Methods for Evaluating the Effectiveness of Self-Media Content Dissemination

Abstract

This invention relates to the field of internet information technology and discloses a method for evaluating the dissemination effectiveness of self-media promotion content. This invention constructs a directed weighted graph by collecting user interaction data and comprehensively calculates three core graph signal indicators: structural cohesion, interaction persistence, and influence anchoring degree. Finally, a comprehensive effectiveness score is obtained through a nonlinear fusion model. This method can effectively identify and filter invalid interactions, quantify the tightness of the community structure, the depth of dialogue, and the anchoring degree of core users. It overcomes the shortcomings of existing technologies, such as poor anti-interference capabilities and limited insight dimensions, achieving a multi-dimensional and interpretable accurate evaluation of community dissemination effectiveness, and providing partners with objective priority decision-making basis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet information technology, and particularly relates to a method for evaluating the propagation efficiency of self-media promotion content. BACKGROUND

[0002] In the field of Internet information technology, especially in the direction of social media data mining and complex network analysis, how to automatically and quantitatively evaluate the interaction quality and structural characteristics of online communities is an important research direction. In the prior art, the evaluation method of the value of such a community mainly relies on statistical calculation of observable and surface interaction data.

[0003] Some prior art attempts to build a user interaction network and calculate basic graph theory indicators such as node degree centrality and network density to evaluate. However, the method based on simple graph theory indicators such as node degree and network density in the prior art can build a user interaction network, but generally lacks the ability to identify edge quality, which makes it difficult to effectively filter a large number of false interaction edges generated by automated scripts, making the network structure vulnerable to noise data pollution, and the calculation result is seriously distorted, making it difficult to truly reflect the organic cohesion of the community. The method based on interaction quantity statistics, such as the total number of comments and the average number of replies, completely ignores the deep structure and semantic information of the interaction data. This method cannot quantify the depth, duration and content quality of the dialogue, and cannot distinguish meaningful in-depth discussions from invalid chain replies. SUMMARY

[0004] The technical problem to be solved by the present application is that the prior art system has obvious deficiencies in anti-interference, structural insight depth and multi-dimensional signal fusion. Therefore, we propose a method for evaluating the propagation efficiency of self-media promotion content.

[0005] In order to achieve the above purpose, the following technical scheme is adopted in the present application: a method for evaluating the propagation efficiency of self-media promotion content, comprising the following steps: S1. Data acquisition: through the official open API interface provided by the self-media platform and the web crawler based on HTTP / HTTPS protocol, obtaining all the published content set of the target community subject within a preset time period T , and the corresponding user interaction data set , the user interaction data contains at least four-tuple information: the unique identifier of the commenter , the identifier of the replied comment , the comment content text and the comment timestamp , if it is a root comment ; S2. Interaction graph construction and signal injection: based on the data obtained in S1, a directed and weighted graph where the node set V is composed of all unique identifiers of users appearing; the edge set E is composed of reply relationships, i.e. if user replies to user , there is a directed edge from to ; the edge weight is determined by a time decay factor and a text quality factor, and the calculation formula is: ; where represents the absolute value of the time difference between the current comment and the parent comment, is a time scale constant; is the character length of the comment content ; and is a length scale constant.

[0006] Preferably, the calculation of the structural cohesion SC in S3 includes the following steps: Step one. Invalid interaction edge identification: for each edge in the graph, calculate its invalid probability , which is determined by a time anomaly factor and a content repetition factor: ; where , is a time anomaly threshold; , is a text similarity threshold; Step two. Network structure uniformity calculation: calculate the local clustering coefficients of all nodes in the graph G, denoted as the set , and the structural uniformity SU is defined as the inverse of the coefficient of variation of this set: , ; this formula shows that the SU value is proportional to the average of the local clustering coefficients and inversely proportional to the standard deviation , which is used to measure the heterogeneity of the network structure; Step three. Structural cohesion SC calculation: , which measures the proportion of suspected non-natural interactions in the network; this formula shows that the structural cohesion SC is proportional to the uniformity of the network structure SU and inversely proportional to the overall proportion of invalid edges in the graph.

