Traffic quality scoring and filtering method for a data management platform

Abstract

The application discloses a traffic quality scoring and screening method of a data management platform, relates to the technical field of traffic quality, and is used for solving the problems of reduced traffic resource utilization efficiency and increased traffic resource waste rate, setting traffic monitoring time for newly published to-be-tested content, collecting recommended exposure data and calculating a recommended display coefficient, monitoring traffic diffusion data and performing ripple layering processing to obtain a reference propagation layer coefficient, combining the recommended display coefficient to calculate a traffic quality level, correcting default access volume based on the traffic quality level, monitoring comment data in real time during traffic distribution and analyzing comment level depth, simultaneously calling historical knowledge data to evaluate topic novelty features, comprehensively determining whether to mark the content as high-quality sediment content, realizing conversion determination from surface clicks to high-quality content sediment, distinguishing between short-term false high traffic caused by algorithm concentrated recommendation and natural propagation sediment type traffic, and improving traffic resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of traffic quality technology, and more specifically, to a method for traffic quality scoring and screening in a data management platform. Background Technology

[0002] With the rapid development of the internet information ecosystem, data management platforms, content communities, and knowledge-based social platforms have gradually become important channels for users to obtain information and exchange knowledge. Through algorithm recommendation mechanisms, these platforms expose newly published content in a concentrated manner within a short period of time in order to improve user activity and the speed of content dissemination.

[0003] The existing technology has the following shortcomings:

[0004] Currently, existing data management platforms typically rely on shallow metrics such as short-term exposure, clicks, and likes when recommending content and distributing traffic. These metrics fail to effectively reflect the depth of content dissemination and knowledge value, and lack a hierarchical evaluation mechanism for traffic quality. This results in reduced efficiency in traffic resource utilization and increased waste of traffic resources. Therefore, this paper proposes a traffic quality scoring and screening method for data management platforms.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a traffic quality scoring and filtering method for a data management platform. By employing a traffic quality hierarchical modeling algorithm based on the fusion of recommendation display coefficient and reference propagation layer coefficient, a dynamic scoring and high-quality filtering and distribution mechanism for content traffic is achieved to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for traffic quality scoring and screening in a data management platform, comprising the following steps:

[0008] Step S1: Set a traffic monitoring time for newly published content to be tested in the topic interest community. During the traffic monitoring time, detect the recommended exposure data and traffic diffusion data, and calculate the recommended display coefficient based on the recommended exposure data.

[0009] Step S2: Perform ripple layering processing on the traffic diffusion data to generate reference propagation layer coefficients, and merge the reference propagation layer coefficients with the recommendation display coefficients to generate the traffic quality level of the content to be tested;

[0010] Step S3: Access the traffic distribution database to obtain the default access volume, correct the default access volume based on the traffic quality level, distribute traffic to the content to be tested according to the correction result, and detect the comment data of the content to be tested in real time and analyze the comment level depth.

[0011] Step S4: Retrieve historical knowledge data from the topic interest community to analyze the topic novelty characteristics of the content to be tested, and combine the topic novelty characteristics with the depth of the comments to determine whether to mark the content to be tested as high-quality accumulated content.

[0012] In a preferred embodiment, in step S1, a preset traffic monitoring time is set for the newly released content to be tested, and the preset traffic monitoring time is divided into multiple monitoring moments.

[0013] The recommended exposure count of the content to be tested is monitored by the content exposure log. The recommended exposure count is the cumulative number of times the content to be tested is pushed to the user's visible interface.

[0014] The exposure fluctuation weighting factor for each monitoring time is calculated based on the maximum recommended exposure count for each monitoring time.

[0015] The recommended display coefficient is calculated based on the weighting factors of each exposure fluctuation and the recommended exposure times.

[0016] In a preferred embodiment, in step S1, the access content forwarding log monitors the forwarding events of the content under test;

[0017] Set hierarchical forwarding weights for forwarding events. Specifically, the hierarchical forwarding weight for the first forwarding event based on the content to be tested is set to 1.

[0018] When a user forwards the message again based on the initial forwarding event, the forwarding weights of different propagation levels are obtained by sequentially increasing the forwarding weights according to the forwarding path.

[0019] Forwarding events are weighted according to their forwarding levels to obtain traffic diffusion data for each forwarding level.

