A new media product user portrait analysis system and method based on media big data

Abstract

The application discloses a new media product user portrait analysis system and method based on media big data, relates to the technical field of data processing, collects user behavior data and platform operation data related to the same news content in a plurality of media platforms, collects and normalizes user feature data and media platform feature data in the plurality of media platforms, obtains demand vectors between the media platforms and the users, identifies different user features, accumulates the difference degrees of the demand vectors between different media platforms based on the demand vectors of the same user in different media platforms, obtains the demand direction difference degrees of each media platform, collects the demand vectors of the users and the user demand direction difference degrees in all media platforms, performs vector clustering on the vectors in the user feature matrix, obtains the portrait of the group users, and aggregates the feature values based on the portrait of the group users to obtain the portrait of the individual users.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically a new media product user profiling analysis system and method based on media big data. Background Technology

[0002] With the rapid development of the new media industry, user profiling analysis is increasingly widely used in media big data. Integrating user behavior data from multiple platforms to gain in-depth insights into customer characteristics and generate unified user profiles has become crucial for improving recommendation accuracy and user experience. However, existing user profiling analysis methods still have many shortcomings in cross-media platform scenarios, particularly regarding the reusability of user profiles, the comprehensiveness of data collection, and the universality of analytical models.

[0003] Existing technologies often construct user profiles by acquiring target users' social media information and comments, combined with sentiment analysis and attribute prediction models. While this approach can improve the efficiency of potential user analysis, its data sources are primarily limited to social media platforms, failing to fully integrate multi-channel media data, thus limiting the comprehensiveness and accuracy of user profiles. Furthermore, this method does not adequately address the differences in user behavior across different media platforms, making it difficult to achieve efficient reuse of user profiles across platforms. Summary of the Invention

[0004] The purpose of this invention is to provide a user profiling analysis system and method for new media products based on media big data, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing user profiles of new media products based on media big data, the method comprising:

[0006] Step S1: Collect user behavior data, user activity data, and platform operation data related to the same news content from several media platforms;

[0007] Step S2: Quantify and extract data features from user behavior data, user activity data, and platform operation data to obtain user feature data and media platform feature data, respectively. User feature data includes user interaction and content preference, while media platform feature data includes user activity index and platform content influence.

[0008] Step S3: Collect and normalize user feature data and media platform feature data from several media platforms to obtain the demand vector between media platforms and users;

[0009] Step S4: Identify different user characteristics, and based on the demand vector of the same user on different media platforms, accumulate the difference in demand vectors between different media platforms to obtain the difference in demand direction of each media platform;

[0010] Step S5: Collect the user demand vectors and the differences in user demand directions across all media platforms to obtain a user feature matrix. Perform vector clustering on the vectors in the user feature matrix to obtain a profile of the group of users. Based on the profile of the group of users, aggregate feature values ​​to obtain an individual user profile.

[0011] Furthermore, step S1 includes:

[0012] Step S1-1: Extract feature words from a news article, encode the feature words or obtain word vectors for the feature words, and record the encoded or word vectors as the identification information of the news article. Collect the identification information of the news article to obtain the content identifier C. id ;

[0013] Step S1-2: Gather several media platforms to form a media platform set M, M = {M1, M2, M3, ..., M}. N}, where N represents the total number of media platforms, N is an integer greater than or equal to 2, and M1, M2, M3... and M... N These represent the 1st, 2nd, 3rd, ..., and Nth media platforms, based on the content identifier C. id In different media platforms, interaction data between users and news content with the same content identifier is collected. The interaction data includes raw behavioral data and raw activity data.

[0014] Step S1-3: Obtain the j-th media platform M from the media platform set. j Behavioral data B of the k-th user Uk within a unit time range. jk and active data A jk And the amount of exposure E of a certain news content on the j-th media platform within a unit time range. j Click-through rate (V) j ;

[0015] Behavioral data B jk Includes: Number of clicks T click Duration of stay D dwell Number of shares (T) share Number of comments (T) comment Number of likes (T) like ;

[0016] Active Data A jk Including: User login frequency F loginThe length of time a user is active within the most recent unit of time, T. last,active User interaction frequency F interaction .

[0017] Furthermore, step S2 includes:

[0018] Step S2-1: Obtain the raw behavioral data B of the kth user. jk Calculate user engagement level I jk I jk =w1·T click +w2·D dwell +w3·T share +w4·T comment +w5·T like ;

[0019] I jk It reflects users' level of interest, engagement, and willingness to share specific news content;

[0020] Where w1, w2, w3, w4 and w5 are the corresponding parameter weight coefficients, and w1+w2+w3+w4+w5=1;

[0021] Step S2-2: The k-th user reads the content identifier C id Within a unit of time, L news articles are collected from the k-th user. The content identifier of each news article is C. id Unlike other news articles, the content representation of the i-th news article in L news articles is denoted as C. history,i Calculate content preference P jk :

[0022]

[0023] Content Preference jk It quantifies user engagement with current news content on a specific platform. id The degree of interest in similar content and the intensity of their historical interactions;

[0024] Where S(C) id C history,i This represents the content identifier Cid and the identifier C of the user's browsing history. history,i The similarity between them, T click,i and D dwell,i μ1 represents the number of clicks and the duration of time a user spends on the i-th news item out of L news items, respectively. μ1 is the weighting coefficient for the number of clicks and μ2 is the weighting coefficient for the duration of time spent.

