User portrait intelligent analysis system and method based on data visualization

Abstract

The invention discloses a user portrait intelligent analysis system and method based on data visualization. The method comprises the following steps: S1, constructing a user multi-source data set; s2, extracting association information among users based on the user multi-source data set, and constructing a user relationship network by using the user association information; s3, dynamically adjusting a community division result based on the similarity of the nodes and the distribution of edge weights; s4, performing feature extraction on the user nodes in each user community, and storing the user features to a tag management module in a classification tag form; s5, based on the user relation network and the community detection result, generating a visual chart through a data visualization module, and presenting the distribution and key features of the user community in a graphical form; and S6, based on the user portrait label and the community grouping result, using a machine learning algorithm to further analyze the user behavior data, and predicting future interest preference or behavior trend of the user. According to the invention, a more efficient tool is provided for enterprises in the aspects of user grouping and precision marketing.

Description

Technical Field

[0001] The present invention relates to the field of user portrait technology, and in particular to a user portrait intelligent analysis system and method based on data visualization. Background Art

[0002] With the rapid development of big data and artificial intelligence technologies, user portraits, as a core data analysis technology, are widely used in precision marketing, recommendation systems, and personalized services. User portraits mine users' multi-dimensional data to characterize their interests, behaviors, and preferences, providing data-driven decision support for enterprises and organizations.

[0003] In the existing technology, user portraits usually rely on a single data source. The analysis method of a single data source limits the comprehensiveness and diversity of user portraits, making it difficult to accurately describe the characteristics of users in different scenarios. In addition, traditional user portrait methods are often based on static data and lack dynamic update capabilities, making it difficult to adapt to rapid changes in user behavior and preferences.

[0004] At the same time, existing user portrait analysis methods lack intuitiveness and operability in result display. Usually, the analysis results are presented in the form of text reports or tables. Analysts need to further extract valuable information. The display method not only reduces the efficiency of data utilization but also increases the professional requirements for analysts. The existing system lacks deep mining of user relationship networks during the analysis process, making it difficult to identify potential correlations between users and unable to effectively realize user segmentation and identification of interest groups.

[0005] In recent years, some user profiling methods based on network analysis and machine learning have gradually emerged, but they still face the following major problems: First, network analysis lacks the ability to comprehensively integrate multi-source data, making it difficult to take into account the multidimensionality and consistency of data; second, when processing user relationship networks, the existing technologies lack the accuracy of community division and feature extraction, resulting in low relevance and reliability of user grouping results.

[0006] In summary, existing technologies have obvious deficiencies in the comprehensiveness, dynamism, visual analysis and user community feature mining of user portrait construction. The existing technical defects not only limit the actual application effect of user portraits, but also bring significant challenges to corporate precision marketing and user behavior analysis. Summary of the invention

[0007] One object of the present invention is to propose a user portrait intelligent analysis system and method based on data visualization, which provides enterprises with more efficient tools in user grouping and precision marketing.

[0008] A user portrait intelligent analysis method based on data visualization according to an embodiment of the present invention includes the following steps:

[0009] S1. Build a user multi-source data set, clean the collected user multi-source data set, and remove noise data, incomplete data, and duplicate data;

[0010] S2. Extract the association information between users based on the user multi-source data set, and use the user association information to build a user relationship network;

[0011] S3. Use the Louvain algorithm to cluster nodes in the user relationship network, generate multiple user communities, and dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights;

[0012] S4. Extract features from user nodes in each user community and store the user features in the form of classification labels in the label management module;

[0013] S5. Based on the user relationship network and community detection results, a visualization chart is generated through the data visualization module to present the distribution and key features of the user community in a graphical form. Different colors and area markers are used to distinguish user communities. The node size and color encode the user feature weight, and the thickness and color of the edge reflect the relationship strength inside and outside the community;

[0014] S6. Based on user portrait tags and community grouping results, use machine learning algorithms to further analyze user behavior data and predict users’ future interest preferences or behavior trends.

[0015] Optionally, the S1 is specifically as follows:

[0016] S11. Obtain user social media activity data, transaction record data, geographic location data, and user interaction behavior data from different data sources through the data collection module to construct an initial user multi-source data set:

[0017] D={D s ,D t ,D l ,D i};

[0018] Among them, D s Represents the user's activity records on social media platforms, including likes, comments, and sharing behavior data. t Represents the user's transaction record, including transaction time, amount and commodity category, D l Indicates the user's geographic location information, including GPS coordinates and timestamp, D i Represents the user's interactive behavior data, including message sending frequency, interactive objects, and time intervals;

[0019] S12, performing data cleaning on the initial user multi-source data set D, marking and removing abnormal data in the initial user multi-source data set D by using a noise detection algorithm, processing incomplete data by using an interpolation method or a deletion method, and determining the uniqueness of data records in the initial user multi-source data set D by using a hash algorithm, removing records with the same hash value, and obtaining a cleaned user multi-source data set D';

[0020] S13, standardize the cleaned user multi-source data set D', unify the format and range of the data, and map the feature values ​​of different data sources to the same range [0,1];

[0021] S14, integrating the cleaned and standardized user multi-source data sets into a unified user multi-source data set D u :

[0022]

[0023] Among them, u k represents the kth user, D′ s (u k ),D′ t (u k ),D′ l (u k ),D′ i (u k ) represent the cleaned and standardized k Related social media activity data, transaction record data, geographic location data and interactive behavior data, where n represents the number of users.

