Method and system for evaluating and recommending self-media cooperation effectiveness based on interaction amount list

Abstract

The application provides a self-media cooperation effectiveness evaluation and recommendation method and system based on an interaction amount list, relates to the technical field of data processing, and the method comprises the following steps: obtaining multi-dimensional sound data of a product and a competitive product on multiple content platforms and receiving a filtering condition; then filtering and aggregating the sound data based on the filtering condition to generate a sound subset, and generating a content weight sequence; then combining the content weight sequence with initial interaction data to generate weighted interaction data, and generating an interaction amount list of the product and the competitive product based on the data; then associating and matching content entries in the list with a self-media information library to obtain self-media information, and calculating a recommendation score of each self-media and the product; finally, evaluating the self-media cooperation effectiveness in combination with the recommendation score, a self-media follower growth rate and content verticality, and generating a self-media recommendation list according to the evaluation result. The application improves the accuracy of self-media cooperation evaluation and the reliability of the recommendation result.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for evaluating and recommending the effectiveness of self-media collaborations based on a list of interaction volume. Background Technology

[0002] The self-media collaboration effectiveness evaluation and recommendation method is an important tool for brands to screen partners and measure campaign effectiveness in social media marketing. With the development of content platforms and the rise of the self-media economy, this method has shown broad application prospects in the field of marketing decision support.

[0003] Existing technologies primarily evaluate self-media collaborations based on content interaction data. For example, they generate interaction rankings by statistically analyzing metrics such as the number of likes, comments, and reposts of content published by self-media, then select collaborating self-media based on these rankings, and finally make a preliminary judgment on the effectiveness of the collaboration by combining the size of the self-media's followers and the content category. This type of method is widely used in marketing practice.

[0004] However, in the process of evaluating self-media based on interaction rankings, the mutual referencing and dissemination relationships between different content can affect the authenticity of interaction data. As a result, rankings based solely on raw interaction values ​​are difficult to accurately reflect the actual influence of self-media. At the same time, multiple factors such as follower growth trends and content professionalism are also difficult to effectively balance in a single evaluation system. Therefore, existing technologies have technical problems with insufficient accuracy in evaluating self-media collaborations. Summary of the Invention

[0005] This application provides a method and system for evaluating and recommending the effectiveness of self-media collaborations based on an interaction volume list, in order to solve the problems of low accuracy in self-media collaboration evaluation and poor reliability of recommendation results in the prior art.

[0006] To address the aforementioned technical issues, firstly, this application provides a method for evaluating and recommending the effectiveness of self-media collaborations based on an interaction volume list, including:

[0007] Obtain multi-dimensional volume data of this product and at least one competitor on at least one content platform, and receive filtering conditions input by users;

[0008] Based on the aforementioned filtering criteria, the multi-dimensional volume data is filtered and aggregated to generate volume sets corresponding to the product and competitors. The PageRank algorithm is then used to rank the content in the volume sets by importance, generating a content weight sequence.

[0009] Based on the content weight sequence and the initial interaction data of each content in the quantum sound set, weighted interaction data is generated, and based on the weighted interaction data, interaction volume lists corresponding to the product and the competitor are generated respectively.

[0010] Each content item in the interaction volume list is associated and matched with a preset self-media information database to determine the corresponding self-media information that generated each content item. Based on the self-media information, a collaborative filtering recommendation algorithm is used to calculate the recommendation score of each self-media and the product.

[0011] Using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the growth rate of followers of the self-media, and the verticality of the content, the effectiveness of the self-media cooperation between the product and the competitor on the content platform is evaluated, and a self-media recommendation list is generated based on the evaluation results.

[0012] Secondly, this application provides a self-media collaboration effectiveness evaluation and recommendation system based on an interaction volume list, including:

[0013] The acquisition module is used to acquire multi-dimensional volume data of this product and at least one competitor on at least one content platform, and to receive filtering conditions input by the user.

[0014] The filtering module is used to filter and aggregate the multi-dimensional volume data based on the filtering conditions, generate volume sets corresponding to the product and competitors, and use the PageRank algorithm to sort the importance of the content in the volume sets to generate a content weight sequence.

[0015] The generation module is used to generate weighted interaction data based on the content weight sequence and the initial interaction data of each content in the sound quantum set, and to generate an interaction volume list for the product and the competitor based on the weighted interaction data.

[0016] The matching module is used to associate and match each content item in the interaction volume list with a preset self-media information database to determine the corresponding self-media information that generated each content item, and to calculate the recommendation score of each self-media and the product based on the self-media information using a collaborative filtering recommendation algorithm.

[0017] The evaluation module is used to evaluate the effectiveness of the product's cooperation with the competitor's self-media on the content platform using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the self-media's follower growth rate and content verticality. Based on the evaluation results, a self-media recommendation list is generated.

[0018] Thirdly, this application provides an electronic device, comprising:

[0019] Memory, used to store computer programs;

[0020] A processor, configured to execute the computer program to implement the steps of the self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list as described in the first aspect above.

[0021] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list as described in the first aspect above.

[0022] The technical solution provided in this application has the following beneficial effects:

[0023] This application provides a comprehensive and targeted data foundation for subsequent evaluation by acquiring multi-dimensional volume data and receiving screening criteria. Then, based on the screening criteria, the data is processed and a PageRank algorithm is used to generate a content weight sequence, effectively identifying content with high influence in the dissemination network. Next, the content weight sequence is combined with initial interaction data to generate weighted interaction data and form an interaction volume list, making the ranking results more reflective of the true dissemination value of the content. Then, the ranking content is correlated and matched with a self-media information database, and a collaborative filtering recommendation algorithm is used to calculate recommendation scores, achieving the goal of discovering potential high-quality self-media based on content performance. Finally, an entropy weight method with multi-criteria decision-making combined with multi-dimensional indicators is used for evaluation and to generate a recommendation list, ensuring the comprehensiveness and objectivity of the evaluation results.

[0024] Furthermore, for each item in the first and second quantum sets, this application obtains its initial interaction data and finds the corresponding importance value from the content weight sequence. Then, it uses an energy propagation algorithm based on a heat conduction model to calculate the conduction heat value of each content node according to the content reference relationship. The initial interaction data, importance value, and conduction heat value are weighted and combined to obtain weighted interaction data. Then, a Kalman filter algorithm is used to dynamically smooth and estimate the state of the weighted interaction data to generate stable interaction data. Finally, the stable interaction data of all content in the two quantum sets are sorted and multiple contents exceeding the preset value are selected to form an interaction volume list corresponding to this product and competitors.

[0025] Furthermore, this method calculates the spread of content by introducing a heat conduction model, enabling interaction data to reflect the diffusion effect of content's influence on the network. At the same time, it uses a Kalman filter algorithm to dynamically smooth the data, effectively eliminating the impact of short-term fluctuations on the evaluation results. The final generated interaction volume list has higher stability and accuracy.