[0025] Step S2-3: Obtain the unique characteristics of the j-th media platform and extract active data A.jk Calculate the platform's user activity index H j And the platform's content influence F j :

[0026] H j =α1·F login +α2·T last,active +α3·F interaction F j =β1·E j +β2·V j ;

[0027] Among them, α1, α2 and α3 are the weight coefficients for calculating the user activity index of the platform, and α1+α2+α3=1, β1 and β2 are the weight coefficients for calculating the content influence of the platform.

[0028] Furthermore, step S3 includes:

[0029] Step S3-1: Transfer the k-th user U k On the j-th media platform M j All features on the vector are concatenated into a single original feature vector X. jk X jk =[I jk P jk H j F j ];

[0030] Step S3-2: Process the original feature vector X jk Each feature dimension in the vector is normalized to obtain a normalized feature vector X′. jk .

[0031] The feature analysis step aims to deeply mine and quantify user behavior features in user behavior data and the media platform's own features contained in activity data, transforming raw, discrete data into computable feature vectors.

[0032] Furthermore, step S4 includes:

[0033] Step S4-1: For the k-th user U k On the j-th media platform M j Above, retrieve the content identified as C id The demand vector D for news jk Demand vector D jk The normalized eigenvector X′ jk ;

[0034] Step S4-2: Identify users with identical authentication information across different media platforms as the same user. User authentication information includes: User ID, Device ID, Cookie ID, or IP address. Calculate the inter-platform demand difference. The inter-platform demand difference refers to calculating the user Uk value for the same user. k M on different media platforms j and M p The difference in demand, of which M p Let J represent the p-th media platform in the set of media platforms, where j ≠ p, and calculate the relationship between user Uk and media platform M. j and M p Demand Vector Difference Diff(D) jk D pk ):

[0035]

[0036] Among them, D s D refers to the number of dimensions in the demand vector. jk,d User U k On platform M j The d-th component of the demand vector, D pk,d User U k On platform M p The d-th component of the demand vector;

[0037] Step S4-3: By aggregating the differences in demand for combinations among all users across all platforms, we obtain the j-th media platform M. j User demand direction difference Demand_Diff j :

[0038] Differences in User Demand Direction (Demand_Diff) j The larger the value, the greater the platform M. j The greater the difference between the user demand patterns in this platform and the average demand patterns on other platforms, the more this metric will serve as a key input for building a unified user profile, ensuring that the profile can capture the differences in user demand across platforms.

[0039] Where K represents the total number of user samples.

[0040] The demand difference analysis step is one of the core design points of this invention. Based on the normalized data, it quantitatively analyzes the differences in the demand direction of the same user for news content on different media platforms, thereby solving the problem that cross-platform profiles are difficult to reuse in the existing technology.

[0041] Furthermore, step S5 includes:

[0042] Step S5-1: Concatenate the user demand direction differences and normalized feature vectors to form a unified user feature matrix U. matrix :

[0043]

[0044] Where, x′ jk User U k On platform M j Normalized eigenvectors on the Demand_Diff N This indicates the degree of difference in user demand direction corresponding to the Nth media platform;

[0045] Step S5-2: For the unified user feature matrix U matrix Clustering is performed on the row vectors to divide users into C user groups G = {G1, G2, G3, ..., G...} C}, where G1, G2, G3, ... and G C These represent the 1st, 2nd, 3rd, ..., and Cth user groups, respectively.

[0046] Step S5-3: Among the C user groups, let the e-th user group be denoted as G. e Calculate the average value of each feature dimension to form the initial profile P of the user group. Ge ;

[0047] Step S5-4: When the k-th user U k Belonging to the e-th user group, user group G e Initial portrait P Ge As a foundation, with user U k Normalized feature vector X′ on all platforms jk The average values ​​are added together to generate an individual user profile P. Uk , Where ω represents a weighting coefficient between 0 and 1.

[0048] The unified user profile generation process, through multi-dimensional analysis, based on the results of demand difference analysis and normalized data, yields a comprehensive, dynamic, and reusable unified user profile, thereby effectively solving the problem of user profiles not being reusable in cross-media platform scenarios.

[0049] To better implement the above methods, a new media product user profiling analysis system based on media big data is also proposed.

[0050] The system includes: a data acquisition module, a feature analysis module, a data aggregation module, a demand difference analysis module, and a unified user profile management module;

[0051] The data acquisition module collects user behavior data, user activity data, and platform operation data related to the same news content from several media platforms. The feature analysis module quantifies and extracts data features from the user behavior data, user activity data, and platform operation data to obtain user feature data and media platform feature data, respectively. The user feature data includes user interaction and content preference, while the media platform feature data includes user activity index and platform content influence. The data aggregation module aggregates and normalizes the user feature data and media platform feature data from several media platforms to obtain a demand vector between media platforms and users. The demand difference analysis module normalizes each feature dimension in the original feature vector to obtain a normalized feature vector. The unified user profile management module aggregates the user demand vectors and the difference in user demand directions from all media platforms to obtain a user feature matrix. Vector clustering is performed on the vectors in the user feature matrix to obtain a profile of the group of users. Based on the profile of the group of users, feature value aggregation is performed to obtain an individual user profile.