[0024] Optionally, S2 is specifically as follows:

[0025] S21. Based on user multi-source dataset D u Extract association information between users, including social relationships, behavioral similarities, and common interest tags;

[0026] S22. For user multi-source dataset D u For every pair of users u i and user u j , calculate the social relationship similarity R s (u i ,u j ), the social relationship similarity is measured by the user social media activity data D′ s (u i ) and D′ s (u j )calculate:

[0027]

[0028] Among them, A(u i ) represents user u i The collection of social media interactions, |A(u i )∩A(u j )| represents user u i and user u j Common social media interactions, |A(u i )∪A(u j )| represents user u i and user u j Total number of social media interactions;

[0029] S23, using cosine similarity based on transaction record data D' t (u i ), D′ t (u j ) and interactive behavior data D′ i (u i ), D′ i (u j ) Calculate user u i and user u j The behavioral similarity between b (u i ,u j ):

[0030]

[0031] Among them, V(u i )=[D′ t (u i ),D′ i (u i )] represents user u i The behavioral characteristic vector, V(u j )=[D′ t (u j ),D′ i (u j )] represents user u j The behavioral feature vector of , · represents the vector dot product, ||V(u i )|| represents the vector V(u i )’s module length;

[0032] S24. Identify user u i and user u j The set of common interest tags between c (u i ,u j ):

[0033] T c (ui ,u j )=T(u i )∩T(u j );

[0034] Among them, T(u i ) indicates that from user u i The set of interest tags extracted from the data, ∩ represents the intersection operation of the set, T c (u i ,u j ) indicates the number of common interests;

[0035] S25. Calculate user u i and user u j The association strength weight w ij :

[0036]

[0037] Among them, α, β, γ are weight coefficients |T(u i )∪T(u j )| represents user u i and user u j The number of elements in the union of interest tags, Indicates the similarity of common interest tags;

[0038] S26. Build a user relationship network:

[0039] G = (V, E);

[0040] Where V = {u1,u2,…,u n} is the user node set, n represents the number of users, E={e ij} is the weighted edge set between users, satisfying the association strength weight w ij ≥δ, where δ is the preset association strength threshold;

[0041] S27. In the user relationship network G, users are represented by nodes, and the relationships between users are represented by weighted edges e. ij The edge weight w is expressed in the form ij Reflect user u i and user u j The strength of the association between them.

[0042] Optionally, the S3 is as follows:

[0043] S31, using the Louvain algorithm to divide the node set V and edge set E in the user relationship network G into communities with the maximization of modularity Q as the objective function;

[0044] S32, initialization, each user u i Considered an independent community i ;

[0045] S33. For each node u in the network i , traverse its neighbor user u j Community C j , calculate the user u i Move to Community C j The modularity gain ΔQ after:

[0046]

[0047] Where m is the total edge weight in the network, w ij For user u i and user u j The edge weight between i For user u i The degree of user u i The sum of the weights of the connected edges, For user u i With Community C j The sum of edge weights between internal nodes, For Community C j The total edge weight inside, For Community C j The sum of the degrees of all nodes in

[0048] S34, user u i Move to community C where the modularity gain ΔQ is maximum and greater than zero j If there is no gain, the original community remains unchanged and step S33 is repeated until the movement of any node in the network cannot increase the modularity Q;

[0049] S35, construct a new aggregation network, regard each community C as a new node, and the new edge weight is the sum of the edge weights between the original communities, forming a new node set V' and edge set E', repeating steps S31 to S34, and dividing the new aggregation network into communities until the modularity Q converges, and obtaining the final community division result;

[0050] S36. Dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights, so that the generated user community has high internal correlation and low external correlation, and meets the following modularity maximization conditions:

[0051]

[0052] Among them, δ(c i ,cj ) is the Kronecker delta function. When user u i and user u j When they belong to the same community, δ(c i ,c j )=1, otherwise δ(c i ,c j )=0.

[0053] Optionally, the S4 is specifically as follows:

[0054] S41, based on the user interest preference data from user u i The interest tag set T(u i ) Extract interest preference features F p (u i ), defined as the weight distribution of users on each interest tag;

[0055] S43, based on the user behavior pattern data from user u i Transaction record data D′ t (u i ) and interactive behavior data D′ i (u i ) Extract behavioral pattern features F b (u i ), defined as the frequency and temporal distribution of user behaviors;

[0056] S44, based on the user geographical distribution data from user u i Geographic location information D′ l (u i ) Extract geographical distribution features F g (u i );

[0057] S45, extracting the interest preference feature F p (u i ), Behavioral pattern characteristics F b (u i ) and geographical distribution characteristics F g (u i ) is integrated into the user feature set F(u i ).