[0020] In a preferred embodiment, in step S2, at the end of the preset traffic monitoring time, the traffic diffusion data is statistically analyzed, and the traffic diffusion data of each forwarding level is combined into a traffic diffusion dataset in ascending order of level.

[0021] Retrieve the traffic diffusion threshold corresponding to each forwarding level in the diffusion threshold database and match it layer by layer with the traffic diffusion dataset;

[0022] If the traffic propagation data is greater than or equal to the traffic propagation threshold, the forwarding level will be marked as a valid propagation level.

[0023] If the traffic diffusion data is less than the traffic diffusion threshold, then the comparison of subsequent propagation levels will stop.

[0024] The number of effective propagation levels is used as the reference propagation level coefficient.

[0025] The traffic quality level is calculated by fusing the standardized reference propagation layer coefficient and the recommendation display coefficient using a nonlinear fusion algorithm.

[0026] In a preferred embodiment, in step S3, the distribution traffic database is accessed to obtain the default access volume corresponding to the content to be tested;

[0027] The default visit count is the baseline visit count set by the platform during the initial content release phase based on topic category, author's historical activity level, and default exposure strategy.

[0028] The default access volume is adjusted based on the traffic quality level to generate a revised access allocation: ;

[0029] in, This is the corrected access allocation. This is the default number of visits. For traffic quality levels, This represents the median of the traffic quality levels. This is the preset visitor volume correction factor.

[0030] In a preferred embodiment, in step S3, after traffic correction and distribution are completed, the comment data of the content to be tested is detected in real time, including the total number of comment chains and the number of comment chain levels;

[0031] The total number of comment chains refers to the total number of independent comment chains formed under the content to be tested based on the hierarchical relationship between comments and replies;

[0032] The number of levels in a comment chain refers to the depth of the hierarchy from the first-level comment node to the last-level reply node in a comment chain;

[0033] Calculate the comment hierarchy depth based on comment data: ;

[0034] in, For comment hierarchy depth, The total number of comment chains. For the number of levels in the comment chain, This is the index value for the comment chain.

[0035] In a preferred embodiment, in step S4, historical knowledge data of the topic interest community is retrieved. The historical knowledge data is a set of structured semantic data that the platform has stored in the knowledge database for a long time.

[0036] The topic tags, title keywords, and semantic features of the content to be tested are vectorized to generate the topic feature vector of the content to be tested.

[0037] Retrieve historical topics from historical knowledge data that have the same topic tags as the content to be tested, and generate a feature vector for each historical topic.

[0038] In a preferred embodiment, in step S4, the semantic similarity between the content to be tested and each historical topic is calculated using cosine similarity based on the topic feature vector of the content to be tested and the feature vector corresponding to each historical topic.

[0039] Compare the semantic similarity between the content to be tested and various historical topics, and select the maximum value as the historical similarity of the content to be tested.

[0040] Subtracting the historical similarity from 1 is taken as the topic novelty feature of the content to be tested.

[0041] In a preferred embodiment, in step S4, the topic novelty feature and the comment level depth are standardized to obtain the topic novelty factor and the comment level factor, respectively.

[0042] A sedimentation quality index is generated by weighted summation of the topic novelty factor and the comment level factor;

[0043] When the sedimentation quality index is greater than or equal to the preset sedimentation judgment threshold, the content to be tested is marked as high-quality sedimentation content.

[0044] When the sedimentation quality index is less than the preset sedimentation judgment threshold, the test content is determined to have not met the sedimentation requirements and will not be marked.

[0045] The technical effects and advantages of this invention are as follows:

[0046] This invention sets a traffic monitoring time for newly released content to be tested, collects recommended exposure data and calculates the recommended display coefficient, monitors traffic diffusion data and performs ripple layering processing to obtain the reference propagation layer coefficient, calculates the traffic quality level based on the recommended display coefficient, and corrects the default access volume based on the traffic quality level. During the traffic distribution process, it monitors comment data in real time and analyzes the comment level depth, while retrieving historical knowledge data to evaluate the novelty characteristics of the topic, and comprehensively determines whether the content should be marked as high-quality content for accumulation. This realizes the transformation judgment from surface clicks to high-quality content accumulation, distinguishes between short-term artificially high traffic caused by algorithm-driven concentrated recommendations and naturally propagated and accumulated traffic, and prioritizes traffic distribution for content with large knowledge increments and high discussion depth, thereby improving the efficiency of traffic resource utilization and promoting the long-term virtuous cycle of the knowledge ecosystem of topic interest communities. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the implementation of a traffic quality scoring and filtering method for a data management platform according to the present invention.