[0052] Furthermore, the data acquisition module includes: a content identification management unit, a cross-platform management unit, and a data management unit;

[0053] The content identification management unit is used to extract feature words from a news content, encode the feature words or obtain word vectors of the feature words, and record the encoded or word vectors as the identification information of a news content. The identification information of a news content is collected to obtain the content identifier. The cross-platform management unit is used to aggregate several media platforms to form a media platform set. On different media platforms, it collects the interaction data between users and news content with the same content identifier. The interaction data includes raw behavioral data and raw activity data. The data management unit is used to manage the behavioral data and activity data of users on the media platform within a unit time range, as well as collect the exposure and click-through rate of a news content on the media platform within a unit time range.

[0054] Furthermore, the feature analysis module includes: a user behavior feature management unit, a content preference management unit, and a media platform feature management unit;

[0055] Among them, the user behavior feature management unit is used to obtain users' raw behavior data and calculate users' interaction participation; the content preference management unit is used to collect the news content read by users, obtain the content identifiers in the news content, and calculate the content preference; the media platform feature management unit is used to obtain the media platform's own feature extraction index and calculate the platform's user activity index and platform content influence based on the activity data.

[0056] Furthermore, the data aggregation module includes: a feature vector management unit and a normalization processing unit;

[0057] The feature vector management unit is used to concatenate all the user's features on the media platform into an original feature vector, and the normalization processing unit is used to normalize each feature dimension in the original feature vector to obtain a normalized feature vector.

[0058] Furthermore, the unified user profile management module includes: a feature space management unit, a cluster analysis unit, a group profile management unit, and an individual user profile management unit;

[0059] The feature space management unit concatenates the differences in user demand directions and normalized feature vectors to form a unified user feature matrix. The clustering analysis unit clusters the row vectors in the unified user feature matrix to divide users into several user groups. The group profile management unit calculates the average value of each feature dimension to form the initial profile of the user group. The individual user profile management unit adds the initial profile of the user group to the average value of the normalized feature vectors of the user on all platforms to generate an individual user profile.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. By quantitatively analyzing the differences in user needs across different media platforms, the problem of the inability to reuse cross-platform user profiles in traditional methods is solved. By utilizing the calculation of demand difference degree and the fusion of unified feature space, standardized modeling of user behavior across multiple platforms is achieved, making a single user profile adaptable to the needs of multiple scenarios.

[0062] 2. By using a normalized difference aggregation method, the differences in data units between platforms can be eliminated, while data noise can be suppressed, and the model's compatibility with heterogeneous data can be improved.

[0063] 3. Establish user segmentation models that combine group profiles and individual profiles, taking into account both group commonalities and individual differences, to support accurate recommendations and strategy formulation. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the structure of a new media product user profiling analysis system based on media big data according to the present invention;

[0065] Figure 2 This is a flowchart illustrating a new media product user profiling analysis method based on media big data, according to the present invention. Detailed Implementation

[0066] 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.

[0067] Example: Figures 1-2 As shown, this invention provides a technical solution: a user profiling analysis system and method for new media products based on media big data. The method includes:

[0068] Step S1: Collect user behavior data, user activity data, and platform operation data related to the same news content from several media platforms;

[0069] Step S1 includes:

[0070] Step S1-1: Extract feature words from a news article, encode the feature words or obtain word vectors for the feature words, and record the encoded or word vectors as the identification information of the news article. Collect the identification information of the news article to obtain the content identifier C. id ;

[0071] Step S1-2: Gather several media platforms to form a media platform set M, M = {M1, M2, M3, ..., M}. N}, where N represents the total number of media platforms, N is an integer greater than or equal to 2, and M1, M2, M3... and M... N These represent the 1st, 2nd, 3rd, ..., and Nth media platforms, based on the content identifier C. id In different media platforms, interaction data between users and news content with the same content identifier is collected. The interaction data includes raw behavioral data and raw activity data.

[0072] Step S1-3: Obtain the j-th media platform M from the media platform set. j The kth user U k Behavioral data B within a unit time range jk and active data A jk And the amount of exposure E of a certain news content on the j-th media platform within a unit time range. j Click-through rate (V) j ;

[0073] Behavioral data B jk Includes: Number of clicks T click Duration of stay D dwell Number of shares (T) share Number of comments (T)comment Number of likes (T) like ;

[0074] Active Data A jk Including: User login frequency F login The length of time a user is active within the most recent unit of time, T. last,active User interaction frequency F interaction .

[0075] Step S2: Quantify and extract data features from user behavior data, user activity data, and platform operation data to obtain user feature data and media platform feature data, respectively. User feature data includes user interaction and content preference, while media platform feature data includes user activity index and platform content influence.

[0076] Step S2 includes:

[0077] Step S2-1: Obtain the raw behavioral data Bjk of the k-th user and calculate the user's interaction engagement level I. jk I jk =w1·T click +w2·D dwell +w3·T share +w4·T comment +w5·T like ;

[0078] Where w1, w2, w3, w4 and w5 are the corresponding parameter weight coefficients, and w1+w2+w3+w4+w5=1;

[0079] Step S2-2: The k-th user reads the content identifier C id Within a unit of time, L news articles are collected from the k-th user. The content identifier of each news article is C. id Unlike other news articles, the content representation of the i-th news article in L news articles is denoted as C. history,i Calculate content preference P jk :

[0080]

[0081] Where S(C) id C history,i ) represents the content identifier C id The identifier C for the user's browsing history history,i The similarity between them, T click,i and D dwell,iμ1 represents the number of clicks and the duration of time a user spends on the i-th news item out of L news items, respectively. μ1 is the weighting coefficient for the number of clicks and μ2 is the weighting coefficient for the duration of time spent.

