CN-121996843-A - Broadband service dynamic compliance user portrait construction and recommendation system

CN121996843ACN 121996843 ACN121996843 ACN 121996843ACN-121996843-A

Abstract

The invention discloses a broadband service dynamic compliance user portrayal construction and recommendation system, which relates to the field of data analysis and management and comprises an acquisition module, a preprocessing module and a processing module, wherein the acquisition module is used for acquiring multi-source heterogeneous modal data of a user in a broadband service scene, performing desensitization processing on the acquired data, the preprocessing module is used for performing missing value filling and outlier detection on the desensitized multi-source heterogeneous modal data, completing standardization regulation and cross-modal dimension adaptation based on semantic attributes of the multi-modal data and outputting semantically preprocessed data of adaptation feature fusion.

Inventors

YUAN LING
YE QI
WANG TING
QIU TIANYUAN
CHEN YUNYUN
YU DONGDONG

Assignees

湖北邮电规划设计有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. A broadband service dynamic compliance user portrayal construction and recommendation system is characterized by comprising: the acquisition module is used for acquiring multi-source heterogeneous modal data of a user in a broadband service scene and performing desensitization processing on the acquired data; The preprocessing module is used for executing missing value filling and abnormal value detection on the desensitized multi-source heterogeneous modal data, completing standardization and cross-modal dimension adaptation based on semantic attributes of each modal data and outputting semantic preprocessing data with adaptation characteristics fused; the fusion module is used for completing cross-modal feature association and depth fusion of semantically preprocessed data based on the transform coding and the attention weight distribution, and generating a high-dimensional fusion feature set with uniform dimension; the construction module is used for carrying out grouping tag identification on the user based on the high-dimensional fusion feature set, extracting user business preference, behavior time sequence rule and compliance attribute information and constructing a dynamic compliance user portrait; The early warning module is used for capturing dynamic change characteristics of user business behavior data through LSTM time sequence modeling, executing real-time early warning analysis of user loss risk and synchronously generating portrait update triggering signals; The recommendation module is used for generating targeted service recommendation according to the latest dynamic compliance user portrait; The multi-source heterogeneous modal data comprises user structured attribute data, network behavior log data, service use track data and terminal association interaction data.
2. The system for constructing and recommending dynamic compliance user portraits for broadband services according to claim 1, wherein said desensitizing process in said acquisition module comprises: for a core sensitive field in the user structured attribute data, a desensitization mode of combining a feature mask and semantic replacement is used; and for the network behavior log data, the service use track data and the terminal association interaction data, a dynamic disturbance desensitization mode is used, and the disturbance amplitude is adaptively adjusted according to the mode type of the data.
3. The broadband business dynamic compliance user portrayal construction and recommendation system of claim 1, wherein the missing value filling of the preprocessing module obeys: ; Wherein: filling the result for the missing value of the jth modal data of the ith user; total number of heterogeneous modes for multiple sources; the cross-modal semantic similarity between the jth modality and the kth modality is obtained; confidence weight for the kth modality; the valid data value for the kth modality for the ith user.
4. The broadband business dynamic compliance user portrayal construction and recommendation system according to claim 3, wherein the preprocessing module is operative to construct a cross-modal semantic mapping matrix based on semantic attributes of each modal data Matrix elements Semantic association strength of the q-th feature of the p-th mode and the unified dimension feature of the target is represented, and the dimension importance score of each mode feature is calculated through information gain And finally, comparing and screening the core features based on the dimension importance scores and a preset screening threshold value, and performing dimension adaptive conversion on the core feature synchronization: ; Wherein: the uniform dimension feature matrix is obtained after the adaptation; A feature matrix standardized for each mode data; Is a dimension importance weight matrix.
5. The broadband service dynamic compliance user portrayal construction and recommendation system of claim 3, wherein the outlier detection by the preprocessing module is compliant with: ; Wherein: an abnormality determination result of the jth modal data of the ith user, 1 indicating that the abnormality is determined, 0 indicating that the abnormality is determined; A statistical deviation degree of the jth modal data of the ith user relative to the modal feature distribution; semantic deviation coefficients for the j-th modality; a preset abnormality judgment threshold value; If it is ∈( , ], For presetting a moderate anomaly threshold, judging the moderate anomaly, and based on cross-modal semantic similarity Confidence weight to modality Selecting the first N modal data with highest semantic association degree to construct a correction standard, calculating a correction value through weighted average, if > And judging that the model is seriously abnormal, and adopting a mode confidence weighting replacement mode, wherein the replacement value is the average result of the statistical mean value and the filling logic of all the normal data in the mode.