[0048] Figure 2 This is a schematic diagram illustrating the steps of a traffic quality scoring and filtering method for a data management platform according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] This invention sets a traffic monitoring time for newly released content to be tested, collects recommended exposure data and calculates the recommended display coefficient, monitors traffic diffusion data and performs ripple layering to obtain the reference propagation layer coefficient, calculates the traffic quality level based on the recommended display coefficient, and corrects the default access volume based on the traffic quality level. During the traffic distribution process, it monitors comment data in real time and analyzes the comment level depth, while retrieving historical knowledge data to evaluate the novelty characteristics of the topic. It comprehensively determines whether to mark the content as high-quality content for accumulation, realizing the transformation judgment from surface clicks to high-quality content accumulation. It distinguishes between short-term artificially high traffic caused by algorithm-driven concentrated recommendations and naturally propagated and accumulated traffic, and prioritizes traffic distribution for content with large knowledge increments and high discussion depth, thereby improving the efficiency of traffic resource utilization.

[0051] Example 1, such as Figures 1 to 2 As shown, a traffic quality scoring and filtering method for a data management platform includes the following steps:

[0052] Step S1: Set a traffic monitoring time for newly published content to be tested in the topic interest community. During the traffic monitoring time, detect the recommended exposure data and traffic diffusion data, and calculate the recommended display coefficient based on the recommended exposure data.

[0053] Step S2: Perform ripple layering processing on the traffic diffusion data to generate reference propagation layer coefficients, and merge the reference propagation layer coefficients with the recommendation display coefficients to generate the traffic quality level of the content to be tested;

[0054] Step S3: Access the traffic distribution database to obtain the default access volume, correct the default access volume based on the traffic quality level, distribute traffic to the content to be tested according to the correction result, and detect the comment data of the content to be tested in real time and analyze the comment level depth.

[0055] Step S4: Retrieve historical knowledge data from the topic interest community to analyze the topic novelty characteristics of the content to be tested, and combine the topic novelty characteristics with the depth of the comments to determine whether to mark the content to be tested as high-quality accumulated content.

[0056] The specific implementation is as follows:

[0057] In step S1, the topic interest community refers to the online interactive platform based on professional fields or vertical interests. Its user group is relatively small and highly professional. The content dissemination relies on the community's internal recommendation algorithm and circle diffusion mechanism. Because some titles attract clicks but the discussions are superficial, they only generate short-term popularity, while content with real knowledge value and in-depth discussion is difficult to gain sustained exposure.

[0058] In the topic-based interest community, a preset traffic monitoring time is set for newly released content to be tested, and the preset traffic monitoring time is divided into multiple monitoring periods.

[0059] Within the preset traffic monitoring period, the recommended exposure count of the content to be tested is monitored through the content exposure log. The recommended exposure count refers to the cumulative number of times the content to be tested is pushed to the user's visible interface.

[0060] The exposure fluctuation weight factor is calculated based on the number of recommended exposures at each monitoring time, reflecting the concentration of push behavior of the recommendation system within the preset traffic monitoring time. The specific steps are as follows:

[0061] Within the preset traffic monitoring period, the exposure fluctuation weighting factor is calculated using the maximum recommended number of exposures: ,in, To recommend the maximum number of exposures, For the first Recommended exposure count for each monitoring moment. For the first Recommended exposure count for each monitoring moment. For the first Exposure fluctuation weighting factor at each monitoring moment;

[0062] Exposure fluctuation weight factor is a coefficient used to characterize the stability of exposure changes of the recommendation algorithm over continuous monitoring time. It reflects the concentration and stability of the platform's recommendation system in pushing content to be tested. When the exposure fluctuation weight factor is close to 1, it indicates that the exposure changes of the content to be tested are small at each monitoring time, and the recommendation push is relatively balanced.

[0063] Calculate the recommendation display coefficient based on the exposure fluctuation weighting factor: Where n is the total number of monitoring times. The recommended display coefficient;

[0064] The higher the recommendation display coefficient, the higher the total recommended exposure of the content under test within the preset traffic monitoring time and the relatively stable distribution, indicating that the content under test has potential for wide dissemination; conversely, the lower the recommendation display coefficient, the more likely the content under test has inflated traffic or rapid decline in popularity.