[0058] Optionally, the S5 is specifically as follows:

[0059] S51. Construct a user community visualization data structure based on the user relationship network G = (V, E) and the community detection results, including the following contents:

[0060] User node u i ∈V user feature set F(u i );

[0061] The edge set E of the user relationship network = {e ij}, each edge e ij The weight w ij Reflects the strength of association between users;

[0062] Community division results {C1, C2, ..., C m}, where each community {C1,C2,…,C m}Contains a subset of user nodes

[0063] S52. Based on the user community visualization data structure, a visualization chart of the user community is generated through a data visualization module and displayed according to the following rules:

[0064] Node: A circle represents a user node u i , the size of the node S(u i ) is proportional to the sum of the feature weights:

[0065] S(u i )=α p ·W p (u i )+α b ·W b (u i )+α g ·W g (u i );

[0066] Among them, W p (u i )、W b (u i )、W g (u i ) represent the total weights of interest preference, behavior pattern and geographical distribution characteristics, α p ,α b ,α g is the display scale factor of feature weight;

[0067] Edge: A straight line represents the relationship between user nodes. ij , the thickness of the edge T(e ij ) and the correlation strength w ij Proportional, colors distinguish the relationship between inside and outside the community;

[0068] Color: Community C k All nodes and edges of are color-coded uniformly, and the brightness of the color is proportional to the mean of the feature weights:

[0069]

[0070] Among them, L(C k ) represents community C k Color brightness;

[0071] S53. Based on the geographical distribution characteristics of the user community F g (u i ), display community C through heat map or geographical distribution map k Geographical features:

[0072] Map node location: determined by the user's geographic feature set F g (u i ) in loc k Decide;

[0073] Popularity: by access frequency freq k Or weight w k Sure;

[0074] S54. Support users' dynamic operations on visual charts through interactive interfaces, including zooming in, zooming out, filtering and focusing functions;

[0075] S55. Generate a dynamic report based on the visualization results of the user community, showing the following:

[0076] Each community C k Key features of the data, including interest tag distribution, behavior pattern distribution, and geographic distribution;

[0077] User node u i Detailed feature data of the , including its association strength and coreness measure in the community.

[0078] A user portrait intelligent analysis system based on data visualization, including the following modules:

[0079] The data collection module is used to collect user data from multiple data sources such as user social media activities, transaction records, geographic location information, and user interaction behaviors, and update the initial user multi-source data set in real time;

[0080] The data cleaning and standardization module is used to clean the initial user multi-source data, remove noise data, incomplete data and duplicate data, and standardize the data;

[0081] The user relationship network construction module is used to extract the association information between users from the cleaned and standardized multi-source data to generate a user relationship network;

[0082] The community detection module is used to cluster nodes in the user relationship network, dynamically adjust the community division results using the Louvain algorithm, and output the division results as node sets and corresponding feature data of multiple user communities;

[0083] The user feature extraction and labeling module is used to extract features of user nodes in each user community, including interest preferences, behavior patterns, and geographic distribution features, generate a user feature set, and store the features in the form of classified labels in the label management module;

[0084] The data visualization module is used to generate visualization charts of user communities based on user relationship networks and community detection results. The data visualization module supports the following functions:

[0085] User node size and color are dynamically encoded according to user feature weights;

[0086] User relationships show the strength of association by edge thickness and color;

[0087] User communities are distinguished by color and regional markers, and the brightness reflects the overall feature weight level of the community;

[0088] Support heat map and geographic distribution map to display user geographic characteristics;

[0089] Provides an interactive interface that supports zooming in, zooming out, filtering, and focusing operations;

[0090] Intelligent analysis and dynamic optimization module, which is used to dynamically analyze and predict user behavior patterns, interest preferences and geographical distribution based on user portrait tags and community grouping results, and respond to real-time changes in data, automatically adjust user relationship networks and community grouping results, and update user portraits;

[0091] The report generation module is used to generate dynamic reports based on the user community analysis results of data visualization, intuitively displaying the key features of each community and the information of user nodes.

[0092] The beneficial effects of the present invention are:

[0093] (1) The present invention comprehensively constructs a multi-source user dataset by integrating users’ social media activity data, transaction record data, geographic location information, and interactive behavior data. Compared with traditional user profiling technology that relies on a single data source, the present invention significantly improves the accuracy and completeness of user profiling by aggregating and extracting features from multidimensional data. It combines interest preference tags, behavioral similarity, and social interaction intensity information in the calculation of user association relationships, which can more accurately reflect the strength of associations between users and lay a high-quality foundation for the construction of user relationship networks.

[0094] (2) The present invention uses the Louvain algorithm to perform dynamic community detection on nodes in the user relationship network, and realizes the accurate division of user communities through modularity gain optimization and layer-by-layer progressive division of the aggregation network. Compared with traditional methods, the present invention can dynamically adjust the internal and external correlations of the community to make the community division results more consistent and reliable. In the community feature extraction, the high similarity and strong correlation of users within the community are ensured by comprehensively considering interest preferences, behavior patterns and geographical distribution, providing enterprises with more efficient tools in user grouping and precision marketing.

[0095] (3) The present invention designs a data visualization module based on user relationship networks and community detection results, which intuitively displays the complex user community distribution and its key features through node size, color, edge thickness and area marking. Compared with the traditional way of presenting user portraits in tables or static charts, the present invention achieves a clear expression of user community characteristics through dynamic visualization charts, and supports interactive operations, which significantly improves the convenience and operability of user portrait analysis. The spatial characteristics of the community can be quickly identified through heat maps and geographic distribution maps, providing strong support for enterprises to optimize resource allocation and geography-related precision delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0097] Figure 1 This is a flow chart of a user portrait intelligent analysis system and method based on data visualization proposed by the present invention. DETAILED DESCRIPTION

[0098] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0099] refer to Figure 1 , a user portrait intelligent analysis method based on data visualization, comprising the following steps:

[0100] S1. Build a user multi-source data set, clean the collected user multi-source data set, and remove noise data, incomplete data, and duplicate data;

[0101] S2. Extract the association information between users based on the user multi-source data set, and use the user association information to build a user relationship network;

[0102] S3. Use the Louvain algorithm to cluster nodes in the user relationship network, generate multiple user communities, and dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights;

[0103] S4. Extract features from user nodes in each user community and store the user features in the form of classification labels in the label management module;

[0104] S5. Based on the user relationship network and community detection results, a visualization chart is generated through the data visualization module to present the distribution and key features of the user community in a graphical form. Different colors and area markers are used to distinguish user communities. The node size and color encode the user feature weight, and the thickness and color of the edge reflect the relationship strength inside and outside the community;

[0105] S6. Based on user portrait tags and community grouping results, use machine learning algorithms to further analyze user behavior data and predict users’ future interest preferences or behavior trends.