[0007] Preferably, the calculation of the interaction persistence IP in S3 includes the following steps: Step one. Conversation tree set construction: extract all root comments from the graph G, i.e. The largest connected subgraph starting from the comments forms the set of dialogue trees. Step 2. Calculation of effective interaction intensity at each level: For each dialogue tree Define the effective interaction strength of its Lth layer. The sum of the weights of the incoming and outgoing edges of all nodes in this layer: ;in, For trees The set of nodes, Let v be the depth of node v in the tree. and These are the sum of the edge weights pointing to node v and the sum of the edge weights pointed to by node v, respectively; Step 3. Fitting the decay model: For each tree Associating its level L with the corresponding The model is obtained by fitting the exponential decay curve: ; The rate at which the interaction popularity of the dialogue tree decays with increasing hierarchy is solved using the nonlinear least squares method to obtain the decay constant. , This is an estimate of the initial interaction strength of the dialogue tree at the root node L=0 level. To fit the constant term in the model, used to capture the constant amount of interaction at the background level in the dialogue tree; Step 4. Global Interaction Persistence IP Calculation: The calculation uses the harmonic mean to ensure that the IP value decreases from the slowest decaying value, i.e. The smallest dialogue tree dominates, thus more accurately reflecting the overall interaction persistence of the community.

[0008] Preferably, the calculation of the influence anchoring degree IA in S3 includes the following steps: Step 1. Definition of user subject association strength: For any user node The strength of its association with community subject s Defined as Where P is the set of content published by subject s. This is an indicator function; it takes the value 1 when user u interacts with content p, and 0 otherwise. Let be the time decay function. , The decay time constant; Step 2. Anchor edge selection: Select the anchor edge set from edge set E. The condition is: for the edge Must meet , For the association strength threshold; Step 3. Anchor subgraph construction: based on anchor edge set Construct anchor subgraph ,in for The set of all edges involved nodes; Step four. Influence anchoring degree IA calculation: in the anchoring subgraph The product of its graph density and the average degree of nodes is calculated: , The formula reflects the tightness of the anchoring community and the individual activity.

[0009] Preferably, the non-linear fusion function in S4 adopts a composite model based on the geometric mean and the logic function, and the calculation formula is: Wherein is an adjustment factor, Indicates the basic synergistic effect of structure and influence, and adopts geometric mean to emphasize the coupling relationship; It is a scaled Sigmoid function, which maps the IP value to the interval (0, 1) as a multiplicative adjustment factor for the entire basic value.

[0010] Preferably, the specific process of obtaining data by web crawler in S1 includes: accessing the target uniform resource locator by simulating browser behavior; parsing the returned hypertext markup language document, locating the document object model node of the comment list; from the document object model node, the unique identifier and content field of the comment are extracted by pattern matching method; for dynamically loaded content, by analyzing the network request data, the application programming interface for obtaining comment data is located, and the page parameter of the interface is iteratively requested to obtain the complete comment data set.

[0011] Preferably, the non-linear least squares fitting in step three is realized by Levenberg-Marquardt algorithm, and the target is to minimize the loss function L: The parameter vector is updated by iteration Until convergence or maximum iteration number is reached.

[0012] Preferably, after generating the cooperation priority list, a result visualization and interactive analysis step is further included, which includes: step one. Multidimensional radar chart generation: a three-dimensional radar chart with SC, IP and IA as axes is constructed, and the index values of each community subject are normalized to the maximum and minimum and plotted on the chart to form its unique "value profile"; Step two. Interactive graph exploration: for the selected community subject, its corresponding interactive graph G is visualized under the force-directed layout algorithm, and the node size is proportional to its PageRank centrality, and the edge thickness is proportional to its weight ; Step three. Time series analysis: divide the time period T into n equal length sub-windows , recalculate SC, IP and IA in each sub-window , and obtain the time series And plot its trend line over time.

[0013] Preferably, when a request through the official API fails, the system automatically downgrades to web crawler mode; after N consecutive failed requests, the collection of data from the target is paused and an exception log is recorded.

[0014] Preferably, for the evaluated community subject, its original data, intermediate graph structure and final indicator results are cached in a distributed cache and a time-to-live (TTL) is set; when an update evaluation request for the subject is received, the validity of the cache is checked first, and a full recalculation is performed only for subjects whose TTL has expired or whose new data volume exceeds the threshold ΔD.