[0026] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating a self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list, provided for embodiments of this application;

[0029] Figure 2 A schematic diagram illustrating a specific implementation of a self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list, provided in this application embodiment;

[0030] Figure 3 This is a schematic diagram of the structure of a self-media collaboration effectiveness evaluation and recommendation system based on an interaction volume list, provided in an embodiment of this application. Detailed Implementation

[0031] In the evaluation and recommendation of self-media collaboration effectiveness, existing technologies mainly rely on interaction rankings to screen and evaluate self-media. Rankings are generated by statistically analyzing interaction data such as likes, comments, and reposts of self-media content, and the value of collaboration is judged by basic indicators such as follower size. However, because the mutual referencing and dissemination relationships between content can affect the authenticity of interaction data, rankings based solely on raw interaction values ​​cannot accurately reflect the actual dissemination influence of self-media. Furthermore, multiple dimensions such as follower growth trends and content professionalism cannot be effectively balanced within a single evaluation system, resulting in significant deficiencies in the accuracy and comprehensiveness of the evaluation results.

[0032] To address the aforementioned issues, this application proposes a method for evaluating and recommending the effectiveness of self-media collaborations based on an interaction volume list. Its core lies in achieving accurate evaluation through multi-dimensional data fusion and multi-stage algorithmic collaboration. Specifically, firstly, multi-dimensional volume data of the product and its competitors are acquired and filtering conditions are received. Then, the data is filtered and aggregated based on the filtering conditions. The PageRank algorithm is used to rank the content by importance, generating a content weight sequence. Next, the content weight sequence is combined with the initial interaction data to generate weighted interaction data and form an interaction volume list. Subsequently, the list content is matched with a self-media information database, and a collaborative filtering recommendation algorithm is used to calculate the recommendation score. Finally, an entropy weight method multi-criteria decision-making method is used to comprehensively evaluate the recommendation score, follower growth rate, and content verticality, generating a self-media recommendation list.

[0033] Therefore, this method, by introducing content dissemination network analysis, multi-indicator weighted fusion, and dynamic data smoothing, enables the interaction volume ranking to truly reflect the dissemination value of the content. At the same time, it comprehensively considers multiple factors such as the growth trend of followers and the professionalism of the content, fundamentally solving the problem of insufficient evaluation accuracy in existing technologies and improving the reliability of self-media cooperation evaluation and the effectiveness of recommendation results.

[0034] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] The core of this application is to provide a method for evaluating and recommending the effectiveness of self-media collaborations based on an interaction volume list. A flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0036] Step 101: Obtain multi-dimensional volume data of this product and at least one competitor on at least one content platform, and receive filtering conditions input by the user.

[0037] In step 101, multi-dimensional voice volume data refers to a series of quantitative indicators generated by the content published by this product and competitors on the content platform. These indicators include the number of content published, the number of content interactions, and the content dissemination range data. The number of interactions includes the number of likes, comments, and reposts, while the dissemination range data includes the content's exposure or readership.

[0038] Filtering criteria refer to the specific parameters set by the user to limit the range of data. These parameters include time range, content platform type, and product category. The time range is used to specify the start and end dates of the data, the content platform type is used to specify which social media or content sharing platforms the data comes from, and the product category is used to specify the product categories related to this product and competitors.

[0039] In this embodiment, multi-dimensional volume data of the product and its competitors are first obtained from various content platforms through data interfaces. At the same time, the time range, content platform type and product category input by the user on the interactive interface are received as filtering conditions to provide basic data sources and parameter constraints for subsequent data processing and evaluation.

[0040] Step 102: Based on the filtering conditions, the multi-dimensional volume data is filtered and aggregated to generate volume sets corresponding to the product and competitors. The PageRank algorithm is used to sort the importance of the content in the volume sets and generate a content weight sequence.

[0041] Among them, the sound quantum set refers to the set of content data related to this product and competitors selected from multi-dimensional sound volume data. This set includes the first sound quantum set and the second sound quantum set. The first sound quantum set corresponds to the content data of this product, and the second sound quantum set corresponds to the content data of competitors. The content weight sequence refers to the sequence formed by arranging all the content in the sound quantum set from high to low importance score.

[0042] In this embodiment, step 102 includes the following process:

[0043] Step 1021: Based on the time range and content platform in the filtering conditions, extract multiple original content records corresponding to this product and each competitor from the multi-dimensional volume data.

[0044] In step 1021, the original content record refers to the content items and related data that meet the time range and content platform conditions selected from the multi-dimensional volume data.

[0045] In this embodiment, firstly, all content items within the time range input by the user are filtered from the multi-dimensional volume data. Then, content items belonging to the specified platform are filtered from the content platform type. Finally, the original content records corresponding to this product and each competitor are obtained.

[0046] Step 1022: Based on the product categories in the filtering conditions, classify the multiple original content records, and accumulate the interaction data belonging to the same content to generate the first quantum set of the product and the second quantum set of the competitor.

[0047] In step 1022, the first quantum set refers to the set of all content items and their interactive data corresponding to this product, and the second quantum set refers to the set of all content items and their interactive data corresponding to the competitor's product.

[0048] In this embodiment, the original content records are classified according to product category. Content records belonging to this product are classified into this product category, and content records belonging to competitors are classified into competitors category. For records of the same content published at different times or through different channels, the corresponding interaction data is accumulated and merged to form the complete interaction data of the content. All classified and merged content entries of this product constitute the first quantum set, and all classified and merged content entries of competitors constitute the second quantum set.

[0049] Step 1023: Construct content reference graphs for the first acoustic quantum set and the second acoustic quantum set respectively. Nodes in the content reference graph represent content, and directed edges between nodes represent reference relationships between content.

[0050] In step 1023, the content reference graph is a network structure graph used to represent the reference relationship between content. The nodes in the graph represent specific content items, and the directed edges between nodes represent the reference direction between content. The directed edge from content A to content B indicates that content A references content B.

[0051] In this embodiment, content reference graphs are constructed for the first quantum set and the second quantum set respectively. For all content entries in the first quantum set, it is analyzed whether there are any reference links or mentions of other content in each content entry. If content A references content B, a directed edge from node A to node B is added to the content reference graph. This process is repeated until all reference relationships in the first quantum set are completely recorded, forming the content reference graph corresponding to the first quantum set. Similarly, the same method is used to construct the corresponding content reference graph for the second quantum set.

[0052] Step 1024: Apply the PageRank algorithm to iteratively calculate the importance score of each node in the content reference graph.

[0053] In step 1024, the importance score is a value calculated using the PageRank algorithm to quantify the importance of each piece of content in the citation network. This value is determined based on the number of citations pointing to the content and the importance of the citation source content.

[0054] In this embodiment, the PageRank algorithm is applied to each content reference graph for iterative calculation. The importance scores of all nodes in the graph are initialized to be equal. Then, the importance score of each node is updated repeatedly according to the iterative rules of the PageRank algorithm. Each time the node is updated, the new score of the node is equal to the sum of the importance scores of all nodes pointing to that node divided by the number of its outgoing edges, multiplied by the damping coefficient, and then added to the base assignment value. After multiple iterations, the importance scores of all nodes tend to stabilize and converge, and the final importance score of each node is obtained.

[0055] Step 1025: Sort the importance scores of all nodes according to the nodes to form a content weight sequence.

[0056] In this embodiment of the application, all nodes in the content reference graph corresponding to the first quantum set are sorted from high to low according to their importance scores to form the content weight sequence corresponding to this product; similarly, all nodes in the content reference graph corresponding to the second quantum set are sorted from high to low according to their importance scores to form the content weight sequence corresponding to the competitor.

[0057] This application generates a quantum set of sound by filtering and aggregating multi-dimensional sound volume data based on screening criteria, and uses the PageRank algorithm to analyze the content citation network. This effectively identifies content with high influence in the dissemination network, providing an accurate basis for the weighting of subsequent interaction data and improving the objectivity and accuracy of content value assessment.