[0082] Step S2-3: Obtain the unique characteristics of the j-th media platform and extract active data A. jk Calculate the platform's user activity index H j And the platform's content influence F j :

[0083] H j =α1·F login +α2·T last,active +α3·F interaction F j =β1·E j +β2·V j ;

[0084] Among them, α1, α2 and α3 are the weight coefficients for calculating the user activity index of the platform, and α1+α2+α3=1, β1 and β2 are the weight coefficients for calculating the content influence of the platform.

[0085] Step S3: Collect and normalize user feature data and media platform feature data from several media platforms to obtain the demand vector between media platforms and users;

[0086] Step S3 includes:

[0087] Step S3-1: Transfer the k-th user U k On the j-th media platform M j All features on the vector are concatenated into a single original feature vector X. jk X jk =[I jk P jk H j F j ];

[0088] Step S3-2: Process the original feature vector X jk Each feature dimension in the vector is normalized to obtain a normalized feature vector X′. jk .

[0089] Step S4: Identify different user characteristics, and based on the demand vector of the same user on different media platforms, accumulate the difference in demand vectors between different media platforms to obtain the difference in demand direction of each media platform;

[0090] Step S4 includes:

[0091] Step S4-1: For the k-th user U kOn the j-th media platform M j Above, retrieve the content identified as C id The demand vector D for news jk Demand vector D jk The normalized eigenvector X′ jk ;

[0092] Step S4-2: Identify users with identical authentication information across different media platforms as the same user. User authentication information includes: User ID, Device ID, Cookie ID, or IP address. Calculate the inter-platform demand difference. The inter-platform demand difference refers to the difference in demand between users for the same user. k Calculate user U k M on different media platforms j and M p The difference in demand, of which M p Let J represent the p-th media platform in the set of media platforms, where j ≠ p, and calculate the relationship between user Uk and media platform M. j and M p Demand Vector Difference Diff(D) jk D pk ):

[0093]

[0094] Among them, D s D refers to the number of dimensions in the demand vector. jk,d User U k On platform M j The d-th component of the demand vector, D pk,d User U k On platform M p The d-th component of the demand vector;

[0095] User authentication information includes, but is not limited to: user ID, such as the hash value of the registered email or mobile phone number; device ID, such as IMEI or device fingerprint; cookie ID or IP address.

[0096] In practical applications, cross-platform association of user identities can be achieved by establishing a unified user identity management system or by using multi-dimensional information matching algorithms, such as establishing a probability matching mechanism.

[0097] Step S4-3: By aggregating the differences in demand for combinations among all users across all platforms, we obtain the j-th media platform M. j User demand direction difference Demand_Diff j :

[0098] Where K represents the total number of user samples.

[0099] Step S5: Collect the user demand vectors and user demand direction differences from all media platforms to obtain a user feature matrix. Perform vector clustering on the vectors in the user feature matrix to obtain a profile of the group users. Based on the profile of the group users, aggregate feature values ​​to obtain an individual user profile.

[0100] Step S5 includes:

[0101] Step S5-1: Concatenate the user demand direction differences and normalized feature vectors to form a unified user feature matrix U. matrix :

[0102]

[0103] Where, X′ jk User U k On platform M j Normalized eigenvectors on the Demand_Diff N This indicates the degree of difference in user demand direction corresponding to the Nth media platform;

[0104] Step S5-2: For the unified user feature matrix U matrix Clustering is performed on the row vectors to divide users into C user groups G = {G1, G2, G3, ..., G...} C}, where G1, G2, G3, ... and G C These represent the 1st, 2nd, 3rd, ..., and Cth user groups, respectively.

[0105] In this embodiment, the K-means clustering algorithm can be used. The goal of cluster analysis is to group users with similar cross-platform behavior patterns and needs into the same group, thereby achieving a macro-level profile description of the user group.

[0106] Before performing K-Means clustering, the optimal number of clusters C can be determined using methods such as the elbow rule or the silhouette coefficient, and then K-Means clustering can be performed.

[0107] Step S5-3: Among the C user groups, let the e-th user group be denoted as G. e Calculate the average value of each feature dimension to form the initial profile P of the user group. Ge ;

[0108] Step S5-4: When the k-th user U k Belonging to the e-th user group, user group G e Initial portrait P Ge As a foundation, with user U kNormalized feature vector X′ on all platforms jk The average values ​​are added together to generate an individual user profile P. Uk , Where ω represents a weighting coefficient between 0 and 1.