[0015] The technical effects and advantages of this invention are as follows: By constructing a user interaction graph and injecting time decay and text quality signals, this invention effectively identifies and filters unnatural interactions. Combined with the extraction and quantification of three graph structure indicators—Structure Cohesion (SC), Interaction Persistence (IP), and Influence Anchoring (IA)—it achieves a multi-dimensional and accurate assessment of community interaction quality. Furthermore, by fusing multiple signals into a comprehensive score (ES) through a nonlinear fusion function, and supplemented by visualization analysis and caching mechanisms, it significantly improves the anti-interference capability, structural insight depth, and system response efficiency of the assessment results. Ultimately, it can provide partners with a true, reliable, and interpretable ranking of community communication effectiveness and decision support. Attached Figure Description

[0016] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0017] Fig. 1 This is a schematic diagram of the steps of the present invention; Fig. 2 This is a schematic diagram of the logical structure of the present invention. Detailed Implementation

[0018] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0019] Reference Figs. 1-2As shown, this invention provides a technical solution: a method for evaluating the dissemination effectiveness of self-media promotion content. The method includes the following steps: S1. Data collection: Using the official open API interfaces provided by the self-media platform and web crawlers based on HTTP / HTTPS protocols, all published content and user interaction data of the target community within a preset time period T are obtained. For self-media platforms that provide official APIs, the system calls these interfaces to obtain the published content set in a structured and efficient manner. and the corresponding user interaction data set The user interaction data It must contain at least a four-tuple of information: a unique identifier for the commenter. Identifier of the comment being replied to Comment content text and comment timestamp If it is a root comment, then For platforms that do not provide official APIs or whose API data is incomplete, the system uses web crawler technology to collect data. In a preferred embodiment, in order to improve the robustness of data collection, when a request through the official API interface fails, the system will automatically downgrade to web crawler mode to try again. After multiple consecutive requests fail, the system will pause the collection of data from the target and record the exception log through the exception logger for manual intervention or subsequent retry.

[0020] S2. Interactive Graph Construction and Signal Injection: Based on the data obtained in S1, construct a directed weighted graph. And inject time decay and text quality signals, where: the construction of node set V: extract all occurrences of user-unique identifiers from user interaction data set C through node generator. The set of nodes V constitutes graph G. Each node represents a user participating in the interaction; the construction of the edge set E: the edge set E is constructed by an edge generator based on the response relationships in the user interaction data. If the user... Reply to user The comment, then, contains a line in graph G from... point to Directed edge ,this means right The comments section was interactive; edge weight calculation and injection: edge weights were calculated and injected for each edge using an edge weight calculator. Assign a weight This weight is jointly determined by the time decay factor and the text quality factor, aiming to quantify the timeliness of interaction and content quality. Its calculation formula is as follows: ;in: This represents the absolute value of the time difference between the current comment and its parent comment, i.e., by... Issued and by The absolute value of the time difference between the sending and receiving events; the smaller the time difference, the more timely the interaction, and the greater the weight. The time scale constant, preferably 3600 seconds, is the reference time unit used to normalize the time difference. The larger the value, the smoother the influence of time on the weight. This means that responses occurring within 1 hour have a higher time decay factor, while responses occurring after 1 hour have a significant decay. For comment content The longer the comment, the higher its information content and quality are generally considered to be. is a length scale constant, a reference value for comment length, used to adjust the saturation rate of text length's contribution to weight. It is preferably 50 characters. When the comment length reaches 50 characters, the text quality factor tends to saturate, and the improvement is not significant after exceeding 50 characters. The time difference and comment length are mapped to the (0,1] interval through an exponential decay function, so that the weight can comprehensively reflect the timeliness of interaction and content quality.

[0021] S3. Graph Structure Signal Extraction and Quantization Steps: Based on the graph G constructed in S2, extract and calculate three core graph structure signal indicators: Structural Cohesion (SC), Interaction Persistence (IP), and Influence Anchoring (IA). Structural Cohesion (SC) measures the tightness of interactions and the uniformity of the structure within a community, while filtering out invalid interactions. Its calculation includes the following steps: Step 1. Identification of Invalid Interaction Edges: For each edge in the graph... Calculate its invalidity probability This probability is determined by both the time anomaly factor and the content duplication factor: ;in , The time anomaly threshold is a critical time interval used to determine whether it is machine-generated commenting. Comment intervals shorter than this value are considered non-human behavior, with 5 seconds being the preferred value. , This is a text similarity threshold used to determine whether comment content is excessively repetitive. Similarity values ​​exceeding this threshold are considered invalid interactions, preferably 0.9. The ratio of Levenshtein distance to the maximum length (i.e., normalized edit distance) is also considered. If this ratio is greater than 0.9, the comment content is considered highly repetitive, and its content repetition factor is the normalized edit distance. By taking the maximum of these two values, it is ensured that if any invalidity condition is met, the edge has a high probability of being invalid. The function is a standard algorithm for calculating string edit distance, which is common knowledge in this field; Step 2. Calculate the uniformity of network structure: Calculate the local clustering coefficients of all nodes in graph G, denoted as set G. The structural uniformity SU is defined as the reciprocal of the coefficient of variation of the set: This formula shows that the SU value is related to the average local clustering coefficient. Proportional to standard deviation The SU value is inversely proportional to the heterogeneity of the network structure. A higher SU value indicates a more homogeneous network structure, dominated by a single type of interaction, such as organic discussion or collaborative data sharing; a lower SU value indicates a network with a mixture of multiple interaction modes and a more heterogeneous structure. Step 3. Calculation of Structural Cohesion (SC): The formula measures the proportion of edges in a network that are suspected of being unnatural interactions, such as bot-generated comments or content duplication. The formula shows that structural cohesion SC is proportional to the uniformity of the network structure SU and inversely proportional to the overall proportion of invalid edges in the graph.