[0106] In this implementation, S1 is specifically as follows:

[0107] S11. Obtain user social media activity data, transaction record data, geographic location data, and user interaction behavior data from different data sources through the data collection module to construct an initial user multi-source data set:

[0108] D={D s ,D t ,D l ,D i};

[0109] Among them, D s Represents the user's activity records on social media platforms, including likes, comments, and sharing behavior data. t Represents the user's transaction record, including transaction time, amount and commodity category, D l Indicates the user's geographic location information, including GPS coordinates and timestamp, D i Represents the user's interactive behavior data, including message sending frequency, interactive objects, and time intervals;

[0110] S12, perform data cleaning on the initial user multi-source data set D, mark and remove abnormal data in the initial user multi-source data set D through a noise detection algorithm, process incomplete data through interpolation or deletion, perform uniqueness judgment on data records in the initial user multi-source data set D through a hash algorithm, remove records with the same hash value, and obtain a cleaned user multi-source data set D';

[0111] S13, standardize the cleaned user multi-source data set D', unify the format and range of the data, and map the feature values ​​of different data sources to the same range [0,1];

[0112] S14, integrating the cleaned and standardized user multi-source data sets into a unified user multi-source data set D u :

[0113]

[0114] Among them, u k represents the kth user, D′ s (u k ),D′ t (u k ),D′ l (u k ),D′ i (u k ) represent the cleaned and standardized k Related social media activity data, transaction record data, geographic location data and interactive behavior data, where n represents the number of users.

[0115] In this implementation, S2 is specifically as follows:

[0116] S21. Based on user multi-source dataset D u Extract association information between users, including social relationships, behavioral similarities, and common interest tags;

[0117] S22. For user multi-source dataset D u For every pair of users u i and user u j , calculate the social relationship similarity R s (u i ,u j ), social relationship similarity is measured through user social media activity data D′ s (u i ) and D′ s (u j )calculate:

[0118]

[0119] Among them, A(u i ) represents user u i The collection of social media interactions, |A(u i )∩A(u j )| represents user u i and user u j Common social media interactions, |A(u i )∪A(uj )| represents user u i and user u j Total number of social media interactions;

[0120] S23, using cosine similarity based on transaction record data D' t (u i ), D′ t (u j ) and interactive behavior data D′ i (u i ), D′ i (u j ) Calculate user u i and user u j The behavioral similarity between b (u i ,u j ):

[0121]

[0122] Among them, V(u i )=[D′ t (u i ),D′ i (u i )] represents user u i The behavioral characteristic vector, V(u j )=[D′ t (u j ),D′ i (u j )] represents user u j The behavioral feature vector of , · represents the vector dot product, ||V(u i )|| represents the vector V(u i )’s module length;

[0123] S24. Identify user u i and user u j The set of common interest tags between c (u i ,u j ):

[0124] T c (u i ,u j )=T(u i )∩T(u j );

[0125] Among them, T(u i ) indicates that from user u i The set of interest tags extracted from the data, ∩ represents the intersection operation of the set, T c(u i ,u j ) indicates the number of common interests;

[0126] S25. Calculate user u i and user u j The association strength weight w ij :

[0127]

[0128] Among them, α, β, γ are weight coefficients |T(u i )∪T(u j )| represents user u i and user u j The number of elements in the union of interest tags, Indicates the similarity of common interest tags;

[0129] S26. Build a user relationship network:

[0130] G = (V, E);

[0131] Where V = {u1,u2,…,u n} is the user node set, n represents the number of users, E={e ij} is the weighted edge set between users, satisfying the association strength weight w ij ≥δ, where δ is the preset association strength threshold;

[0132] S27. In the user relationship network G, users are represented by nodes, and the relationships between users are represented by weighted edges e. ij The edge weight w is expressed in the form ij Reflect user u i and user u j The strength of the association between them.

[0133] In this implementation, S3 is specifically as follows:

[0134] S31, using the Louvain algorithm to divide the node set V and edge set E in the user relationship network G into communities with the maximization of modularity Q as the objective function;

[0135] S32, initialization, each user u i Considered an independent community i ;

[0136] S33. For each node u in the network i , traverse its neighbor user u j Community C j , calculate the user u i Move to Community Cj The modularity gain ΔQ after:

[0137]

[0138] Where m is the total edge weight in the network, w ij For user u i and user u j The edge weight between i For user u i The degree of user u i The sum of the weights of the connected edges, For user u i With Community C j The sum of edge weights between internal nodes, For Community C j The total edge weight inside, For Community C j The sum of the degrees of all nodes in

[0139] S34, user u i Move to community C where the modularity gain ΔQ is maximum and greater than zero j If there is no gain, the original community remains unchanged and step S33 is repeated until the movement of any node in the network cannot increase the modularity Q;

[0140] S35, construct a new aggregation network, regard each community C as a new node, and the new edge weight is the sum of the edge weights between the original communities, forming a new node set V' and edge set E', repeating steps S31 to S34, and dividing the new aggregation network into communities until the modularity Q converges, and obtaining the final community division result;

[0141] S36. Dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights, so that the generated user community has high internal correlation and low external correlation, and meets the following modularity maximization conditions:

[0142]

[0143] Among them, δ(c i ,c j ) is the Kronecker delta function. When user u i and user u j When they belong to the same community, δ(c i ,c j )=1, otherwise δ(c i ,c j )=0.