[0109] The system includes: a data acquisition module, a feature analysis module, a data aggregation module, a demand difference analysis module, and a unified user profile management module;

[0110] Among them, the data collection module is used to collect user behavior data, user activity data and platform operation data related to the same news content from several media platforms;

[0111] The data acquisition module includes a content identification management unit, a cross-platform management unit, and a data management unit. The content identification management unit is used to extract feature words from a news content, encode the feature words or obtain word vectors of the feature words, and record the encoded or word vectors as the identification information of a news content. The identification information of a news content is collected to obtain a content identifier. The cross-platform management unit is used to aggregate several media platforms to form a media platform set. On different media platforms, it collects the interaction data between users and news content with the same content identifier. The interaction data includes raw behavioral data and raw activity data. The data management unit is used to manage the behavioral data and activity data of users on the media platform within a unit time range, as well as collect the exposure and click-through rate of a news content on the media platform within a unit time range.

[0112] The feature analysis module is used to quantify and extract features from user behavior data, user activity data, and platform operation data, resulting in user feature data and media platform feature data. User feature data includes user interaction and content preference, while media platform feature data includes user activity index and platform content influence.

[0113] The feature analysis module includes: a user behavior feature management unit, a content preference management unit, and a media platform feature management unit. The user behavior feature management unit is used to acquire users' raw behavior data and calculate users' interaction participation. The content preference management unit is used to collect the news content read by users, obtain content identifiers in the news content, and calculate content preference. The media platform feature management unit is used to acquire the media platform's own feature extraction index and calculate the platform's user activity index and platform content influence based on activity data.

[0114] Among them, the data aggregation module is used to aggregate and normalize user feature data and media platform feature data from several media platforms to obtain the demand vector between media platforms and users.

[0115] The data aggregation module includes a feature vector management unit and a normalization processing unit. The feature vector management unit is used to concatenate all the features of the user on the media platform into an original feature vector, and the normalization processing unit is used to normalize each feature dimension in the original feature vector to obtain a normalized feature vector.

[0116] Among them, the demand difference analysis module is used to normalize each feature dimension in the original feature vector to obtain a normalized feature vector;

[0117] The demand difference analysis module includes: a user demand vector management unit, a user authentication unit, and a total platform demand difference management unit. The user demand vector management unit is used to analyze the demand vector of users on media platforms for obtaining news. The demand vector is a normalized feature vector. The user authentication unit is used to identify users with the same user authentication information on different media platforms as the same user, calculate the demand difference on different media platforms, and calculate the degree of difference of user demand vectors on different media platforms. The total platform demand difference management unit is used to aggregate the degree of difference of combined demand of all users on all platforms to obtain the degree of difference of user demand direction on media platforms.

[0118] Among them, the unified user profile management module is used to collect the demand vectors and the degree of difference in user demand direction of users in all media platforms to obtain the user feature matrix, perform vector clustering on the vectors in the user feature matrix to obtain the profile of the group users, and perform feature value aggregation based on the profile of the group users to obtain the profile of the individual user.

[0119] The unified user profile management module includes: a feature space management unit, a clustering analysis unit, a group profile management unit, and an individual user profile management unit. The feature space management unit concatenates the differences in user demand directions and normalized feature vectors to form a unified user feature matrix. The clustering analysis unit clusters the row vectors in the unified user feature matrix to divide users into several user groups. The group profile management unit calculates the average value of each feature dimension to form the initial profile of the user group. The individual user profile management unit adds the initial profile of the user group to the average value of the normalized feature vectors of the user across all platforms to generate an individual user profile.

[0120] The implementation process of this application is illustrated below through a specific embodiment:

[0121] S1 Data Acquisition:

[0122] For a news article, word vectors are extracted using the BERT model and aggregated into a vector representation of C_id (0.85, 0.72, 0.61, 0.9).

[0123] News content was distributed across three media platforms, M1, M2, and M3, and raw activity and behavioral data were collected from users U1 and U2 respectively.

[0124] Raw active data is represented as follows:

[0125] For user U1, in M1: Flogin=5, Tlast,active=180min, Finteraction=15; in M2: Flogin=10, Tlast,active=60min, Finteraction=30; in M3: Flogin=3, Tlast,active=240min, Finteraction=10.

[0126] For user U2, in M1: Flogin=8, Tlast,active=120min, Finteraction=20; in M2: Flogin=12, Tlast,active=90min, Finteraction=40; ​​in M3: Flogin=2, Tlast,active=90min, Finteraction=5.

[0127] The raw behavioral data is represented as follows:

[0128] For user U1, in M1: Tclick=5, Ddwell=120s, Tshare=1, Tcomment=2, Tlike=3; in M2: Tclick=1, Ddwell=15s, Tshare=0, Tcomment=0, Tlike=1; in M3: Tclick=3, Ddwell=180s, Tshare=2, Tcomment=4, Tlike=5.

[0129] For user U2, in M1: Tclick=2, Ddwell=60s, Tshare=0, Tcomment=1, Tlike=2; in M2: Tclick=4, Ddwell=45s, Tshare=1, Tcomment=2, Tlike=4; in M3: Tclick=1, Ddwell=30s, Tshare=0, Tcomment=0, Tlike=1.

[0130] Platform exposure and click-through rate are expressed as follows:

[0131] E1=10000, V1=0.08, E2=50000, V2=0.02, E3=5000, V3=0.15.

[0132] S2 Feature Analysis:

[0133] Obtain the weight coefficients ω1 = 0.15, ω2 = 0.3, ω3 = 0.2, ω4 = 0.2, ω5 = 0.15;

[0134] For user U1, in M1, I11 = 0.15 × 5 + 0.3 × 120 + 0.2 × 1 + 0.2 × 2 + 0.15 × 3 = 37.8.