[0022] Persistent Interaction IP is used to measure the depth and duration of community conversations. Its calculation includes the following steps: Step 1. Constructing the Conversation Tree Set: Extract all root comments from graph G, i.e. The largest connected subgraph starting from the comments forms the set of dialogue trees. Each dialogue tree represents an independent discussion thread; Step 2. Calculation of effective interaction intensity at each level: For each dialogue tree Define the effective interaction strength of its Lth layer. The sum of the weights of the incoming and outgoing edges of all nodes in this layer: ;in, For trees The set of nodes, Let v be the depth of node v in the tree. This value represents the level of attention node v receives in the dialogue tree. This value represents the level of activity of node v in actively participating in the discussion within the dialogue tree. and These are the sum of the edge weights pointing to node v and the sum of the edge weights pointed to by node v, respectively. This strength comprehensively considers the quality and timeliness of all comments in this layer; Step 3. Decay model fitting: For each tree Associating its level L with the corresponding Linear exponential decay fitting, the model is: ; To determine the rate at which the interaction popularity of the dialogue tree decays with increasing hierarchy, the decay constant is solved using the nonlinear least squares method. , This is an estimate of the initial interaction strength of the dialogue tree at the root node L=0 level. The constant term in the fitting model is used to capture the constant amount of interaction at the background level in the dialogue tree; in a preferred embodiment, the nonlinear least squares fitting is implemented using the Levenberg-Marquardt algorithm, the objective of which is to minimize the loss function L: By iteratively updating the parameter vector Until convergence or the maximum number of iterations is reached; S34. Global Interactive Persistent IP Calculation: The calculation uses the harmonic mean to ensure that the IP value decreases from the slowest decaying value, i.e. Smaller dialogue trees dominate, thus more accurately reflecting the overall interactivity and persistence of the community.

[0023] Influence Anchoring Strength (IA) measures the degree of connection and activity between core users and the main body of a community. Its calculation includes the following steps: Step 1. Definition of User-Main Body Connection Strength: For any user node... The strength of its association with community subject s Defined as Where P is the set of content published by subject s. This is an indicator function; it takes the value 1 when user u interacts with content p, and 0 otherwise. Let be the time decay function. , The longer the influence lasts, the better, preferably 7 days; Step 2. Anchor edge selection: Select the anchor edge set from edge set E. The condition is: for the edge Must meet , For the correlation strength threshold, This is the minimum qualification threshold used to determine whether a user has a strong connection with the core community, used to filter out the core group. The preferred version is 2.0, meaning that only users who receive a response... When the correlation strength with the main body of the community reaches a certain threshold, the related interaction edges are considered anchor edges; Step 3. Construction of anchor subgraph: based on the anchor edge set Construct anchor subgraph ,in for The set of nodes involved by all edges in the graph; the anchored subgraph represents the core user group and its interaction network closely related to the main body of the community; Step 4. Influence Anchoring Degree (IA) Calculation: In the anchored subgraph Calculate the product of its graph density and the average degree of its nodes: ; This formula comprehensively reflects both the cohesion of the anchored community and the level of individual activity.

[0024] S4. Multi-signal fusion decision: The three indicators SC, IP, and IA calculated in S3 are used as inputs. A comprehensive evaluation score ES is calculated through a nonlinear fusion function. All community entities to be evaluated are ranked according to ES to generate a cooperation priority list.