[0065] Access content forwarding logs monitor forwarding events of the content under test. Forwarding events refer to the sharing, re-propagation, or referencing operations performed by users on the target content.

[0066] For detected forwarding events, hierarchical forwarding weights are set. Specifically, the hierarchical forwarding weight of the first-level forwarding event based on the content to be tested is set to 1. When the user forwards again based on the first forwarding event, the hierarchical forwarding weights are increased sequentially according to the forwarding path. For example, the hierarchical forwarding weight of the second level is 2, the hierarchical forwarding weight of the third level is 3, and so on, forming a multi-level propagation chain.

[0067] Among them, the first-level forwarding event refers to the initial forwarding behavior of a user directly forwarding the content to be tested after it has been published. It is not generated based on the forwarding links of other users and is located at the first level in the propagation link.

[0068] We obtain the traffic diffusion data of the content under test by weighting all detected forwarding events according to their hierarchical forwarding weights: ,in, This represents the forwarding weight of the i-th layer. Let i be the number of forwarding events at layer i. This represents the flow diffusion data for the i-th layer.

[0069] Traffic diffusion data reflects the speed at which the content under test spreads among user groups and is used to assess the diffusion trend of the content.

[0070] It should be explained that the preset traffic monitoring time can be set according to the user activity cycle of the topic interest community and the push rhythm of the platform recommendation algorithm; the content exposure log refers to the set of data records used to record the display records of the content to be tested being pushed in the user interface; the content forwarding log refers to the set of data records used to record the operation records of users' sharing, re-propagation, citation or forwarding behavior of the content to be tested.

[0071] In step S2, at the end of the preset flow monitoring time, the flow diffusion data is statistically analyzed, and the flow diffusion data of each layer is combined into a flow diffusion dataset in ascending order of the layers.

[0072] The traffic propagation threshold corresponding to each forwarding level in the propagation threshold database is retrieved and matched layer by layer with the traffic propagation dataset. Ripple layer processing is then used to generate reference propagation layer coefficients. The specific process is as follows:

[0073] The traffic diffusion data at each propagation level is compared sequentially with the corresponding traffic diffusion threshold:

[0074] If the traffic propagation data is greater than or equal to the traffic propagation threshold, the propagation level is marked as a valid forwarding level.

[0075] If the traffic diffusion data is less than the traffic diffusion threshold, then the comparison of subsequent forwarding levels will stop.

[0076] The number of effective forwarding layers is used as the reference propagation layer coefficient.

[0077] The citation propagation layer coefficient reflects the diffusion depth and activity of the tested content. The larger the citation propagation layer coefficient, the higher the diffusion contribution of the tested content in multiple propagation layers, the longer the propagation chain, the stronger the ability to reach across user circles, and the ability to continuously trigger secondary dissemination and extended discussions. The smaller the citation propagation layer coefficient, the lower the diffusion of the tested content, the shallower the propagation layer, the limited the activity of the propagation, the content popularity mainly depends on the initial exposure or algorithm recommendation, and the subsequent diffusion ability is weak.

[0078] It should be explained that ripple layering processing refers to abstracting the propagation process of the content to be tested into a water surface ripple structure with hierarchical progression. The initial propagation behavior is regarded as the ripple source point, and subsequent propagation behaviors form multiple diffusion layers in the propagation chain. The diffusion threshold database is a data set used to store threshold data corresponding to different propagation levels. The traffic diffusion threshold is the critical value for determining whether each propagation level constitutes effective diffusion. It is set by professionals based on community size, user activity cycle, and historical propagation model, and then stored in the diffusion threshold database. It will not be elaborated here.

[0079] After standardizing the reference propagation layer coefficient and the recommendation display coefficient separately, a nonlinear fusion algorithm is used to combine the reference propagation layer coefficient and the recommendation display coefficient to calculate the traffic quality level of the content under test: ,in, As a penalty item, For collaborative terms, , and As a preset adjustment factor, The reference propagation layer coefficient after standardization. The standardized recommendation display coefficient. For traffic quality levels;

[0080] When both the reference propagation layer coefficient and the recommendation display coefficient are high, the traffic quality level is improved through synergistic terms, reflecting the resonance amplification effect formed between algorithm exposure and natural propagation of the content under test;

[0081] When the recommendation display coefficient is high but the reference propagation layer coefficient is low, the penalty term is used to suppress the artificially high performance brought about by concentrated recommendations, reduce the traffic quality level of the content under test, and distinguish between naturally accumulated content and title-attractive content during the fusion calculation process, so as to achieve priority identification of truly disseminated content.