[0135] Similarly, in M2, I21 = 4.8, and in M3, I31 = 56.4;

[0136] Calculate U2 in the same way. In M1, I12 = 18.8, and in M2, I22 = 15.3 and I32 = 9.3.

[0137] Obtain the weight coefficients μ1 = 0.4 and μ2 = 0.6;

[0138] User U1 recently read L=3 news items in M1, namely Ch1, Ch2 and Ch3;

[0139] S(Cid, Ch1) = 0.9, Tclick, 1 = 4, Ddwell, 1 = 90s, S (Cid, Ch2) = 0.7, Tclick, 2 = 2, Ddwell, 2 = 50s, S (Cid, Ch3) = 0.5, Tclick, 3 = 1, Ddwell, 3 = 30s;

[0140] Calculate the preference degree P11 of U1 in M1, P11 = (0.9 × (0.4 × 4 + 0.6 × 90) + 0.7 × (0.4 × 2 + 0.6 × 50) + 0.5 × (0.4 × 1 + 0.6 × 30)) ÷ 3 = 26.93;

[0141] User U1 recently read L=2 news items in M2, Ch4 and Ch5;

[0142] S(Cid, Ch4) = 0.6, Tclick, 4 = 2, Ddwell, 4 = 20s, S (Cid, Ch5) = 0.4, Tclick, 5 = 1, Ddwell, 5 = 10s;

[0143] Calculate the preference degree of U1 in M2, P21 = 5.12;

[0144] Similarly, the preference scores for U1 in M3, U2 in M1, U2 in M2, and U2 in M3 are P31 = 35.5, P12 = 15.8, P22 = 10.2, and P32 = 8.1.

[0145] The weights of H are obtained as α1 = 0.3, α2 = 0.4, and α3 = 0.3, and the weights of F are β1 = 0.00006 and β2 = 0.4. Since the original exposure volume has a large dimension, the data is scaled to a uniform range by adjusting the ratio of β1 and β2.

[0146] Get the average data of active users in M1: Flogin=6, Tlast,active=150min, Finteraction=18;

[0147] Get the average data of active users in M2: Flogin = 11, Tlast,active = 75min, Finteraction = 35;

[0148] Obtain the average data of active users in M3: Flogin = 2.5, Tlast,active = 165min, Finteraction = 7.5;

[0149] Calculate H1 = 43.8, H2 = 43.8, H3 = 69, F1 = 0.632, F2 = 3.008, F3 = 0.36.

[0150] S3 Data Aggregation and Normalization:

[0151] Constructing feature vectors:

[0152] X11=(37.8,26.93,67.2,0.632), X21=(4.8,5.12,43.8,3.008), X31=(56.4,35.5,69,0.36),

[0153] X12=(18.8,15.8,67.2,0.632), X22=(15.3.10.2.43.8.3.008), X32=(9.3,8.1,69,0.36),

[0154] The feature vector is normalized based on the maximum and minimum values ​​of each column of data. For example, the maximum value of the first column is 56.4 and the minimum value is 4.8.

[0155] The normalized vector is used as the user requirement vector:

[0156] D11=(0.6395,0.7178,0.9286,0.1027), D21=(0,0,0,1),

[0157] D31=(1,1,1,0), D12=(0.2713,0.3512,0.9286,0.1027),

[0158] D22=(0.2035,0.1669,0,1), D32=(0.0872,0.0981,1,0);

[0159] Differences in requirements between computing platforms:

[0160] Diff(D11, D21)=1.61, Diff(D11, D31)=0.4746, Diff(D12, D22)=1.3061,

[0161] Diff(D12, D32)=0.3371, Diff(D22, D32)=1.4206;

[0162] Summary of platform requirements differences:

[0163] Demand_Diff1=0.93195, Demand_Diff2=1.584175, Demand_Diff3=1.058075;

[0164] In this embodiment, to reduce computational load, dimensionality reduction of the matrix can be employed, such as using principal component analysis (PCA) to reduce the dimensionality of the vectors.

[0165] The eigenvector corresponding to U1 is (0.6395, 0.7178, 0.9286, 0.1027, 0, 0, 0, 1);

[0166] The eigenvector corresponding to U2 is (0.2713, 0.3512, 0.9286, 0.1027, 0.2035, 0.1669, 0, 1);

[0167] The dimensions are reduced to U1' = (-1.5, 0.8) and U2' = (1.2, -0.6) respectively;

[0168] After calculating the mean of the user demand vectors D11, D21, and D31 related to U1, the dimensionality is reduced to obtain the feature mean vector (-0.8, 0.3).

[0169] After calculating the mean of the user demand vectors D12, D22, and D32 related to U2, the dimensionality is reduced to obtain the feature mean vector (0.9, -0.4);

[0170] Get ω = 0.7,

[0171] The final individual profile of user U1 is PU1 = 0.7 × (-1.5, 0.8) + 0.3 × (-0.8, 0.3) = (-1.29, 0.65);

[0172] The final individual profile of user U2 is PU2 = 0.7 × (1.2, -0.6) + 0.3 × (0.9, -0.4) = (1.11, -0.54).