[0025] The nonlinear fusion function adopts a composite model based on geometric mean and logistic function, and its calculation formula is as follows: ,in This is a modulating factor used to control the rate of change of interaction persistence IP in the Sigmoid function. The larger the value, the more significant the positive impact of a small increase in IP on the final score ES. The geometric mean is used to emphasize the coupling relationship between the basic synergistic effect of structure and influence. The geometric mean can ensure that when either SC or IA is low, their product will significantly lower the basic value, reflecting the importance of both being indispensable. It is a scaled Sigmoid function that maps IP values ​​to the (0,1) interval. As a multiplicative adjustment factor for the entire base value, the Sigmoid function can smoothly introduce the nonlinear effects of IP into the composite score. When IP is low, the adjustment factor is close to 0, and when IP is high, the adjustment factor is close to 1, thus effectively regulating the base synergistic effect.

[0026] Finally, the sorting and list generator sorts all the community entities to be evaluated based on the calculated comprehensive evaluation score ES, and generates a cooperation priority list. The higher the ES score, the better the dissemination effectiveness of the community entity's promotional content, and the higher the cooperation priority.

[0027] After generating the cooperation priority list, this invention also includes a result visualization and interactive analysis step, executed by the visualization and interactive analysis module, designed to provide an intuitive display of evaluation results and in-depth analytical capabilities: Step 1. Multidimensional Radar Chart Generation: Construct a three-dimensional radar chart with SC, IP, and IA as axes, and plot the index values ​​of each community entity on the chart after max-min normalization, forming its unique "value profile"; Step 2. Interactive Graph Exploration: For the selected community entity, visualize its corresponding interaction graph G under the force-directed layout algorithm, where the node size is proportional to its PageRank centrality and the edge thickness is proportional to its weight. Step 3. Time series analysis: Divide the time period T into n equal-length sub-windows. In each child window Recalculate SC, IP, and IA to obtain the time series. And plot its trend line over time.

[0028] In a preferred embodiment, the present invention further includes cache management. For evaluated community subjects, their original data, intermediate graph structure, and final indicator results are cached in a distributed cache, and a time-to-live (TTL) is set. When an update evaluation request for a subject is received, the cache checker first checks the cache validity. Only for subjects whose TTL has expired or whose new data volume exceeds the threshold ΔD will the data update trigger trigger a full recalculation. This mechanism significantly improves the response speed and resource utilization of the evaluation system and avoids unnecessary duplicate calculations.

[0029] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for evaluating the propagation efficiency of self-media promotion content, characterized in that, The method comprises the following steps: S1. Data collection: Through the official open API interface provided by the self-media platform and the web crawler based on HTTP / HTTPS protocol, all the published content sets of the target community subject within the preset time period T are obtained , and the corresponding user interaction data set , the user interaction data at least contains four-tuple information: the unique identifier of the commenter , the identifier of the replied comment , the comment content text and the comment timestamp , if it is a root comment ; S2. Interaction graph construction and signal injection: Based on the data obtained in S1, a directed and weighted graph is constructed , where the node set V is composed of all unique identifiers of users appearing; the edge set E is composed of reply relationships, i.e. if user replies to user , there is a directed edge from to ; the edge weight is determined by a time decay factor and a text quality factor, and its calculation formula is: ; where represents the absolute value of the time difference between the current comment and the parent comment, is the time scale constant; is the character length of the comment content ; and is the length scale constant; S3. Graph structure signal extraction and quantification: based on the graph G constructed in S2, three core graph structure signal indicators are extracted and calculated: structure cohesion SC, interaction persistence IP and influence anchoring degree IA; S4. Multi-signal fusion decision: taking the three indicators SC, IP and IA calculated in S3 as input, a comprehensive evaluation score ES is calculated through a nonlinear fusion function, and all the evaluated social subjects are sorted according to ES to generate a cooperation priority list; The calculation of the structural cohesion SC in S3 includes the following steps: Step 1. Invalid interaction edge identification: for each edge in the graph , calculate its invalid probability , which is determined by the time anomaly factor and the content repetition factor: ; wherein , is the time anomaly threshold; , is the text similarity threshold; Step 2. Network structure uniformity calculation: calculate the local clustering coefficient of all nodes in the graph G, denoted as the set , and the structure uniformity SU is defined as the inverse of the coefficient of variation of this set: , , ; this formula shows that the SU value is proportional to the average of the local clustering coefficient , and inversely proportional to the standard deviation , which is used to measure the heterogeneity of the network structure; Step 3. Structural cohesion SC calculation: ; measures the suspected non-natural interaction in the network; this formula shows that the structural cohesion SC is proportional to the uniformity of the network structure SU, and inversely proportional to the overall proportion of invalid edges in the graph.