[0058] Step 103: Based on the content weight sequence and the initial interaction data of each content in the quantum sound set, generate weighted interaction data, and based on the weighted interaction data, generate interaction volume lists for the product and the competitor respectively.

[0059] Among them, weighted interaction data refers to the interaction value obtained by combining the initial interaction data of the content with the importance value and heat value in the content weight sequence. This value is used to comprehensively reflect the actual influence of the content in the dissemination network. The interaction volume list refers to the list formed by sorting the weighted interaction data from high to low after dynamic smoothing and selecting multiple contents that exceed the preset value. This list includes the first interaction volume list corresponding to this product and the second interaction volume list corresponding to competitors.

[0060] In this embodiment, such as Figure 2 As shown, step 103 includes the following process:

[0061] Step 1031: For each item in the first and second quantum sets, obtain the initial interaction data of the corresponding item and find the importance value of the corresponding item from the content weight sequence.

[0062] In step 1031, the initial interaction data refers to the raw interaction values ​​directly obtained from the quantum sound set, which include the number of likes, comments, and shares of the content.

[0063] In this embodiment, each item in the first quantum set is traversed first, and the initial interaction data is read from the data record corresponding to the content. At the same time, the importance value corresponding to the content is found in the content weight sequence according to the content identifier. The same method is used to traverse each item in the second quantum set to obtain the initial interaction data and the corresponding importance value of each item.

[0064] In practical applications, taking the content of a beauty brand's product A and its competitor B on a short video platform as an example, for a piece of content "lipstick swatch video" in the First Quantum Set, the initial interaction data read from the data records includes 12,000 likes, 320 comments, and 560 shares; based on the title and posting link of the content, the importance value of the content is found to be 0.85 in the content weight sequence.

[0065] Step 1032: Using an energy propagation algorithm based on a heat conduction model, calculate the heat conduction value of each content node according to the reference relationship between content. The heat conduction value reflects the heat spread obtained by the content from other high-weight content.

[0066] In step 1032, the energy propagation algorithm based on the heat conduction model is a calculation method that simulates the heat transfer process in a physical medium. This method compares the content reference network to a heat conduction medium and the importance value of the content node to heat. By iteratively calculating the cumulative effect of heat transfer along the reference relationship from high-weight content to content directly or indirectly related to it, the method finally obtains the heat conduction value obtained by each content node from the network.

[0067] The heat transfer value is a numerical value calculated using a heat transfer model to quantify the heat transfer effect that content gains from the citation network. This value is determined based on the importance values ​​of other content citing the content and the length of the citation path. High-weight content refers to content items that rank high in importance scores in the content weight sequence. These contents have many citation sources in the content citation network or are cited by many other high-weight contents, thus having a stronger dissemination influence on other content in the network.

[0068] In this embodiment of the application, an energy propagation algorithm based on a heat conduction model is applied to calculate the content reference graphs in the first and second quantum sets. For each content node, all reference source nodes pointing to that node are identified, the importance values ​​of these source nodes are obtained, and the importance values ​​of each source node are multiplied by the heat conduction coefficient and then summed to obtain the heat conduction value of the content node. This process is repeated until the heat conduction value of all content nodes is calculated.

[0069] In practical applications, analyzing the content citation graph of the aforementioned "lipstick swatch video" reveals three citations pointing to it: one from the "Celebrity Lipstick Recommendation" video (importance value 0.92), another from the "Affordable Lipstick Review" video (importance value 0.78), and yet another from the "Lipstick Swatch Collection" video (importance value 0.65). Assuming a thermal conductivity coefficient of 0.7, the thermal conductivity of this content is 0.92×0.7 + 0.78×0.7 + 0.65×0.7, resulting in a value of 1.645.

[0070] Step 1033: The initial interaction data, importance value and thermal conductivity value are weighted and combined to obtain weighted interaction data.

[0071] In this embodiment, the initial interaction data, importance value, and thermal conductivity value are assigned combined weights. The initial interaction data of each item is multiplied by the first weight, the importance value by the second weight, and the thermal conductivity value by the third weight. The three product results are then added together to obtain the weighted interaction data corresponding to that item. The same operation is performed on all items in sequence to complete the calculation of the weighted interaction data of all items in the first and second quantum sets.

[0072] In practical applications, the initial interaction data is weighted at 0.4, the importance value at 0.3, and the heat transfer value at 0.3. For the "lipstick swatch video" content, the initial 12,000 interactions are converted into a standard score of 40. The importance value is 0.85 × 0.3 = 0.255, and the heat transfer value is 1.645 × 0.3 = 0.4935. Therefore, the weighted interaction data is 40 × 0.4 + 0.255 + 0.4935, and the calculated result is 16.7485.

[0073] Step 1034: Use the Kalman filter algorithm to dynamically smooth and estimate the state of the weighted interaction data to generate stable interaction data.

[0074] Among them, stable interaction data refers to the interaction value obtained by dynamically smoothing the weighted interaction data through the Kalman filter algorithm. This value eliminates the impact of short-term fluctuations and can more stably reflect the long-term interaction trend of the content.

[0075] Step 1034 may specifically include the following steps:

[0076] A1: Define a state vector and establish a state change equation, which describes the rules by which the state vector changes over time.

[0077] In step A1, the state vector refers to a set of parameters used to describe the content interaction state, including the interaction value and its rate of change; the state change equation refers to the mathematical relationship describing the evolution of the content interaction state over time. This equation predicts the interaction level at the current moment based on the interaction level and rate of change at the previous moment, where the amount of change in the interaction level is determined by the rate of change at the previous moment, and the rate of change remains constant during the prediction process. This linear recursive relationship enables dynamic modeling of the content interaction trend.

[0078] It should be noted that the embodiments of this application do not specifically limit the specific form of the state change equation, and can be set accordingly according to the actual situation.

[0079] In this embodiment, a state vector is first defined for each content. The state vector contains two components: the first component represents the current interaction level of the content, and the second component represents the rate of change of the interaction level. A state change equation is established, which represents the linear relationship between the state vector at the current moment and the state vector at the previous moment. The change in the interaction level is affected by both the interaction level and the rate of change at the previous moment, while the rate of change remains relatively stable.

[0080] In practical applications, for "lipstick swatch video" content, its state vector is defined as [ , ],in This represents the interaction level at time k. The rate of change at time k is represented; the state change equation is established as follows: , ,in Indicates the first The level of interaction at any given moment Indicates the first rate of change at time, Indicates a time interval.

[0081] A2: Based on the state change equation, calculate the predicted state value at the current moment using the state data from the previous moment.

[0082] In step A2, the state prediction value refers to the result obtained by predicting the state vector at the current moment through the state change equation.

[0083] In this embodiment of the application, the state vector of the previous moment is read, and the state vector is substituted into the state change equation to calculate the state prediction value of the current moment. The state prediction value includes the prediction value of the interaction level and the prediction value of the rate of change.

[0084] In practical applications, the state vector at the previous moment... For example, if the time interval is set to 1 time unit, then the predicted state value at the current moment... , ,in Indicates time Interaction level prediction Indicates time Predicted rate of change.

[0085] A3: Use the weighted interactive data as observations and calculate the Kalman gain.