[0173] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for analyzing user portraits of new media products based on media big data, characterized in that: The method comprises: Step S1: collecting user behavior data, user activity data and platform operation data related to the same news content in a plurality of media platforms; Step S2: quantifying and extracting data features of the user behavior data, the user activity data and the platform operation data to obtain user feature data and media platform feature data, wherein the user feature data includes user interactive participation and content preference, and the media platform feature data includes user activity index and platform content influence; Step S3: collecting and normalizing the user feature data and the media platform feature data in the plurality of media platforms to obtain demand vectors between the media platforms and the users; Step S4: identifying different user features, accumulating the difference degree of demand vectors between different media platforms based on the demand vectors of the same user in different media platforms, and obtaining demand direction difference degrees of each media platform; Step S5: collecting the demand vectors of the users in all media platforms and the user demand direction difference degrees to obtain a user feature matrix, performing vector clustering on the vectors in the user feature matrix to obtain a portrait of a group of users, and performing feature value aggregation based on the portrait of the group of users to obtain an individual user portrait. 2.The method of claim 1, wherein the method further comprises: Step S1 comprises: Step S1-1: feature word extraction is performed on a certain news content, the feature words are encoded or word vectors of the feature words are obtained, the encoding or word vectors are recorded as identification information of the certain news content, and the identification information of the certain news is collected to obtain a content identifier C id ; Step S1-2: Collecting several media platforms to form a media platform set M, M = {M1, M2, M3, …, MN}, where N represents the total number of media platforms, N is an integer greater than or equal to 2, where M1, M2, M3, …, and MN represent the first, second, third, …, and Nth media platforms, respectively, based on the content identifier C N , the content identifier C N is used to identify the news content, and the content identifier C id is used to identify the news content in different media platforms. Collect the interaction data of users with news content with the same content identifier, including original behavior data and original active data. Step S1-3: Obtain the j-th media platform M from the media platform set. j The kth user U k Behavioral data B within a unit time range jk and active data A jk And the exposure volume E of the news content on the j-th media platform within a unit time range. j Click-through rate (V) j ; The behavior data B jk Including: click times T click , stay duration D dwell , sharing times T share , comment quantity T comment , like times T like ; The active data A jk comprising: a user login frequency F login , a user active time length T in a recent unit time range last,active , a user content interaction frequency F interaction . 3.The method of claim 2, wherein the method further comprises: determining a user portrait of the new media product based on the media big data. Step S2 comprises: Step S2-1: Obtain the original behavior data B of the kth user. jk Calculate user engagement level I jk I jk =w1·T click +w2·D dwell +w3·T share +w4·T comment +w5·T like ; Wherein, w1, w2, w3, w4 and w5 are corresponding parameter weight coefficients, and w1+w2+w3+w4+w5=1; Step S2-2: When the kth user reads the content identifier C id In a unit time range of the news content, L pieces of news content read by the kth user are collected, and the content identifier of the news content is C id Different, the content identifier of the ith piece of news content in the L pieces of news content is recorded as C history,i , the content preference degree P jk is calculated wherein S(C id , C history,i ) represents the similarity between the content identifier C id and the identifier C history,i of the historical browsing content of the user, T click,i and D dwell,i respectively represent the number of clicks and the duration of stay of the user on the i-th piece of news content among the L pieces of news content, μ1 is the weight coefficient of the number of clicks, and μ2 is the weight coefficient of the duration of stay. Step S2-3: Obtain the own characteristics of the jth media platform, extract active data A jk , calculate the platform user activity index H j and the platform content influence F j : H j = a1 · F login + a2 · T last,active + a3 · F interaction , F j = b1 · E j + b2 · V j ; Wherein, α1, α2 and α3 are weight coefficients for calculating the platform user activity index, and α1+α2+α3=1, and β1 and β2 are weight coefficients for calculating the platform content influence.