[0082] It should be explained that the standardization methods include, but are not limited to, standard linear transformation based on interval scaling, statistical Z-Score standardization method, or normalization method based on nonlinear mapping function. The application methods of standardization will not be elaborated here. The preset adjustment factor can be adjusted and configured according to the platform operation strategy or traffic resource allocation preference. The value range is 0 to 1. For example, in the community scenario that emphasizes knowledge accumulation and in-depth discussion, the preset adjustment factor corresponding to the citation propagation layer coefficient can be increased.

[0083] In step S3, the distribution traffic database is accessed to obtain the default access volume corresponding to the content to be tested. The default access volume is the benchmark access volume data set by the platform in the initial content release stage based on topic classification, author's historical activity and default exposure strategy. It is generated by the historical distribution records accumulated in the distribution traffic database over a long period of time and reflects the average access distribution level of similar content by the platform under no-intervention conditions.

[0084] It should be noted that the distribution traffic database is the core dataset within the data management platform used to record and store historical information related to content traffic allocation. It is used to support data retrieval in content recommendation, exposure control, and traffic correction. The default exposure strategy is a basic display mechanism set by the platform in the initial stage of content release based on content category, community topic popularity, and user interest model. It is used to ensure that all newly released content receives a fair basic exposure opportunity in the initial stage.

[0085] The default access volume is adjusted based on the traffic quality level to generate a revised access allocation. The adjustment process is implemented by setting a correction function, the expression of which is as follows:

[0086] ;

[0087] in, This is the corrected access allocation. This is the default number of visits. For traffic quality levels, This represents the median of the traffic quality levels. This is a preset access volume correction factor used to control the gain of high-quality content in traffic allocation.

[0088] It should be noted that, based on the symmetry and statistical equilibrium principle of the traffic quality level value range, the median of the value range is selected as the adjustment term in the correction function. The median of the traffic quality level indicates that the content quality is at the platform average level, neither high-quality nor low-quality content. When the traffic quality level is greater than the median, the correction term is positive, indicating that the content will receive additional distribution traffic compensation. When the traffic quality level is less than the median, the correction term is negative, indicating that the distribution weight of the content is reduced to suppress low-quality traffic.

[0089] The revised access allocation is the final access allocation for content entering the recommendation system, used to control the display frequency of the content in the recommendation list and exposure positions.

[0090] After traffic correction and distribution are completed, the comment data of the content to be tested is monitored in real time. The comment data includes the total number of comment chains generated according to the comment chain structure and the number of comment chain levels.

[0091] The total number of comment chains refers to the total number of independent comment chains formed under the content to be tested, based on the hierarchical relationship between comments and replies. Each comment chain starts with a primary comment (i.e., a comment that directly addresses the content itself) and includes all consecutive reply nodes at all levels under that primary comment and their derived child reply structures. When there are multiple primary comments under the content to be tested, each primary comment and all its corresponding replies together constitute an independent comment chain.

[0092] The comment chain hierarchy refers to the depth of a comment chain from the first comment node to the last reply node. When constructing the comment chain structure, the hierarchy is calculated recursively based on the referencing relationships between comments. The hierarchy of the first comment is denoted as 1, the hierarchy of a direct reply to the first comment is denoted as 2, and so on, until the last reply node in the comment chain. The comment chain hierarchy is used to characterize the extent of the extension of ideas and the depth of interaction within a single discussion path. The larger the hierarchy, the more thorough the exchange of views among users within the discussion chain and the deeper the topic extension.

[0093] The comment hierarchy depth is calculated based on comment data, reflecting the depth of the comment interaction structure. The expression is defined as follows:

[0094] ;

[0095] in, For comment hierarchy depth, The total number of comment chains. For the number of levels in the comment chain, This is the index value for the comment chain.

[0096] Comment depth is used to measure the actual depth of discussion and the duration of interaction among users, reflecting the breadth of community users' participation and the extent of their understanding of the content.

