, organizations, and locations). A graph database is used to store nodes and edges. Edge types include "topic association" (based on keyword similarity), "temporal association" (the temporal relationship between events and news), and "entity co-occurrence" (the number and types of entities shared between news and events). This creates a dynamic network of associations between news and events, providing a foundation for subsequent analysis.

[0042] Event trigger detection: A BERT-based event type classifier categorizes and extracts real-time news and events. When a hot event, such as a "rebound in the epidemic," is detected, the system uses the knowledge graph to query historical news collections that share at least two core entities with the event (e.g., "remote work," "enterprise," and "employee"). For example, a 2020 news story about "remote work technology" could trigger a secondary spread if triggered by a 2022 epidemic rebound. This method enables accurate identification of news with potential for long-tail dissemination.

[0043] Step 2: Dynamic weight allocation of temporal attention mechanism: Time decay function improvements:; The event correlation factor is introduced to improve the traditional exponential decay function. The specific formula is as follows: ; Formula parameter explanation: : represents the weight of a piece of news dissemination data at time t. This weight measures the importance of the news to the overall dissemination effect evaluation at the current point in time. A higher value indicates a greater proportion of the news in the current evaluation and a more significant impact on the final dissemination effect evaluation results.

[0044] : is a constant attenuation coefficient. Its value range usually needs to be determined by combining actual application scenarios and repeated debugging and optimization through a large amount of experimental data. It is mainly used to adjust the weight of news dissemination data at a basic level, and directly affects the overall magnitude of the attenuation of news dissemination data weight over time. In different types of news dissemination platforms or fields, The value of will vary. For example, on social media platforms where information updates very quickly and hot topics are short-lived, in order to highlight the value of new news, It may be set relatively large, causing the weight of news dissemination data to decay faster; in the field of professional in-depth news, due to the relatively long timeliness of news value, It will be set smaller.

[0045] : is the time decay parameter, which is mainly used to accurately control the weight of news dissemination data with time interval The decay rate of the change. The larger the value, the The weight of news dissemination data is increasing. The faster the decay rate, the faster the influence of the news will decline over time; conversely, The smaller the value, the slower the news spread data weight decays, and the news can maintain a higher weight for a longer period of time. This parameter can be flexibly adjusted according to the spread of different types of news. For example, the popularity of entertainment news dissipates quickly, so a larger value can be set. For news on in-depth political analysis and academic research, the impact lasts longer, so a smaller value.

[0046] : represents the time interval between the news release time and the current time. The time unit can be set according to the actual data collection and analysis needs. Common ones are seconds, minutes, hours, days, etc. For example, when the unit is days and When , it means that the news was released 1 day away from the current time; The larger the value, the longer the news was released. In the traditional exponential decay function, the lower the weight of its dissemination data will be.

[0047] Event relevance score: It is a quantitative value calculated based on the length of the event-news path in the knowledge graph, and is specifically used to measure the closeness of the association between historical news and current events. The specific calculation rules are: when the path length between news and events in the knowledge graph is less than or equal to 2 hops ("hop" means the number of connections from one node to another in the knowledge graph, for example, the news node is connected to the event node through an intermediate entity node, and the path length is 2 hops), the score is greater than or equal to 0.8; the longer the path length, the lower the score. The role of this score is to dynamically adjust the weight of news dissemination data based on the relevance of news to current events on the basis of traditional time decay. When historical news is closely related to current events, the event relevance score is high, which will make Increase, that is, the weight of the news dissemination data is automatically increased, which effectively avoids the underestimation of news with potential value in long-tail dissemination due to simple time factors, and realizes the quantitative evaluation of the empowerment effect of historical news in current dissemination.

[0048] Through the above improvements, when historical news is closely related to current events, the weight of its dissemination data will be automatically increased. This can not only effectively avoid the problem of underestimation of the value of long-tail dissemination, but also accurately quantify the enabling effect of historical news on current dissemination, greatly improving the accuracy and effectiveness of the assessment of the value of long-tail dissemination news.