[0144] In this implementation, S4 is specifically as follows:

[0145] S41, based on the user interest preference data from user u i The interest tag set T(u i ) Extract interest preference features F p (u i ), defined as the weight distribution of users on each interest tag;

[0146] S43, based on the user behavior pattern data from user u i Transaction record data D′ t (u i ) and interactive behavior data D′ i (u i ) Extract behavioral pattern features F b (u i ), defined as the frequency and temporal distribution of user behaviors;

[0147] S44, based on the user geographical distribution data from user u i Geographic location information D′ l (u i ) Extract geographical distribution features F g (u i );

[0148] S45, extracting the interest preference feature F p (u i ), Behavioral pattern characteristics F b (u i ) and geographical distribution characteristics F g (u i ) is integrated into the user feature set F(u i ).

[0149] In this implementation, S5 is specifically as follows:

[0150] S51. Construct a user community visualization data structure based on the user relationship network G = (V, E) and the community detection results, including the following contents:

[0151] User node u i ∈V user feature set F(u i );

[0152] The edge set E of the user relationship network = {e ij}, each edge e ij The weight w ij Reflects the strength of association between users;

[0153] Community division results {C1, C2, ..., C m}, where each community {C1,C2,…,C m}Contains a subset of user nodes

[0154] S52. Based on the user community visualization data structure, a visualization chart of the user community is generated through a data visualization module and displayed according to the following rules:

[0155] Node: A circle represents a user node u i , the size of the node S(u i ) is proportional to the sum of the feature weights:

[0156] S(u i )=α p ·W p (u i )+α b ·W b (u i )+α g ·W g (u i );

[0157] Among them, W p (u i )、W b (u i )、W g (u i ) represent the total weights of interest preference, behavior pattern and geographical distribution characteristics, α p ,α b ,α g is the display scale factor of feature weight;

[0158] Edge: A straight line represents the relationship between user nodes. ij , the thickness of the edge T(e ij ) and the correlation strength w ij Proportional, colors distinguish the relationship between inside and outside the community;

[0159] Color: Community C k All nodes and edges of are color-coded uniformly, and the brightness of the color is proportional to the mean of the feature weights:

[0160]

[0161] Among them, L(C k ) represents community C k Color brightness;

[0162] S53. Based on the geographical distribution characteristics of the user community F g (u i ), display community C through heat map or geographical distribution mapk Geographical features:

[0163] Map node location: determined by the user's geographic feature set F g (u i ) in loc k Decide;

[0164] Popularity: by access frequency freq k Or weight w k Sure;

[0165] S54. Support users' dynamic operations on visual charts through interactive interfaces, including zooming in, zooming out, filtering and focusing functions;

[0166] S55. Generate a dynamic report based on the visualization results of the user community, showing the following:

[0167] Each community C k Key features of the data, including interest tag distribution, behavior pattern distribution, and geographic distribution;

[0168] User node u i Detailed feature data of the , including its association strength and coreness measure in the community.

[0169] A user portrait intelligent analysis system based on data visualization, including the following modules:

[0170] The data collection module is used to collect user data from multiple data sources such as user social media activities, transaction records, geographic location information, and user interaction behaviors, and update the initial user multi-source data set in real time;

[0171] The data cleaning and standardization module is used to clean the initial user multi-source data, remove noise data, incomplete data and duplicate data, and standardize the data;

[0172] The user relationship network construction module is used to extract the association information between users from the cleaned and standardized multi-source data to generate a user relationship network;

[0173] The community detection module is used to cluster nodes in the user relationship network, dynamically adjust the community division results using the Louvain algorithm, and output the division results as node sets and corresponding feature data of multiple user communities;

[0174] The user feature extraction and labeling module is used to extract features of user nodes in each user community, including interest preferences, behavior patterns, and geographic distribution features, generate a user feature set, and store the features in the form of classified labels in the label management module;

[0175] The data visualization module is used to generate visualization charts of user communities based on user relationship networks and community detection results. The data visualization module supports the following functions:

[0176] User node size and color are dynamically encoded according to user feature weights;

[0177] User relationships show the strength of association by edge thickness and color;

[0178] User communities are distinguished by color and regional markers, and the brightness reflects the overall feature weight level of the community;

[0179] Support heat map and geographic distribution map to display user geographic characteristics;

[0180] Provides an interactive interface that supports zooming in, zooming out, filtering, and focusing operations;

[0181] Intelligent analysis and dynamic optimization module, which is used to dynamically analyze and predict user behavior patterns, interest preferences and geographical distribution based on user portrait tags and community grouping results, and respond to real-time changes in data, automatically adjust user relationship networks and community grouping results, and update user portraits;

[0182] The report generation module is used to generate dynamic reports based on the user community analysis results of data visualization, intuitively displaying the key features of each community and the information of user nodes.