4. The new media product user portrait analysis method based on media big data according to claim 3, characterized in that: Step S3 comprises: Step S3-1: the kth user U k On the jth media platform M j All features are spliced into an original feature vector X jk , X jk = [I jk , P jk , H j , F j ]; Step S3-2: normalizing each feature dimension in the original feature vector X jk to obtain a normalized feature vector X' jk . 5.The method of claim 4, wherein the method further comprises: Step S4 comprises: Step S4-1 : for the kth user U k At the jth media platform M j , the demand vector D id for news with content identifier C jk is obtained, said demand vector D jk being the normalized feature vector X' jk ; Step S4-2: Identify users with identical user authentication information across different media platforms as the same user. The user authentication information includes: User ID, Device ID, Cookie ID, or IP address. Calculate the inter-platform demand difference, where the inter-platform demand difference refers to calculating the difference between the user's ID and device ID for the same user (Uk). k M on different media platforms j and M p The difference in demand, of which M p Let j represent the p-th media platform in the media platform set, where j ≠ p, and calculate the value for user U. k For media platform M j and M p Demand Vector Difference Diff(D) jk D pk ): where D s denotes the dimensionality of the demand vector, D jk,d denotes the user U k on the platform M j the d-th component of the demand vector, D pk,d denotes the user U k on the platform M p the d-th component of the demand vector; Step S4-3: Obtain the demand direction difference Demand_Diff of the jth media platform M by aggregating the demand direction difference of all users on all platforms for the combination j j :​ where K represents the total number of user samples. 6.The method of claim 5, wherein the method further comprises: Step S5 comprises: Step S5-1: splice the user demand direction difference degree and the normalized feature vector to form a unified user feature matrix U matrix : wherein X' jk refers to the user U k on the platform M j normalized feature vector, Demand_Diff N represents the user demand direction difference degree corresponding to the Nth media platform; Step S5-2: clustering the row vectors in the unified user feature matrix U matrix , and dividing the users into C user groups G = {G1, G2, G3,..., GC}, where G1, G2, G3,..., and GCrepresent the 1st, 2nd, 3rd,..., and Cth user groups, respectively. C C ​​ Step S5-3: Among the C user groups, the e-th user group is denoted as G e , and the average value of each feature dimension is calculated to form the initial image P of the user group Ge ; Step S5-4: When the k-th user U k Belonging to the e-th user group, user group G e Initial portrait P Ge As a foundation, with user U k Normalized feature vector X′ on all platforms jk The average values ​​are added together to generate an individual user profile P. Uk , Where ω represents a weighting coefficient between 0 and 1. 7.A media big data-based new media product user portrait analysis system for performing the media big data-based new media product user portrait analysis method of any one of claims 1-6. The system comprises: A data collection module, a feature analysis module, a data collection module, a demand difference analysis module and a unified user portrait management module; The data collection module is used for collecting user behavior data, user activity data and platform operation data related to the same news content in a plurality of media platforms, the feature analysis module is used for quantifying and extracting data features of the user behavior data, the user activity data and the platform operation data to obtain user feature data and media platform feature data, wherein the user feature data includes user interactive participation and content preference, and the media platform feature data includes user activity index and platform content influence, the data collection module is used for collecting and normalizing the user feature data and the media platform feature data in the plurality of media platforms to obtain demand vectors between the media platforms and the users, the demand difference analysis module is used for normalizing each feature dimension in the original feature vector to obtain a normalized feature vector, and the unified user portrait management module is used for collecting the demand vectors of the users in all media platforms and the user demand direction difference degrees to obtain a user feature matrix, performing vector clustering on the vectors in the user feature matrix to obtain a portrait of a group of users, and performing feature value aggregation based on the portrait of the group of users to obtain an individual user portrait.

8. The new media product user portrait analysis system based on media big data according to claim 7, characterized in that: The data collection module comprises a content identifier management unit, a cross-platform management unit and a data management unit, wherein the content identifier management unit is configured to extract feature words from a certain news content, encode the feature words or obtain word vectors of the feature words, record the encoding or the word vectors as identifier information of the certain news content, and collect the identifier information of the certain news to obtain a content identifier; the cross-platform management unit is configured to collect a plurality of media platforms to form a media platform set, collect interaction data of a user with the news content having the same content identifier in different media platforms, and the interaction data comprises original behavior data and original active data; and the data management unit is configured to manage behavior data and active data of the user on the media platform in a unit time range, and collect exposure and click conversion rate of the media platform to the certain news content in the unit time range. The feature analysis module comprises a user behavior feature management unit, a content preference degree management unit and a media platform feature management unit, wherein the user behavior feature management unit is configured to obtain the original behavior data of the user, and calculate an interaction participation degree of the user; the content preference degree management unit is configured to collect news content read by the user, obtain a content identifier in the news content, and calculate a content preference degree; and the media platform feature management unit is configured to obtain media platform self-feature extraction, and calculate a platform user active index and a platform content influence based on the active data.

9. The new media product user portrait analysis system based on media big data according to claim 7, characterized in that: The data collection module comprises a content identifier management unit, a cross-platform management unit and a data management unit, wherein the content identifier management unit is configured to extract feature words from a certain news content, encode the feature words or obtain word vectors of the feature words, record the encoding or the word vectors as identifier information of the certain news content, and collect the identifier information of the certain news to obtain a content identifier; the cross-platform management unit is configured to collect a plurality of media platforms to form a media platform set, collect interaction data of a user with the news content having the same content identifier in different media platforms, and the interaction data comprises original behavior data and original active data; and the data management unit is configured to manage behavior data and active data of the user on the media platform in a unit time range, and collect exposure and click conversion rate of the media platform to the certain news content in the unit time range.

10. The new media product user portrait analysis system based on media big data according to claim 7, characterized in that: The demand difference analysis module comprises a user demand vector management unit, a user authentication unit and a total platform demand difference management unit, wherein the user demand vector management unit is configured to obtain a demand vector of the user on the media platform, and the demand vector is the normalized feature vector; the user authentication unit is configured to identify users with the same user authentication information in different media platforms as the same user, calculate demand differences in different media platforms, and calculate a demand vector difference degree of the user on different media platforms; and the total platform demand difference management unit is configured to aggregate the demand vector difference degrees of all users on all platforms to obtain a user demand direction difference degree of the media platform. The unified user portrait management module comprises a feature space management unit, a clustering analysis unit, a group portrait management unit and an individual user portrait management unit, wherein the feature space management unit splices the user demand direction difference and the normalized feature vector to form a unified user feature matrix, the clustering analysis unit is used for clustering the row vectors in the unified user feature matrix to divide users into several user groups, the group portrait management unit is used for calculating the average value of each feature dimension to form an initial portrait of the user group, and the individual user portrait management unit is used for adding the average value of the normalized feature vector of the user on all platforms to the initial portrait of the user group as a basis to generate an individual user portrait.

Images