[0086] In step A3, the observed value refers to the weighted interactive data calculated in step 1033, which serves as the actual measurement input for the Kalman filter algorithm; the Kalman gain value refers to the weighting coefficient calculated based on the prediction error and the observation error, which is used to balance the contribution ratio of the predicted state value to the observed state value in the state estimation.

[0087] In this embodiment, the weighted interaction data at the current moment is obtained as the observation value. The Kalman gain value is calculated based on the error covariance of the state prediction and the observation error covariance. The magnitude of the gain value reflects the degree of trust in the observation value. The larger the gain value, the more trust in the observation value, and the smaller the gain value, the more trust in the state prediction value.

[0088] In practical applications, for the "lipstick swatch video" content, the weighted interactive data observation value at the current moment is 16.7485. Setting the prediction error covariance to 0.5 and the observation error covariance to 0.2, the Kalman gain value is... The calculated result is approximately 0.714.

[0089] A4: Based on the predicted state value, the observed value, and the Kalman gain value, calculate the optimal state estimate for the current time, and output the interaction value component in the optimal state estimate as stable interaction data.

[0090] In step A4, the optimal state estimate refers to the state vector obtained by weighted averaging of the state prediction and the observed values. This vector is the optimal estimate of the true state at the current moment.

[0091] In this embodiment, the difference between the predicted state value and the observed value is multiplied by the Kalman gain to obtain a correction value. This correction value is then added to the predicted state value to obtain the optimal state estimate at the current moment. The interaction level component is extracted from the optimal state estimate and used as the stable interaction data output for the content at the current moment. The same operation is performed on all content in sequence to obtain the stable interaction data for all content.

[0092] In practical applications, for the "lipstick swatch video" content, the difference between the observed value and the predicted state value is 16.7485 - 15.5 = 1.2485, and the correction is 1.2485 × 0.714 ≈ 0.891. Therefore, the interaction level component in the optimal state estimate at the current moment... This value will be used as the stable interactive data output for this content.

[0093] Step 1035: Sort the stable interaction data of all contents in the first quantum set and the second quantum set from high to low, and select multiple contents with stable interaction data exceeding a preset value from the sorting results to form the first interaction volume list corresponding to the product and the second interaction volume list corresponding to the competitor.

[0094] In step 1035, the preset value refers to the pre-set threshold value of stable interactive data used to filter content to enter the list.

[0095] The embodiments of this application do not specifically limit the value of the preset value; it can be set according to the actual situation.

[0096] In this embodiment, the stable interaction data of all content in the first quantum set are arranged in descending order. Starting from the beginning of the sorting results, multiple contents with stable interaction data greater than a preset value are selected in sequence, and these contents are grouped into the first interaction volume list corresponding to this product. The same method is used to process all content in the second quantum set to form the second interaction volume list corresponding to the competitor.

[0097] In practical applications, a preset value of 10 is set. After sorting the stable interaction data of all content in the first quantum set, 15 pieces of content with stable interaction data greater than 10 are selected. These pieces of content include "lipstick swatch videos" (16.391), "celebrity lipstick recommendations" (22.56), and "affordable lipstick reviews" (19.21), etc. These pieces of content are grouped into the first interaction volume list corresponding to this product. The same operation is performed on the second quantum set to obtain the second interaction volume list corresponding to the competitor products.

[0098] This application generates weighted interaction data by combining the content weight sequence with the initial interaction data, introduces a heat conduction model to calculate the spread popularity between content, and uses a Kalman filter algorithm for dynamic smoothing, so that the generated interaction volume list can truly reflect the spread value and long-term trend of the content, thereby improving the stability and accuracy of the list.

[0099] Step 104: Match each content item in the interaction volume list with the preset self-media information database to determine the corresponding self-media information that generated each content item, and calculate the recommendation score of each self-media and the product based on the self-media information using a collaborative filtering recommendation algorithm.

[0100] The self-media information database refers to a pre-stored database containing self-media information and its related attributes, with each record in the database corresponding to a self-media outlet. Self-media information refers to various data related to the self-media pre-stored in the self-media information database, including the self-media identifier, nickname, number of followers, content category, historical content records, and historical interaction data. The recommendation score is a numerical value calculated using a collaborative filtering recommendation algorithm to quantify the degree of matching between the self-media and the product; a higher score indicates that the self-media is more suitable for cooperation with the product.

[0101] In this embodiment, step 104 includes the following process:

[0102] Step 1041: Query the self-media information corresponding to each content item in the first interaction volume list and the second interaction volume list from the self-media information database, wherein the self-media information includes a self-media identifier.

[0103] In step 1041, the self-media identifier refers to the identity code used to uniquely identify each self-media, which can be numbers, letters, or a combination of both.

[0104] In this embodiment of the application, each content item in the first interaction volume list and the second interaction volume list is first traversed. Based on the publisher information of each content item, a search is performed in the self-media information database to find the self-media identity information corresponding to the content item. The self-media identifier is extracted from the found self-media identity information to establish the correspondence between the content item and the self-media identifier.

[0105] In practical applications, taking the "lipstick swatch video" content item in the first interaction volume list as an example, based on the publisher account "Beauty Self-Media Xiao A" of this content item, a search is conducted in the self-media information database, and the corresponding self-media identity information is found, including self-media identifier D1001, self-media nickname "Beauty Self-Media Xiao A", number of followers 850,000, and content category beauty. The self-media identifier D1001 is extracted from this information.

[0106] Step 1042: Extract topic keywords from the content entries, construct content topic vectors, and use the latent Dirichlet assignment model to perform cluster analysis on the content topic vectors to generate content topic distribution.

[0107] In step 1042, the topic keywords refer to words that can represent the core theme of the content extracted from the title, description or body of the content item; the content topic vector refers to the result of mapping the topic keywords of the content item into a numerical vector according to a preset thesaurus, and this vector is used to quantitatively represent the topic category to which the content belongs.

[0108] The Latent Dirichlet Allocation Model (LDA) is a probabilistic generative model used to mine latent topics from text collections. This model infers the topic distribution of documents by analyzing the co-occurrence patterns of words across different documents. This application does not impose specific limitations on the model type, internal structure design, parameter design, or training process of the LDA; these can be set according to actual circumstances.

[0109] Content topic distribution refers to the probability distribution of each content item belonging to various topics, calculated using the Latent Dirichlet Allocation Model. This distribution is a multi-dimensional vector, with each dimension corresponding to the probability value of a topic.

[0110] In this embodiment, text analysis is performed on each content item in the first and second interaction lists to extract keywords that represent the core theme of the content from its title and description. These keywords are then mapped into numerical vectors to obtain the content theme vector for each content item. All content theme vectors are input into a pre-trained latent Dirichlet allocation model, and the probability value of each content item belonging to each preset theme is calculated through model iteration. These probability values ​​are then used to form the content theme distribution of the content item.

[0111] In practical applications, text analysis is performed on the titles and descriptions of content items related to "lipstick swatch videos" to extract keywords such as "lipstick," "swatches," "lip makeup," and "lasting power." These keywords are then mapped into numerical vectors to obtain the content topic vector for this content item. This vector is input into a latent Dirichlet assignment model, which pre-defines five topic categories: beauty tutorials, product reviews, daily sharing, fashion styling, and entertainment / humor. The model calculates the following content topic distributions for this content item: beauty tutorials (0.45 probability), product reviews (0.35 probability), daily sharing (0.15 probability), fashion styling (0.03 probability), and entertainment / humor (0.02 probability).