[0183] Embodiment 1:

[0184] To verify the feasibility and effectiveness of the present invention, this embodiment 1 takes the user portrait analysis of a certain e-commerce platform as an application scenario, integrates, analyzes and visualizes its multi-source data to achieve accurate user segmentation and personalized marketing decisions.

[0185] The e-commerce platform has an average of about 500,000 active users per day, and users are mainly distributed in many regions across the country. The platform hopes to use user portrait intelligent analysis technology to explore users' shopping behavior characteristics, interest preferences and geographical distribution, discover potential high-value user groups and formulate more precise marketing strategies.

[0186] In actual application, the method of the present invention starts with the following data: user's social media interaction data (likes, comments, reposts), platform transaction records (purchase frequency, order amount, product category), geographic location information (delivery address, active areas) and user interaction behavior (product browsing records, customer service consultation records). 5 million pieces of data were collected from January 2023 to June 2023, of which each user's social media records accounted for approximately 10%, transaction records accounted for 50%, and geographic data and behavior records accounted for 20% respectively.

[0187] First, the collected multi-source data is cleaned and standardized through the data acquisition module of the present invention. The shopping record of user A in February 2023 shows that the order amount of his home appliances is 1,500 yuan, and the order amount of clothing is 800 yuan. The geographical location of user A shows that his main active area is Area A, and he has mentioned home appliance promotions many times in the comment area. After cleaning duplicate data and formatting, user A's multi-source data is stored as standardized records, and the data feature values ​​are uniformly normalized to the range of [0,1].

[0188] Next, we built a user relationship network based on multi-source data. The similarity between user A and user B in purchasing home appliances was 0.85, and both were active in area A. We divided the user relationship network into communities using the Louvain algorithm, and found that user A and user B belonged to the same community, while user C (mainly purchased clothing products and was active in area B) belonged to another community. During the community division process, we found that the internal correlation of the community reached 0.92 and the external correlation was less than 0.15 through modularity optimization. Finally, we divided users into 350 communities, with an average of 142 users in each community.

[0189] In the process of extracting user features, the present invention extracts the user's consumption behavior features from the transaction records. The average amount per order of user A is 450 yuan, the average monthly purchase frequency is 4 times, and the interest tags are concentrated on "home appliance promotion" and "environmental protection and energy saving" products. In the geographical distribution features, the activity heat of user A is mainly concentrated in areas A and C. The extracted user features are stored in the tag management module in the form of classification tags for subsequent analysis and call.

[0190] A visualization chart of the user community was generated through the data visualization module. In the chart, the community where user A is located is marked in blue, the size of the node reflects the total amount of consumption of the users, and the thickness of the edge reflects the purchase similarity between users. Through the interactive interface, the platform can quickly locate high-value user groups. The average monthly consumption of the 20 users in the blue community reaches 5,000 yuan, mainly concentrated in home appliances and smart home products, showing extremely high consumption potential.

[0191] In order to verify the effectiveness of the present invention, a comparison was made with the traditional method. The traditional method is based on a single data source and relies on a rule classification algorithm to perform user grouping. On the same user data set, the number of communities grouped by the traditional method is 200, with an average internal community correlation of 0.74 and an external correlation of 0.26. It is impossible to effectively distinguish the consumption behavior characteristics of some users. In addition, the user analysis report generated by the traditional method is presented in a static table form, and analysts need to spend extra time to extract key data. In contrast, the method of the present invention increases the number of communities to 350, the internal correlation is increased to 0.92, and the external correlation is reduced to 0.15 through multi-source data integration, dynamic community division and visual display, and the generation time of the user analysis report is shortened by about 70%.

[0192] Experimental data show that after using the method of the present invention, the platform's precision marketing conversion rate increased from 7.4% of the traditional method to 11.6%. For home appliance promotion activities, the average order amount of users in the blue community increased by 15%, and the number of active users increased by 18%. In the promotion of clothing products, the purchase frequency in the orange community increased by 22%.

[0193] In summary, this Example 1 proves the feasibility and effectiveness of the present invention through actual scenarios, solves the problems of insufficient multi-source data integration, low community division accuracy, and non-intuitive data display in traditional methods in user portrait analysis, and provides efficient support for the platform's marketing decisions.

[0194] The present invention comprehensively constructs a user multi-source dataset by integrating users' social media activity data, transaction record data, geographic location information and interactive behavior data. Compared with traditional user portrait technology that relies on a single data source, the present invention significantly improves the accuracy and completeness of user portraits through the aggregation and feature extraction of multi-dimensional data. It combines interest preference tags, behavioral similarity and social interaction intensity information in the calculation of user association relationships, which can more accurately reflect the strength of associations between users and lay a high-quality foundation for the construction of user relationship networks.

[0195] The present invention adopts the Louvain algorithm to perform dynamic community detection on nodes in the user relationship network, and realizes the accurate division of user communities through modularity gain optimization and layer-by-layer progressive division of the aggregation network. Compared with the traditional method, the present invention can dynamically adjust the internal and external correlations of the community to make the community division results more consistent and reliable. In the community feature extraction, the high similarity and strong correlation of users within the community are ensured by comprehensively considering interest preferences, behavior patterns and geographical distribution, providing enterprises with more efficient tools in user grouping and precision marketing.

[0196] The present invention designs a data visualization module based on user relationship networks and community detection results, which intuitively displays complex user community distribution and its key features through node size, color, edge thickness and area marking. Compared with the traditional way of presenting user portraits in tables or static charts, the present invention achieves a clear expression of user community characteristics through dynamic visualization charts, and supports interactive operations, which significantly improves the convenience and operability of user portrait analysis. The spatial characteristics of the community can be quickly identified through heat maps and geographic distribution maps, providing strong support for enterprises to optimize resource allocation and geography-related precision delivery.