[0049] Cross-period communication network integration: The initial dissemination stage of news and the long-tail propagation stage The unified model is a spatiotemporal heterogeneous graph. In this spatiotemporal heterogeneous graph, nodes include various entities such as users, news, and events, while edges include different types of relationships such as forwarding, comments, and event associations.

[0050] Specifically, time slicing technology is used to divide the entire news dissemination process into multiple time segments in chronological order. Graph Neural Networks (GNNs) are then used to conduct in-depth analysis of the structural differences in the dissemination network during different time periods. For example, in the initial dissemination stage of news, the dissemination network is typically dominated by media nodes, relying on the influence of media organizations and dissemination channels for diffusion. However, in the long-tail dissemination stage, the dissemination network shifts to being dominated by user-generated content (UGC) nodes, relying more on spontaneous sharing and interactive dissemination among users.

[0051] Finally, the system uses graph neural networks (GNN) to comprehensively analyze and process the communication network structure of each time period, and outputs a comprehensive value of cross-time communication efficiency, thereby achieving a comprehensive and accurate evaluation of the long-tail communication effect, effectively solving the technical problem that traditional static network models cannot describe the evolution of cross-time communication network structures.

[0052] 2. System Example 1. System architecture: This system consists of a data acquisition module, a distortion analysis module, a long-tail propagation modeling module and a comprehensive evaluation module.

[0053] Data collection module: This module obtains multi-source data through social media API interfaces (such as Weibo and WeChat), news client log collection systems, and third-party data platforms (such as iResearch). It also uses a heterogeneous data cleaning engine to perform data format conversion, deduplication, and time alignment to ensure data accuracy and consistency.

[0054] Distortion Analysis Module: This module includes a multi-dimensional distortion detection submodule and a cognitive bias calculation submodule. The multi-dimensional distortion detection submodule implements semantic, emotional, and logical distortion detection models; the cognitive bias calculation submodule calculates the cognitive bias index using a user behavior database and a causal inference engine.

[0055] Long-tail propagation modeling module: This module consists of a dynamic knowledge graph construction submodule and a spatiotemporal heterogeneous graph neural network submodule. The dynamic knowledge graph construction submodule is responsible for real-time event extraction and graph updates, while the spatiotemporal heterogeneous graph neural network submodule performs time slicing processing and attention weight calculation.

[0056] Comprehensive evaluation module: Utilizes a multi-dimensional weighted algorithm and reinforcement learning optimizer to calculate the communication quality score, and displays the evaluation results in the form of a communication effect heat map, distortion-cognitive impact correlation matrix, etc. through a visual output unit.

[0057] It is necessary to add that: Distortion Analysis Module: This module combines knowledge graph entity relationship detection with causal inference technology for the first time, breaking through the traditional linear evaluation framework based on text similarity. Through multi-dimensional distortion detection and causal inference-driven cognitive impact quantification, it fully quantifies the causal chain from "information distortion to cognitive bias," providing an innovative approach for evaluating the effectiveness of secondary news creation and dissemination.

[0058] Long-tail Communication Module: This module proposes an "event-correlation-driven temporal attention mechanism," effectively addressing structural differences in communication networks across time periods through the cross-domain integration of dynamic knowledge graphs and spatiotemporal graph neural networks (GNNs). It accurately captures delayed value-added effects that traditional short-term statistics cannot identify, enabling efficient evaluation of long-tail communication.

[0059] Further explanation is needed: Existing news communication effectiveness evaluation technologies primarily focus on formalized statistics and short-term data on dissemination reach, but fail to address the actual impact of content distortion on audience perception or the long-tail nature of dissemination caused by cross-period event correlations. This solution innovatively integrates multiple technologies, including knowledge graphs, causal inference, and spatiotemporal GNNs, to construct a dynamic, multi-dimensional, and cross-period communication effectiveness evaluation system. Core concepts such as "distortion vector space" and "event-driven temporal attention" are unprecedented in existing technologies.