[0112] Step 1043: Using the self-media identifier as the row and the content item as the column, construct a basic interaction matrix based on the weighted interaction data of each content item in the first interaction volume list and the second interaction volume list.

[0113] In step 1043, the basic interaction matrix is ​​a matrix that represents the interaction relationship between the self-media and its published content in the form of a two-dimensional table. The rows of the matrix represent self-media identifiers, the columns represent content items, and each element value in the matrix represents the weighted interaction data of the corresponding content item published by the corresponding self-media.

[0114] In this embodiment of the application, the self-media identifiers and weighted interaction data of all content items in the first interaction list and the second interaction list are collected. The self-media identifiers are used as the row indexes of the matrix, and the content items are used as the column indexes of the matrix. The weighted interaction data of each content item is filled into the intersection of the corresponding row and column to construct the basic interaction matrix.

[0115] In practical applications, the first and second interaction lists contain a total of 30 content items, involving 20 self-media accounts. These 20 self-media accounts are used as the rows of the basic interaction matrix, and the 30 content items are used as the columns of the basic interaction matrix. The weighted interaction data of each content item is filled into the intersection of the row of the corresponding self-media account and the column of the content item, thus constructing a 20-row, 30-column basic interaction matrix. The element value of the "lipstick swatch video" content item at the intersection of the row of self-media account D1001 and the column of the content item is 16.391.

[0116] Step 1044: Expand the dimensions of the basic interaction matrix according to the content topic distribution to generate an enhanced interaction matrix, which includes a topic relevance dimension.

[0117] In step 1044, the enhanced interaction matrix refers to the extended matrix formed by adding a topic relevance dimension to the basic interaction matrix. In addition to retaining the original self-media-content interaction relationship, this matrix also adds information on the degree of correlation between the self-media published content and each topic category. The topic relevance dimension refers to the vector dimension added to the enhanced interaction matrix to represent the degree of correlation between the self-media and each content topic. This dimension is composed of the average probability value of all content items published by the self-media on a specific topic. Each topic corresponds to an independent dimension. The larger the dimension value, the more obvious the self-media's content creation tendency in the corresponding topic area.

[0118] In this embodiment of the application, based on the content theme distribution of each content item, the average theme probability of each self-media in each theme category is calculated to obtain the theme preference vector of each self-media; the basic interaction matrix is ​​concatenated with the theme preference vector of the self-media to form an enhanced interaction matrix, each row of which contains the basic interaction data of the self-media and the theme preference value of the self-media in each theme category.

[0119] In practical application, for the self-media identifier D1001, it published three pieces of content in the first and second interaction lists: "Lipstick Swatch Video", "Foundation Review", and "Eye Makeup Tutorial". The average distribution of the content themes of these three pieces of content yielded the self-media's theme preference vector as follows: Makeup Tutorial Theme 0.5, Product Review Theme 0.3, Daily Sharing Theme 0.15, Fashion Matching Theme 0.03, and Entertainment and Humor Theme 0.02. This theme preference vector was then concatenated with the 30 weighted interaction data values ​​of the row containing the self-media identifier D1001 in the basic interaction matrix to form an enhanced interaction matrix row vector containing 35 elements.

[0120] Step 1045: Apply the collaborative filtering recommendation algorithm to calculate the multidimensional preference similarity between different self-media identifiers for the enhanced interaction matrix.

[0121] In step 1045, the multidimensional preference similarity refers to the numerical value calculated by the collaborative filtering recommendation algorithm to quantify the degree of consistency of preferences between two self-media in multiple dimensions. This value is determined based on the similarity between the row vectors of the two self-media in the enhanced interaction matrix.

[0122] In this embodiment of the application, the cosine similarity calculation method in the collaborative filtering recommendation algorithm is adopted. For each pair of self-media identifiers in the enhanced interaction matrix, the cosine similarity of the row vectors corresponding to the two self-media identifiers is calculated to obtain the multi-dimensional preference similarity between the two self-media. This process is repeated to calculate the similarity between all pairs of self-media to form a self-media similarity matrix.

[0123] In practical applications, for self-media identifiers D1001 and D1002, their row vectors in the enhanced interaction matrix are obtained respectively. The row vector of D1001 is [16.391, 8.245, 12.673, ..., 0.5, 0.3, 0.15, 0.03, 0.02], and the row vector of D1002 is [0, 14.562, 0, ..., 0.2, 0.6, 0.1, 0.05, 0.05]. The cosine value of the angle between these two vectors is calculated using the cosine similarity formula, and the multidimensional preference similarity between D1001 and D1002 is 0.76.

[0124] It should be noted that the specific form of the cosine similarity formula is not limited in the embodiments of this application, and can be set accordingly according to the actual situation.

[0125] Step 1046: Based on the multidimensional preference similarity and the historical interaction between the self-media identifier and the product content, predict the recommendation score of each self-media identifier for the product content.

[0126] In step 1046, historical interaction information refers to the record of content related to this product in the content published by the self-media in the past. This record includes the content items of this product published by the self-media and their interaction data.

[0127] In this embodiment of the application, for each self-media identifier, the top K self-media with the highest similarity to the self-media identifier are identified, and the weighted interaction data of these K self-media related to the content of this product in the historical interactions are obtained. The weighted interaction data is weighted and averaged according to the similarity to obtain the recommendation score of the self-media identifier for the content of this product. The same operation is performed on all self-media identifiers in turn to obtain the recommendation scores of all self-media.

[0128] In practical application, for the self-media identifier D1001, the top 3 self-media with the highest similarity are found: D1002 (0.76 similarity), D1005 (0.68 similarity), and D1008 (0.62 similarity). The weighted interaction data related to this product's content in the historical interactions of these three self-media are as follows: D1002 has two pieces of content related to this product, with weighted interaction data of 12.45 and 8.92 respectively. The average of these two data points is taken as the representative value for D1002, i.e., (12.45+8.92) / 2=10.685. D1005 has one piece of content related to this product, with a weighted interaction data of 15.23. D1008 has no content related to this product; its contribution value is 0. Therefore, the recommendation score calculation process for D1001 is as follows: The recommended score for D1001 was calculated to be... .

[0129] This application enhances the interaction matrix by associating and matching the interaction volume list with the self-media information database, introducing content theme analysis, and using a collaborative filtering recommendation algorithm to calculate multi-dimensional preference similarity. Based on content theme features and the historical behavior of self-media, it can accurately predict the matching degree between self-media and the product, thereby improving the accuracy and targeting of self-media recommendations.

[0130] Step 105: Using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the growth rate of followers of the self-media, and the verticality of the content, evaluate the effectiveness of the self-media cooperation between the product and the competitor on the content platform, and generate a self-media recommendation list based on the evaluation results.

[0131] Among them, the entropy weight method is a multi-criteria decision-making method that determines the weight of indicators based on the degree of dispersion of indicator data. This method measures the degree of variation of indicators by calculating the information entropy of each indicator. The smaller the information entropy, the greater the amount of information provided by the indicator, and the higher the corresponding weight.

[0132] Multi-criteria decision-making refers to a decision-making method that comprehensively evaluates and ranks alternatives under multiple conflicting evaluation indicators; follower growth rate refers to the percentage increase in the number of followers of a self-media within a specific time period, which reflects the speed of audience expansion of the self-media; content verticality refers to the degree of concentration of the content published by a self-media in a specific topic area, which reflects the degree of focus of the self-media in a professional field.