[0197] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A user portrait intelligent analysis method based on data visualization, characterized in that: The steps include: S1. Build a user multi-source data set, clean the collected user multi-source data set, and remove noise data, incomplete data, and duplicate data; S2. Extract the association information between users based on the user multi-source data set, and use the user association information to build a user relationship network; S3. Use the Louvain algorithm to cluster nodes in the user relationship network, generate multiple user communities, and dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights; S4. Extract features from user nodes in each user community and store the user features in the form of classification labels in the label management module; S5. Based on the user relationship network and community detection results, a visualization chart is generated through the data visualization module to present the distribution and key features of the user community in a graphical form. Different colors and area markers are used to distinguish user communities. The node size and color encode the user feature weight, and the thickness and color of the edge reflect the relationship strength inside and outside the community; S6. Based on user portrait tags and community grouping results, use machine learning algorithms to further analyze user behavior data and predict users’ future interest preferences or behavior trends.

2. According to the user portrait intelligent analysis method based on data visualization according to claim 1, it is characterized in that: The S1 is specifically as follows: S11. Obtain user social media activity data, transaction record data, geographic location data, and user interaction behavior data from different data sources through the data collection module to construct an initial user multi-source data set: D={D s ,D t ,D l ,D i }; Among them, D s Represents the user's activity records on social media platforms, including likes, comments, and sharing behavior data. t Represents the user's transaction record, including transaction time, amount and commodity category, D l Indicates the user's geographic location information, including GPS coordinates and timestamp, D i Represents the user's interactive behavior data, including message sending frequency, interactive objects, and time intervals; S12, performing data cleaning on the initial user multi-source data set D, marking and removing abnormal data in the initial user multi-source data set D by using a noise detection algorithm, processing incomplete data by using an interpolation method or a deletion method, and determining the uniqueness of data records in the initial user multi-source data set D by using a hash algorithm, removing records with the same hash value, and obtaining a cleaned user multi-source data set D'; S13, standardize the cleaned user multi-source data set D', unify the format and range of the data, and map the feature values ​​of different data sources to the same range [0,1]; S14, integrating the cleaned and standardized user multi-source data sets into a unified user multi-source data set D u : Among them, u k represents the kth user, D′ s (u k ),D′ t (u k ),D′ l (u k ),D′ i (u k ) represent the cleaned and standardized k Related social media activity data, transaction record data, geographic location data and interactive behavior data, where n represents the number of users.

3. According to the user portrait intelligent analysis method based on data visualization according to claim 1, it is characterized in that: The S2 is specifically as follows: S21. Based on user multi-source dataset D u Extract association information between users, including social relationships, behavioral similarities, and common interest tags; S22. For user multi-source dataset D u For every pair of users u i and user u j , calculate the social relationship similarity R s (u i ,u j ), the social relationship similarity is measured by the user social media activity data D′ s (u i ) and D′ s (u j )calculate: Among them, A(u i ) represents user u i The collection of social media interactions, |A(u i )∩A(u j )| represents user u i and user u j Common social media interactions, |A(u i )∪A(u j )| represents user u i and user u j Total number of social media interactions; S23, using cosine similarity based on transaction record data D' t (u i ), D′ t (u j ) and interactive behavior data D′ i (u i ), D′ i (u j ) Calculate user u i and user u j The behavioral similarity between b (u i ,u j ): Among them, V(u i )=[D′ t (u i ),D′ i (u i )] represents user u i The behavioral characteristic vector, V(u j )=[D′ t (u j ),D′ i (u j )] represents user u j The behavioral feature vector of , · represents the vector dot product, ||V(u i )|| represents the vector V(u i )’s module length; S24. Identify user u i and user u j The set of common interest tags between c (u i ,u j ): T c (u i ,u j )=T(u i )∩T(u j ); Among them, T(u i ) indicates that from user u i The set of interest tags extracted from the data, ∩ represents the intersection operation of the set, T c (u i ,u j ) indicates the number of common interests; S25. Calculate user u i and user u j The association strength weight w ij : Among them, α, β, γ are weight coefficients |T(u i )∪T(u j )| represents user u i and user u j The number of elements in the union of interest tags, Indicates the similarity of common interest tags; S26. Build a user relationship network: G = (V, E); Where V = {u1,u2,…,u n } is the user node set, n represents the number of users, E={e ij } is the weighted edge set between users, satisfying the association strength weight w ij ≥δ, where δ is the preset association strength threshold; S27. In the user relationship network G, users are represented by nodes, and the relationships between users are represented by weighted edges e. ij The edge weight w is expressed in the form ij Reflect user u i and user u j The strength of the association between them.