[0060] This solution effectively addresses the subjectivity inherent in the traditional "self-assessment" model. By objectively quantifying "cognitive bias" and "long-tail value," it enables news providers to accurately measure the actual contribution of "communication effectiveness" to news value. Field verification has shown that compared to existing technologies, this solution improves the accuracy of evaluating secondary creative content by over 40% and its sensitivity in identifying long-tail content by 60%, significantly reducing misjudgments such as "highly forwarded content with low value" and "ignoring old news with new buzz."

[0061] The combination of causal inference and natural language processing (NLP), and the integration of spatiotemporal GNNs and knowledge graphs, is a systematic solution designed to address specific technological gaps in communication effectiveness evaluation. These technologies work together to create a "1+1>2" synergistic effect. For example, causal inference provides a theoretical foundation for distortion assessment, while spatiotemporal GNNs provide algorithmic support for long-tail communication modeling.

Claims

1. A method for determining the effect of news dissemination, characterized in that: The following steps are involved: Multi-dimensional distortion detection of secondary creative content is performed to generate a distortion vector including semantic distortion, emotional distortion, and logical distortion. Analyze the correlation between the distortion vector and the user's in-depth behavior data through a causal inference model to calculate the cognitive bias index; Build a dynamic knowledge graph to identify the entity association between historical news and current events, and dynamically adjust the weight of historical communication data based on the event correlation; The original dissemination range, distortion vector, cognitive bias index and long-tail dissemination weight are integrated to output a comprehensive score of news dissemination effect.

2. The method for determining news dissemination effect according to claim 1, wherein: Multi-dimensional distortion detection includes: Entity relationship extraction technology based on knowledge graphs detects semantic distortion and identifies missing entities or distorted relationships in secondary creations; A multimodal model that integrates BERT text sentiment analysis, emoji recognition, and the sentiment intensity of forwarded messages to detect sentiment distortion; Combine the logic rule engine with the deep learning model to detect logical distortions and identify logical fallacies such as taking things out of context or replacing concepts.

3. The method for determining news dissemination effect according to claim 1, wherein: User in-depth behavior data includes: Knowledge verification behavior data, such as querying records of key entities through search engines; Opinion expression behavior data, such as semantic similarity analysis results for repeated or distorted core information in user comments; Decision-related behavior data, such as browsing logs of visits to relevant professional content pages.

4. The method for determining news dissemination effect according to claim 1, wherein: The dynamic knowledge graph contains news entities, event entities, and their subject associations, time associations, and entity co-occurrence relationships. It is stored in a graph database and updated in real time.

5. A system for determining news dissemination effect, characterized in that: include: Distortion Analysis Module: This module is used to detect semantic, emotional, and logical distortion in secondary creative content, generate distortion vectors, and calculate the cognitive bias index through causal inference. Long-tail communication modeling module: used to build a dynamic knowledge graph to identify the relationship between historical news and current events, and calculate the cross-period communication effectiveness through spatiotemporal heterogeneous graph neural networks; Comprehensive evaluation module: used to integrate the original dissemination range, distortion vector, cognitive bias index and long-tail dissemination weight to output a comprehensive score of news dissemination effect.

6. A system for determining news dissemination effect according to claim 5, characterized in that: The distortion analysis module includes: Multi-dimensional distortion detection submodule: This module includes a semantic distortion detection unit based on knowledge graph, a multimodal sentiment transfer model, and a logical distortion detection unit that integrates logic rules and deep learning. Cognitive bias calculation submodule: Contains a user behavior database and a causal inference engine, and analyzes the causal effect of distortion types on user behavior based on propensity score matching.

7. The system for determining news dissemination effect according to claim 5, wherein: The long-tail propagation modeling module includes: Dynamic knowledge graph construction submodule: used to extract events in real time and update the knowledge graph containing news, events, entities and their relationships; Spatiotemporal heterogeneous graph neural network submodule: used to model the propagation process as a heterogeneous graph containing time slices, and analyze the structural differences of the propagation network in different time periods through graph neural networks.

8. The system for determining news dissemination effect according to claim 5, wherein: It also includes a data collection module, which is used to obtain heterogeneous propagation data through multi-platform API interfaces, log collection systems and third-party data platforms, and unify the format, deduplicate and align the data in time.

Images