[0133] The explanations and specific implementation methods of the entropy weight method and multi-criteria decision-making method can be found in relevant technologies, and will not be elaborated here.

[0134] The self-media recommendation list is a sequence formed by sorting multiple self-media icons from high to low according to their comprehensive evaluation scores. Each self-media icon in the sequence corresponds to a specific self-media. The order of the sequence reflects the priority level of these self-media in cooperation based on a comprehensive consideration of three evaluation indicators: recommendation score, follower growth rate, and content verticality.

[0135] In this embodiment, step 105 includes the following process:

[0136] Step 1051: Obtain the follower growth rate and content verticality corresponding to each self-media identifier.

[0137] In this embodiment, the historical data of the number of followers corresponding to each self-media identifier is first read from the self-media information database, and the ratio of the increase in followers within a specified time period to the initial number of followers is calculated to obtain the follower growth rate corresponding to each self-media identifier. At the same time, based on the topic distribution data of each self-media identifier in its historical published content, the degree of concentration of its content on each topic is calculated to obtain the content verticality corresponding to each self-media identifier.

[0138] Step 1052: Construct an initial evaluation matrix, in which the behavior of the initial evaluation matrix is ​​identified by self-media, and the three evaluation indicators are recommendation score, follower growth rate and content verticality.

[0139] In step 1052, the initial evaluation matrix refers to a matrix that represents self-media and their evaluation index values ​​in the form of a two-dimensional table. The rows of the matrix represent different self-media identifiers, and the columns represent the three evaluation indicators: recommendation score, follower growth rate, and content verticality. Each element value in the matrix represents the specific value of the corresponding self-media on the corresponding indicator.

[0140] In this embodiment of the application, the recommendation score, follower growth rate and content verticality of all self-media identifiers are collected. The self-media identifier is used as the row index of the matrix, and the recommendation score, follower growth rate and content verticality are used as the column index of the matrix. The three index values ​​of each self-media are filled into the intersection of the corresponding row and column to construct an initial evaluation matrix.

[0141] Step 1053: Identify outlier data points in the initial evaluation matrix, correct the outlier data points, and generate a standard evaluation matrix.

[0142] In step 1053, outlier data points refer to index values ​​in the initial evaluation matrix that are significantly different from the distribution of other data points. These values ​​may deviate from the normal range due to data collection errors or special events. The standard evaluation matrix refers to the normalized evaluation matrix obtained after identifying and correcting outlier data points in the initial evaluation matrix.

[0143] In this embodiment of the application, the statistical analysis of each column of index data in the initial evaluation matrix is ​​performed to calculate the mean and standard deviation of each index. Values ​​that deviate from the mean by more than a preset multiple of the standard deviation are marked as abnormal data points. For the identified abnormal data points, the mean of the index or the interpolation of adjacent data points is used to replace them to complete the correction process of the abnormal data points and obtain the standard evaluation matrix.

[0144] Step 1054: For the standard evaluation matrix, calculate the information entropy of each evaluation index, and calculate the corresponding entropy weight based on the information entropy of each evaluation index.

[0145] In step 1054, information entropy refers to a metric used to measure the dispersion of indicator data. The larger the value, the smaller the degree of variation of the indicator data and the less information it provides. Entropy weight refers to the indicator weight calculated based on information entropy. This weight reflects the importance of the indicator in the comprehensive evaluation. The smaller the information entropy, the larger the entropy weight of the indicator.

[0146] In this embodiment, the index data in each column of the standard evaluation matrix is ​​normalized to obtain the probability distribution of each index; the information entropy value of each index is calculated based on the probability distribution, and the formula for calculating the information entropy is as follows: ,in Indicates the first Information entropy of the indicator Indicates the first The first evaluation object, the first The normalized probability value of the item indicator. This represents the total number of evaluation objects; the entropy weight is calculated based on the information entropy value of each indicator, and the formula for calculating the entropy weight is as follows: ,in Indicates the first Entropy weight of the indicator This indicates the total number of evaluation indicators.

[0147] Step 1055: Introduce fuzzy hierarchical analysis and, in conjunction with preset importance preferences, adjust the entropy weights to generate the final weights.

[0148] In step 1055, the preset importance preference refers to the relative importance of various evaluation indicators pre-set by the decision-maker according to business needs, and the final weight refers to the indicator weights used for comprehensive evaluation determined by combining the entropy weight method and the fuzzy hierarchical analysis method.

[0149] In this embodiment of the application, a fuzzy judgment matrix between each evaluation index is constructed according to a preset importance preference, and the subjective weight of each index is calculated by the fuzzy hierarchical analysis method. The objective weight obtained by the entropy weight method and the subjective weight obtained by the fuzzy hierarchical analysis method are weighted and combined to obtain the final weight of each index.

[0150] Step 1056: Based on the final weight, perform a weighted summation of the evaluation index values ​​corresponding to each self-media identifier in the standard evaluation matrix to obtain the comprehensive evaluation value of each self-media identifier, and sort all self-media identifiers from high to low according to the comprehensive evaluation value, and generate a self-media recommendation list based on the sorting result.

[0151] In step 1056, the comprehensive evaluation value refers to the comprehensive score obtained by multiplying the values ​​of the self-media on each evaluation indicator by the corresponding final weight and summing them. This score is used to measure the overall cooperative value of the self-media.

[0152] In this embodiment of the application, the recommendation score, follower growth rate and content verticality corresponding to each self-media identifier in the standard evaluation matrix are multiplied by their respective final weights, and the three product results are added together to obtain the comprehensive evaluation value of the self-media identifier; all self-media identifiers are arranged in descending order of comprehensive evaluation value, and the top-ranked self-media identifiers are selected to form a self-media recommendation list.

[0153] In this embodiment, after step 105, the following process is also included:

[0154] B1: Using the multi-armed slot machine algorithm in reinforcement learning, based on historical collaborative data, calculate dynamic weight values ​​for each self-media in the self-media recommendation list, forming a self-media list with dynamic weight values.

[0155] In step B1, the multi-armed slot machine algorithm is a reinforcement learning algorithm used to dynamically allocate resources among multiple options. This algorithm maximizes cumulative gains by balancing the exploration of new options with the utilization of known good options. The dynamic weight value refers to the self-media priority value that is dynamically adjusted by the multi-armed slot machine algorithm based on historical cooperation results. This value is used to guide the allocation of subsequent cooperation resources. The self-media list with dynamic weight values ​​refers to the list formed by adding dynamic weight values ​​to each self-media based on the self-media recommendation list.

[0156] The explanation and specific implementation of the multi-armed slot machine algorithm can be found in relevant technologies, and will not be elaborated here.

[0157] In this embodiment, each self-media in the self-media recommendation list is first considered as one arm of a multi-armed slot machine algorithm. The historical average revenue and number of collaborations for each self-media are calculated based on its historical collaboration data. Then, the upper bound strategy of the confidence interval in the multi-armed slot machine algorithm is applied to calculate a dynamic weight value for each self-media. The formula for this dynamic weight value is as follows: ,in Indicates the first The dynamic weight value of a self-media account. Indicates the first The historical average revenue of a self-media account Indicates the exploration coefficient. Indicates the total number of collaborations. Indicates the first The number of historical collaborations of each self-media outlet is calculated; the calculated dynamic weight value is then appended to the information of each self-media outlet in the self-media recommendation list, forming a self-media list with dynamic weight values.