[0097] In step S4, historical knowledge data from the topic interest community is retrieved to perform topic novelty feature analysis on the content to be tested. The historical knowledge data is a set of structured semantic data that the platform has stored in the knowledge database for a long time. It includes the title, body text, topic tags, keyword vectors and related discussion summaries of the content published in the community, which are used to reflect the community's knowledge accumulation and content distribution in a specific topic area.

[0098] After vectorizing the topic tags, title keywords, and semantic features of the content to be tested, a topic feature vector is generated. Subsequently, historical topics with the same topic tags as the content to be tested are retrieved from historical knowledge data, and a feature vector corresponding to each historical topic is generated. The semantic similarity between the content to be tested and each historical topic is calculated using cosine similarity, and the calculation formula is as follows:

[0099] ;

[0100] in, The semantic similarity between the content to be tested and historical topics. and These are the inner product and the magnitude of the topic feature vectors of the content to be tested, respectively. and These are the inner product and the magnitude of the feature vectors corresponding to the historical topics, respectively.

[0101] It should be noted that vectorization refers to the process of converting the text information of the content to be tested, including topic tags, title keywords and semantic features of the body text, into numerical feature vectors through natural language processing and feature extraction. Specifically, the text content is segmented, stop words are removed and standardized, and then each word or phrase is mapped to a numerical vector of fixed dimensions.

[0102] Compare the semantic similarity between the content to be tested and various historical topics, and select the maximum value as the historical similarity of the content to be tested.

[0103] Subtracting historical similarity from 1 is taken as the topic novelty feature of the content to be tested;

[0104] The topic novelty feature reflects the degree of semantic difference between the content to be tested and historical knowledge data. The larger the value, the higher the novelty and originality of the content to be tested in terms of topic expression, viewpoint structure or knowledge coverage.

[0105] The topic novelty factor and the comment level depth are obtained by standardizing the topic novelty feature and the comment level factor respectively;

[0106] A weighted summation of the topic novelty factor and the comment level factor is used to generate a sedimentation quality index, which measures the comprehensive performance of content in terms of knowledge contribution and discussion depth. The calculation formula is as follows:

[0107] ;

[0108] in, The sedimentation quality index, For the novelty factor of the topic, As a comment level factor, and This is a weighting factor used to balance the influence of discussion depth and topic novelty on the evaluation of content accumulation value.

[0109] The precipitation quality index is compared with the preset precipitation judgment threshold:

[0110] When the sedimentation quality index is greater than or equal to the preset sedimentation judgment threshold, the content to be tested is marked as high-quality sedimentation content.

[0111] When the sedimentation quality index is less than the preset sedimentation judgment threshold, the test content is determined to have not met the sedimentation requirements and will not be marked.

[0112] It should be noted that the weighting coefficients are proportional parameters set when calculating the sedimentation quality index. These coefficients are based on historical data statistics from the platform, and their contribution is calculated through regression analysis of a large number of known high-quality content samples. Weighting coefficients are then assigned according to the contribution, and the following conditions must be met: To ensure that the sedimentation quality index remains normalized in terms of dimensions; the sedimentation judgment threshold is a critical judgment standard used to distinguish between high-quality sedimentation content and ordinary content. It is set by statistically analyzing the numerical distribution of the sedimentation quality index of high-quality content in historical samples, and selecting the percentile (such as the 75th percentile) in the numerical distribution as the sedimentation judgment threshold.

[0113] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0114] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0115] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0116] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0117] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for traffic quality scoring and filtering in a data management platform, characterized in that: Includes the following steps: Step S1: Set a traffic monitoring time for newly published content to be tested in the topic interest community. During the traffic monitoring time, detect the recommended exposure data and traffic diffusion data, and calculate the recommended display coefficient based on the recommended exposure data. Step S2: Perform ripple layering processing on the traffic diffusion data to generate reference propagation layer coefficients, and merge the reference propagation layer coefficients with the recommendation display coefficients to generate the traffic quality level of the content to be tested; Step S3: Access the traffic distribution database to obtain the default access volume, correct the default access volume based on the traffic quality level, distribute traffic to the content to be tested according to the correction result, and detect the comment data of the content to be tested in real time and analyze the comment level depth. Step S4: Retrieve historical knowledge data from the topic interest community to analyze the topic novelty characteristics of the content to be tested, and combine the topic novelty characteristics with the depth of the comments to determine whether to mark the content to be tested as high-quality accumulated content; In step S4, historical knowledge data from the topic interest community is retrieved. Historical knowledge data is a set of structured semantic data that the platform has stored in the knowledge database for a long time. The topic tags, title keywords, and semantic features of the content to be tested are vectorized to generate the topic feature vector of the content to be tested. Retrieve historical topics from historical knowledge data that have the same topic tags as the content to be tested, and generate feature vectors for each historical topic; In step S4, the semantic similarity between the content to be tested and each historical topic is calculated using cosine similarity based on the topic feature vector of the content to be tested and the feature vector corresponding to each historical topic. Compare the semantic similarity between the content to be tested and various historical topics, and select the maximum value as the historical similarity of the content to be tested. Subtracting historical similarity from 1 is taken as the topic novelty feature of the content to be tested; In step S4, the topic novelty feature and comment level depth are standardized to obtain the topic novelty factor and comment level factor, respectively. A sedimentation quality index is generated by weighted summation of the topic novelty factor and the comment level factor; When the sedimentation quality index is greater than or equal to the preset sedimentation judgment threshold, the content to be tested is marked as high-quality sedimentation content. When the sedimentation quality index is less than the preset sedimentation judgment threshold, the test content is determined to have not met the sedimentation requirements and will not be marked.