2. The method of claim 1, wherein: The calculation of persistent interactive IP in S3 includes the following steps: Step 1. Constructing the dialogue tree set: Extract all root comments from graph G, i.e. The largest connected subgraph starting from the comments forms the set of dialogue trees. Step 2. Calculation of effective interaction intensity at each level: For each dialogue tree Define the effective interaction strength of its Lth layer. The sum of the weights of the incoming and outgoing edges of all nodes in this layer: ;in, For trees The set of nodes, Let v be the depth of node v in the tree. and These are the sum of the edge weights pointing to node v and the sum of the edge weights pointed to by node v, respectively; Step 3. Decay model fitting: For each tree Assign its level L to the corresponding The model is obtained by fitting the exponential decay curve: ; To determine the rate at which the interaction popularity of the dialogue tree decays with increasing hierarchy, the decay constant is solved using the nonlinear least squares method. , This is an estimate of the initial interaction strength of the dialogue tree at the root node L=0 level. To fit the constant term in the model, used to capture the constant amount of interaction at the background level in the dialogue tree; Step 4. Global Interaction Persistence IP Calculation: The calculation uses the harmonic mean to ensure that the IP value decreases from the slowest decaying value, i.e. The smallest dialogue tree dominates, thus more accurately reflecting the overall interaction persistence of the community.

3. The method of claim 1, wherein: The calculation of the influence anchoring degree IA in S3 includes the following steps: Step 1. User subject association strength definition: for any user node , the association strength with the community subject s is defined as , , where P is the content set published by the subject s, is an indicator function, taking 1 when the user u has an interaction under the content p, otherwise taking 0; is a time decay function, , is the decay time constant; Step 2. Anchoring edge screening: screening the anchoring edge set from the edge set E, with the condition that for edge , it must satisfy , is the association strength threshold; Step 3. Anchoring subgraph construction: based on the anchoring edge set , the anchoring subgraph is constructed, where is the set of nodes involved in all edges in ; Step 4. Influence anchoring degree IA calculation: on the anchoring subgraph , calculate the product of its graph density and the average degree of nodes: , ; , the formula comprehensively reflects the closeness of the anchoring community and the individual activity.

4. The method of claim 1, wherein: The nonlinear fusion function in S4 adopts a composite model based on geometric mean and logical function, and its calculation formula is: wherein is an adjustment factor, represents the basic synergistic effect of structure and influence, and the geometric mean is used to emphasize the coupling relationship; is a scaled Sigmoid function, which maps the IP value to the interval (0, 1) as a multiplicative adjustment factor for the overall basic value.

5. The method of claim 1, wherein: The specific process of obtaining data in S1 through the web crawler includes: accessing the target uniform resource locator by simulating browser behavior; parsing the returned hypertext markup language document, locating the document object model node of the comment list; from the document object model node, the unique identifier and content field of the comment are extracted through pattern matching method; for dynamically loaded content, the application programming interface for obtaining comment data is located by analyzing network request data, and the complete comment data set is obtained by iteratively requesting the paging parameters of the interface.

6. The method of claim 2, wherein: The non-linear least squares fitting in step three is implemented using the Levenberg-Marquardt algorithm, which aims to minimize the loss function L: by iteratively updating the parameter vector until convergence or a maximum number of iterations is reached.

7. The method of claim 6, wherein: After generating the cooperation priority list, a result visualization and interactive analysis step is also included, which includes: step one. Multidimensional radar chart generation: a three-dimensional radar chart is constructed with SC, IP, and IA as the axes, and the index values of each community subject are normalized to the maximum and minimum and plotted on the chart to form its unique "value profile"; step two. Interactive graph exploration: for the selected community subject, its corresponding interaction graph G is visualized under the force-directed layout algorithm, and the node size is proportional to its PageRank centrality, and the edge thickness is proportional to its weight ; step three. Time series analysis: divide the time period T into n equal length sub-windows , recalculate SC, IP, and IA in each sub-window , obtain the time series , and draw its trend line over time.

8. The method of claim 7, wherein the method further comprises: When the request through the official API interface fails, it is automatically downgraded to the web crawler mode; when the request fails continuously for N times, the collection of the target is suspended, and an exception log is recorded.

9. The method of claim 8, wherein: For the evaluated social subjects, their original data, intermediate graph structure and final indicator results are cached in the distributed cache, and the time to live TTL is set; when receiving an update evaluation request for the subject, first check the cache validity, and only for the subjects whose TTL has expired or the new data volume exceeds the threshold ΔD, perform full re-computation.

Images