[0158] B2: Introduce a community detection algorithm based on graph neural networks to divide the self-media in the self-media list into communities, in order to identify potential cooperative relationships between self-media and generate a self-media relationship graph.

[0159] In step B2, a graph neural network is a deep learning model that can directly process graph structure data. This model learns the feature representation of nodes by aggregating information from neighboring nodes.

[0160] This application does not impose specific limitations on the type of graph neural network, the structural design of its internal structure, the parameter design, the training process, etc., and corresponding settings can be made according to the actual situation.

[0161] Community detection algorithms are algorithms used to divide nodes in a network into several sets of nodes with tight internal connections and sparse external connections. The explanation and specific implementation of community detection algorithms can be found in relevant technologies, which will not be elaborated here.

[0162] Potential cooperative relationships refer to the possible relationships between self-media such as mutual referencing, joint cooperation, or audience overlap. A self-media relationship graph is a network graph structure built with self-media identifiers as nodes and various relationships between self-media as edges.

[0163] In this embodiment, interaction data between each self-media in the self-media list is collected. This interaction data includes the number of times they reference each other and the number of times they appear together. An initial self-media relationship network is constructed using self-media identifiers as nodes and interaction data as edge weights. The initial self-media relationship network is then input into a pre-trained graph neural network model, and the feature vector of each self-media is learned through multi-layer graph convolution operations of the model. A community detection algorithm is applied to perform cluster analysis on the learned self-media feature vectors, and self-media with similar features are divided into the same community, generating a self-media relationship graph with community labels.

[0164] B3: Based on the aforementioned self-media relationship diagram, and combining the content verticality and follower overlap of the self-media, calculate the synergistic effect score among the self-media.

[0165] In step B3, follower overlap refers to the proportion of the number of followers shared by two self-media accounts to their respective total followers. This value reflects the degree of overlap in the audiences of the two self-media accounts. Synergy score is a value used to quantify the additional effects that may be generated when multiple self-media accounts collaborate. The higher the score, the greater the overall effect that multiple self-media accounts may produce.

[0166] In this embodiment of the application, the community tags of each self-media in the self-media relationship graph are used to identify self-media pairs belonging to the same community. For each pair of self-media belonging to the same community, their content verticality value and follower overlap value are obtained. The product of the content verticality and the follower overlap are weighted and combined to obtain the synergy effect score of the pair of self-media. The synergy effect scores of all self-media pairs in the same community are summarized to obtain the synergy effect score of each community.

[0167] B4: Based on the synergy score and the list of self-media, determine the final collaborative recommendation result.

[0168] In step B4, the final cooperation recommendation result refers to the cooperation plan determined after comprehensively considering the individual value of the self-media, dynamic weight, and synergy effect. This result includes a list of self-media identifiers recommended for cooperation and suggestions on the cooperation methods.

[0169] In this embodiment, each self-media in the list of self-media with dynamic weight values ​​is sorted from high to low according to the dynamic weight value, and the top-ranked self-media with the highest dynamic weight values ​​are selected as the basic recommendation set; from the basic recommendation set, self-media combinations belonging to the same community and with high synergy effect scores are identified, and these combinations are marked as priority recommendation partners; the remaining self-media in the basic recommendation set are combined with the recommended self-media combinations to form the final cooperation recommendation result, which includes a list of recommended self-media identifiers and cooperation suggestions for self-media combinations in the same community.

[0170] This application uses a multi-criteria decision-making method based on entropy weighting to comprehensively evaluate recommendation scores, follower growth rate, and content verticality. It introduces anomaly data identification and fuzzy hierarchical analysis to optimize weight allocation, and combines multi-armed slot machine algorithm and community detection algorithm to dynamically adjust and collaboratively optimize the recommendation results, thereby improving the comprehensiveness of the evaluation of self-media cooperation effectiveness and the scientific nature of the recommendation results.

[0171] Figure 3 A schematic diagram of the structure of a self-media collaboration effectiveness evaluation and recommendation system based on an interaction volume list is provided in this application embodiment, as shown below. Figure 3 As shown, the system includes:

[0172] The acquisition module 31 is used to acquire multi-dimensional volume data of this product and at least one competitor on at least one content platform, and to receive filtering conditions input by the user.

[0173] The filtering module 32 is used to filter and aggregate the multi-dimensional volume data based on the filtering conditions, generate volume sets corresponding to the product and competitors, and use the PageRank algorithm to sort the importance of the content in the volume sets to generate a content weight sequence.

[0174] The generation module 33 is used to generate weighted interaction data based on the content weight sequence and the initial interaction data of each content in the sound quantum set, and to generate an interaction volume list for the product and the competitor based on the weighted interaction data.

[0175] The matching module 34 is used to associate and match each content item in the interaction volume list with a preset self-media information database to determine the corresponding self-media information that generated each content item, and to calculate the recommendation score of each self-media and the product based on the self-media information using a collaborative filtering recommendation algorithm.

[0176] Evaluation module 35 is used to evaluate the effectiveness of the product's cooperation with the competitor's self-media on the content platform using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the self-media follower growth rate and content verticality, and to generate a self-media recommendation list based on the evaluation results.

[0177] The self-media collaboration effectiveness evaluation and recommendation system based on the interaction volume list in this application embodiment is used to implement the aforementioned self-media collaboration effectiveness evaluation and recommendation method based on the interaction volume list. Therefore, the specific implementation of the self-media collaboration effectiveness evaluation and recommendation system based on the interaction volume list can be found in the embodiment section of the self-media collaboration effectiveness evaluation and recommendation method based on the interaction volume list above. The specific implementation can be referred to the description of the corresponding embodiment, and will not be repeated here.

[0178] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the self-media collaboration effectiveness evaluation and recommendation method based on the interaction volume list described above.

[0179] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for evaluating and recommending the effectiveness of self-media collaboration based on an interaction volume list.

[0180] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0181] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list.

[0182] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0183] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0184] The foregoing has provided a detailed description of a method and system for evaluating and recommending the effectiveness of self-media collaboration based on an interaction volume list, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for evaluating and recommending the effectiveness of self-media collaborations based on an interaction volume list, characterized in that, include: Obtain multi-dimensional volume data of this product and at least one competitor on at least one content platform, and receive filtering conditions input by users; Based on the filtering conditions, the multi-dimensional volume data is filtered and aggregated to generate volume sets corresponding to this product and competitors. The PageRank algorithm is used to sort the importance of the content in the volume sets and generate a content weight sequence. Based on the content weight sequence and the initial interaction data of each content in the quantum sound set, weighted interaction data is generated, and based on the weighted interaction data, interaction volume lists corresponding to the product and the competitor are generated respectively. Each content item in the interaction volume list is associated and matched with a preset self-media information database to determine the corresponding self-media information that generated each content item. Based on the self-media information, a collaborative filtering recommendation algorithm is used to calculate the recommendation score of each self-media and the product. Using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the growth rate of followers of the self-media, and the verticality of the content, the effectiveness of the self-media cooperation between the product and the competitor on the content platform is evaluated, and a self-media recommendation list is generated based on the evaluation results. The step involves generating weighted interaction data based on the content weight sequence and the initial interaction data of each content item in the sound quantum set, and then generating interaction volume lists for the product and its competitors based on the weighted interaction data, including: For each item in the first and second quantum sets, obtain the initial interaction data of the corresponding item, and find the importance value of the corresponding item from the content weight sequence; Using an energy propagation algorithm based on a heat conduction model, the heat conduction value of each content node is calculated according to the reference relationship between content. The heat conduction value reflects the heat spread obtained by the content from other high-weight content. The initial interaction data, importance value, and thermal conductivity value are weighted and combined to obtain weighted interaction data; The Kalman filter algorithm is used to dynamically smooth and estimate the state of the weighted interaction data to generate stable interaction data. The stable interaction data of all contents in the first and second acoustic quantum sets are sorted from high to low, and multiple contents with stable interaction data exceeding a preset value are selected from the sorting results to form the first interaction volume list corresponding to the product and the second interaction volume list corresponding to the competitor.