4. The user portrait intelligent analysis method based on data visualization according to claim 1 is characterized in that: The S3 is as follows: S31, using the Louvain algorithm to divide the node set V and edge set E in the user relationship network G into communities with the maximization of modularity Q as the objective function; S32, initialization, each user u i Considered an independent community i ; S33. For each node u in the network i , traverse its neighbor user u j Community C j , calculate the user u i Move to Community C j The modularity gain ΔQ after: Where m is the total edge weight in the network, w ij For user u i and user u j The edge weight between i For user u i The degree of user u i The sum of the weights of the connected edges, For user u i With Community C j The sum of edge weights between internal nodes, For Community C j The total edge weight inside, For Community C j The sum of the degrees of all nodes in S34, user u i Move to community C where the modularity gain ΔQ is maximum and greater than zero j If there is no gain, the original community remains unchanged and step S33 is repeated until the movement of any node in the network cannot increase the modularity Q; S35, construct a new aggregation network, regard each community C as a new node, and the new edge weight is the sum of the edge weights between the original communities, forming a new node set V' and edge set E', repeating steps S31 to S34, and dividing the new aggregation network into communities until the modularity Q converges, and obtaining the final community division result; S36. Dynamically adjust the community division results based on the similarity of nodes and the distribution of edge weights, so that the generated user community has high internal correlation and low external correlation, and meets the following modularity maximization conditions: Among them, δ(c i ,c j ) is the Kronecker delta function. When user u i and user u j When they belong to the same community, δ(c i ,c j )=1, otherwise δ(c i ,c j )=0.

5. The user portrait intelligent analysis method based on data visualization according to claim 1 is characterized in that: The S4 is specifically as follows: S41, based on the user interest preference data from user u i The interest tag set T(u i ) Extract interest preference features F p (u i ), defined as the weight distribution of users on each interest tag; S43, based on the user behavior pattern data from user u i Transaction record data D′ t (u i ) and interactive behavior data D′ i (u i ) Extract behavioral pattern features F b (u i ), defined as the frequency and temporal distribution of user behaviors; S44, based on the user geographical distribution data from user u i Geographic location information D′ l (u i ) Extract geographical distribution features F g (u i ); S45, extracting the interest preference feature F p (u i ), Behavioral pattern characteristics F b (u i ) and geographical distribution characteristics F g (u i ) is integrated into the user feature set F(u i ).

6. The user portrait intelligent analysis method based on data visualization according to claim 1 is characterized in that: The S5 is specifically as follows: S51. Construct a user community visualization data structure based on the user relationship network G = (V, E) and the community detection results, including the following contents: User node u i ∈V user feature set F(u i ); The edge set E of the user relationship network = {e ij }, each edge e ij The weight w ij Reflects the strength of association between users; Community division results {C1, C2, ..., C m }, where each community {C1,C2,…,C m }Contains a subset of user nodes S52. Based on the user community visualization data structure, a visualization chart of the user community is generated through a data visualization module and displayed according to the following rules: Node: A circle represents a user node u i , the size of the node S(u i ) is proportional to the sum of the feature weights: S(u i )=a p ·W p (u i )+a b ·W b (u i )+a g ·W g (u i ); Among them, W p (u i )、W b (u i )、W g (u i ) represent the total weights of interest preference, behavior pattern and geographical distribution characteristics, α p ,α b ,α g is the display scale factor of feature weight; Edge: A straight line represents the relationship between user nodes. ij , the thickness of the edge T(e ij ) and the correlation strength w ij Proportional, colors distinguish the relationship between inside and outside the community; Color: Community C k All nodes and edges of are color-coded uniformly, and the brightness of the color is proportional to the mean of the feature weights: Among them, L(C k ) represents community C k Color brightness; S53. Based on the geographical distribution characteristics of the user community F g (u i ), display community C through heat map or geographical distribution map k Geographical features: Map node location: determined by the user's geographic feature set F g (u i ) in loc k Decide; Popularity: by access frequency freq k Or weight w k Sure; S54. Support users' dynamic operations on visual charts through interactive interfaces, including zooming in, zooming out, filtering and focusing functions; S55. Generate a dynamic report based on the visualization results of the user community, showing the following: Each community C k Key features of the data, including interest tag distribution, behavior pattern distribution, and geographic distribution; User node u i Detailed feature data of the , including its association strength and coreness measure in the community.

7. A user portrait intelligent analysis system based on data visualization, characterized in that: Includes the following modules: The data collection module is used to collect user data from multiple data sources such as user social media activities, transaction records, geographic location information, and user interaction behaviors, and update the initial user multi-source data set in real time; The data cleaning and standardization module is used to clean the initial user multi-source data, remove noise data, incomplete data and duplicate data, and standardize the data; The user relationship network construction module is used to extract the association information between users from the cleaned and standardized multi-source data to generate a user relationship network; The community detection module is used to cluster nodes in the user relationship network, dynamically adjust the community division results using the Louvain algorithm, and output the division results as node sets and corresponding feature data of multiple user communities; The user feature extraction and labeling module is used to extract features of user nodes in each user community, including interest preferences, behavior patterns, and geographic distribution features, generate a user feature set, and store the features in the form of classified labels in the label management module; The data visualization module is used to generate visualization charts of user communities based on user relationship networks and community detection results. The data visualization module supports the following functions: User node size and color are dynamically encoded according to user feature weights; User relationships show the strength of association by edge thickness and color; User communities are distinguished by color and regional markers, and the brightness reflects the overall feature weight level of the community; Support heat map and geographic distribution map to display user geographic characteristics; Provides an interactive interface that supports zooming in, zooming out, filtering, and focusing operations; Intelligent analysis and dynamic optimization module, which is used to dynamically analyze and predict user behavior patterns, interest preferences and geographical distribution based on user portrait tags and community grouping results, and respond to real-time changes in data, automatically adjust user relationship networks and community grouping results, and update user portraits; The report generation module is used to generate dynamic reports based on the user community analysis results of data visualization, intuitively displaying the key features of each community and the information of user nodes.

Images