2. The traffic quality scoring and filtering method for a data management platform according to claim 1, characterized in that: In step S1, a preset traffic monitoring time is set for the newly released content to be tested, and the preset traffic monitoring time is divided into multiple monitoring moments; The recommended exposure count of the content to be tested is monitored by the content exposure log. The recommended exposure count is the cumulative number of times the content to be tested is pushed to the user's visible interface. The exposure fluctuation weighting factor for each monitoring time is calculated based on the maximum recommended exposure count for each monitoring time. The recommended display coefficient is calculated based on the weighting factors of each exposure fluctuation and the recommended exposure times.

3. The traffic quality scoring and screening method for a data management platform according to claim 1, characterized in that: In step S1, the access content forwarding log monitors the forwarding events of the content under test; Set hierarchical forwarding weights for forwarding events. Specifically, the hierarchical forwarding weight for the first forwarding event based on the content to be tested is set to 1. When a user forwards the message again based on the initial forwarding event, the forwarding weights of different propagation levels are obtained by sequentially increasing the forwarding weights according to the forwarding path. Forwarding events are weighted according to their forwarding levels to obtain traffic diffusion data for each forwarding level.

4. The traffic quality scoring and screening method for a data management platform according to claim 3, characterized in that: In step S2, at the end of the preset traffic monitoring time, the traffic diffusion data is statistically analyzed, and the traffic diffusion data of each forwarding level is combined into a traffic diffusion dataset in ascending order of level. Retrieve the traffic diffusion threshold corresponding to each forwarding level in the diffusion threshold database and match it layer by layer with the traffic diffusion dataset; If the traffic propagation data is greater than or equal to the traffic propagation threshold, the forwarding level will be marked as a valid propagation level. If the traffic diffusion data is less than the traffic diffusion threshold, then the comparison of subsequent propagation levels will stop. The number of effective propagation levels is used as the reference propagation level coefficient. The traffic quality level is calculated by fusing the standardized reference propagation layer coefficient and the recommendation display coefficient using a nonlinear fusion algorithm.

5. The traffic quality scoring and screening method for a data management platform according to claim 1, characterized in that: In step S3, the distribution traffic database is accessed to obtain the default access volume corresponding to the content to be tested; The default visit volume is the baseline visit volume data set by the platform during the initial content release stage based on topic category, author's historical activity level, and default exposure strategy; The default access volume is adjusted based on the traffic quality level to generate a revised access allocation: ; in, This is the corrected access allocation. This is the default number of visits. For traffic quality levels, This represents the median of the traffic quality levels. This is the preset visitor volume correction factor.

6. The traffic quality scoring and screening method for a data management platform according to claim 1, characterized in that: In step S3, after traffic correction and distribution are completed, the comment data of the content to be tested is detected in real time, including the total number of comment chains and the number of comment chain levels; The total number of comment chains refers to the total number of independent comment chains formed under the content to be tested based on the hierarchical relationship between comments and replies; The number of levels in a comment chain refers to the depth of the hierarchy from the first-level comment node to the last-level reply node in a comment chain; Calculate the comment hierarchy depth based on comment data: ; in, For comment hierarchy depth, The total number of comment chains. For the number of levels in the comment chain, This is the index value for the comment chain.

Images