2. The method according to claim 1, characterized in that, The step of using the Kalman filter algorithm to dynamically smooth and estimate the state of the weighted interaction data to generate stable interaction data includes: Define a state vector and establish a state change equation, which describes the rule by which the state vector changes over time. Based on the state change equation, the predicted state value at the current moment is calculated using the state data from the previous moment. The weighted interactive data is used as the observation value, and the Kalman gain value is calculated. Based on the predicted state value, the observed value, and the Kalman gain value, the optimal state estimate for the current time moment is calculated, and the interaction value component in the optimal state estimate is output as stable interaction data.

3. The method according to claim 1, characterized in that, The step of associating and matching each content item in the interaction volume list with a preset self-media information database to determine the corresponding self-media information that generated each content item, and calculating the recommendation score of each self-media and the product using a collaborative filtering recommendation algorithm based on the self-media information, includes: The self-media information is retrieved from the self-media information database. The self-media information includes a self-media identifier. Thematic keywords are extracted from the content entries to construct content theme vectors. Then, the content theme vectors are clustered using the latent Dirichlet assignment model to generate content theme distribution. Using self-media identifiers as rows and content items as columns, a basic interaction matrix is ​​constructed based on the weighted interaction data of each content item in the first and second interaction lists; Based on the content topic distribution, the basic interaction matrix is ​​expanded in dimensions to generate an enhanced interaction matrix, which includes a topic relevance dimension. A collaborative filtering recommendation algorithm is applied to calculate the multidimensional preference similarity between different self-media identifiers for the enhanced interaction matrix; Based on the multidimensional preference similarity and the historical interaction between the self-media identifier and the product content, the recommendation score of each self-media identifier for the product content is predicted.

4. The method according to claim 1, characterized in that, The multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the self-media follower growth rate, and content verticality, evaluates the effectiveness of the product's self-media cooperation with the competitor on the content platform. Based on the evaluation results, a self-media recommendation list is generated, including: Get the follower growth rate and content verticality corresponding to each self-media identifier; An initial evaluation matrix is ​​constructed, which identifies the behavior of self-media and lists three evaluation indicators: recommendation score, follower growth rate, and content verticality. Identify outlier data points in the initial evaluation matrix, correct the outlier data points, and generate a standard evaluation matrix. For the standard evaluation matrix, calculate the information entropy of each evaluation indicator, and calculate the corresponding entropy weight based on the information entropy of each evaluation indicator; By introducing fuzzy hierarchical analysis and combining it with preset importance preferences, the entropy weights are adjusted to generate the final weights; Based on the final weight, the evaluation index values ​​corresponding to each self-media identifier in the standard evaluation matrix are weighted and summed to obtain the comprehensive evaluation value of each self-media identifier. All self-media identifiers are then sorted from high to low according to the comprehensive evaluation value, and a self-media recommendation list is generated based on the sorting results.

5. The method according to claim 1, characterized in that, Based on the filtering conditions, the multi-dimensional volume data is filtered and aggregated to generate volume sets corresponding to the product and competitors. The PageRank algorithm is then used to rank the content in the volume sets by importance, generating a content weight sequence, including: Based on the time range and content platform in the filtering conditions, extract multiple original content records corresponding to this product and each competitor from the multi-dimensional volume data; Based on the product categories in the filtering conditions, the multiple original content records are categorized, and the interactive data belonging to the same content are accumulated to generate the first quantum set of the product and the second quantum set of the competitor. Construct content reference graphs for the first quantum set and the second quantum set respectively. Nodes in the content reference graph represent content, and directed edges between nodes represent reference relationships between content. The PageRank algorithm is applied to iteratively calculate the importance score of each node in the content reference graph; The importance scores of all nodes are sorted according to the nodes to form a content weight sequence.

6. The method according to claim 1, characterized in that, After generating the self-media recommendation list based on the evaluation results, it also includes: Using the multi-armed slot machine algorithm in reinforcement learning, based on historical collaborative data, a dynamic weight value is calculated for each self-media in the self-media recommendation list, forming a self-media list with dynamic weight values; A community detection algorithm based on graph neural networks is introduced to divide the self-media in the self-media list into communities in order to identify potential cooperative relationships between self-media and generate a self-media relationship graph. Based on the aforementioned self-media relationship diagram, and combining the content verticality and follower overlap of the self-media, a synergistic effect score among the self-media is calculated. The final collaborative recommendation result is determined based on the synergy score and the list of self-media.

7. A self-media collaboration effectiveness evaluation and recommendation system based on an interaction volume list, characterized in that, include: The acquisition module is used to acquire multi-dimensional volume data of this product and at least one competitor on at least one content platform, and to receive filtering conditions input by the user. The filtering module is used to filter and aggregate the multi-dimensional volume data based on the filtering conditions, generate volume sets corresponding to the product and competitors, and use the PageRank algorithm to sort the importance of the content in the volume sets to generate a content weight sequence. The generation module is used to generate weighted interaction data based on the content weight sequence and the initial interaction data of each content in the sound quantum set, and to generate an interaction volume list for the product and the competitor based on the weighted interaction data. The matching module is used to associate and match each content item in the interaction volume list with a preset self-media information database to determine the corresponding self-media information that generated each content item, and to calculate the recommendation score of each self-media and the product based on the self-media information using a collaborative filtering recommendation algorithm. The evaluation module is used to evaluate the effectiveness of the product's cooperation with the competitor's self-media on the content platform using a multi-criteria decision-making method based on entropy weight, combined with the recommendation score, the self-media follower growth rate and content verticality. Based on the evaluation results, a self-media recommendation list is generated. The step involves generating weighted interaction data based on the content weight sequence and the initial interaction data of each content item in the sound quantum set, and then generating interaction volume lists for the product and its competitors based on the weighted interaction data, including: For each item in the first and second quantum sets, obtain the initial interaction data of the corresponding item, and find the importance value of the corresponding item from the content weight sequence; Using an energy propagation algorithm based on a heat conduction model, the heat conduction value of each content node is calculated according to the reference relationship between content. The heat conduction value reflects the heat spread obtained by the content from other high-weight content. The initial interaction data, importance value, and thermal conductivity value are weighted and combined to obtain weighted interaction data; The Kalman filter algorithm is used to dynamically smooth and estimate the state of the weighted interaction data to generate stable interaction data. The stable interaction data of all contents in the first and second acoustic quantum sets are sorted from high to low, and multiple contents with stable interaction data exceeding a preset value are selected from the sorting results to form the first interaction volume list corresponding to the product and the second interaction volume list corresponding to the competitor.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list as described in any one of claims 1 to 6, when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the self-media collaboration effectiveness evaluation and recommendation method based on an interaction volume list as described in any one of claims 1 to 6.

Images