6. The broadband service dynamic compliance user portrayal construction and recommendation system according to claim 1, wherein the cross-modal feature fusion of the fusion module is based on a dual-driven attention mechanism of modal contribution degree-feature correlation, and the attention weights of all modes are fused in depth by combining with a transducer code ; Wherein: attention weight for the j-th modality; A contribution coefficient for the j-th modality; the average characteristic mutual information of the j-th mode and all other modes; Is the total number of multi-source heterogeneous modes.
7. The system for constructing and recommending dynamic compliance user portraits for broadband services according to claim 1, wherein the user grouping tag identification in the construction module determines compliance priority weights based on compliance attribute information in a high-dimensional fusion feature set, and then applies user service preference features and behavior time sequence rule features to divide users into a plurality of groupings with similar features through an adaptive density clustering algorithm, and for each grouping, extracts core features to generate a multi-level tag system, wherein the multi-level tag system comprises a compliance level tag, a service preference tag and a behavior mode tag, and the weights of the tags are as follows: ; Wherein: Final weight for the tag; Priority coefficients for compliance attributes; Is a compliance priority weight; the association weight of the tag and the compliance attribute is; and the association weight of the label and the business behavior characteristic.
8. The broadband service dynamic compliance user portrayal construction and recommendation system according to claim 1, wherein the process of capturing the dynamic change characteristics of the user service behavior data in the early warning module is as follows: S1, performing time sequence regulation on preprocessed user business behavior data, and aligning the data according to uniform time granularity; S2, self-adaptively dividing time sequence windows based on user behavior liveness, so that each window contains complete behavior period characteristics; S3, extracting business behavior core characteristics in each time sequence window; S4, fitting behavior core feature sequences in each time sequence window through linear regression, and calculating to obtain a user behavior trend slope; S5, calculating Euclidean distance of a behavior feature vector of a current time sequence window and a previous time sequence window, and synchronously executing normalization processing and marking as a behavior mutation coefficient; S6, feature fusion is carried out on the behavior trend slope and the behavior mutation coefficient to form a dynamic change feature set of the user business behavior data; the user loss risk early warning triggering conditions are as follows: ; Wherein: Early warning results for loss risk; A user behavior trend slope within a current time sequence window; A preset trend slope minimum threshold value; Is the coefficient of the behavior mutation; The method comprises the steps of setting a preset highest threshold value of the behavior mutation coefficient; When the early warning trigger condition is met, generating an portrait update trigger signal, and synchronously outputting loss risk early warning information, wherein the early warning information is based on a preset early warning level threshold value and a preset early warning level threshold value And comparing the determined early warning grades.
9. The broadband service dynamic compliance user portrayal construction and recommendation system according to claim 1, wherein the recommendation module generates the targeted service recommendation according to the user service preference, the behavior timing rule and the compliance attribute information corresponding to the latest dynamic compliance user portrayal after receiving the latest dynamic compliance user portrayal: Each user service preference, behavior time sequence rule and compliance attribute information are manually pre-bound with a unique service list by a system end user, the occurrence times of each matched service list are inquired based on the user service preference, behavior time sequence rule and compliance attribute information corresponding to the latest dynamic compliance user portrait, and the service list with the largest occurrence times is used as a targeted service recommendation target to be pushed to the user.
10. The system for constructing and recommending dynamic compliance user portraits for broadband services according to claim 1, wherein the acquisition module is interactively connected with the preprocessing module through a wireless network, the preprocessing module is interactively connected with the fusion module through a wireless network, the fusion module is interactively connected with the construction module through a wireless network, the construction module is interactively connected with the early warning module through a wireless network, and the early warning module is interactively connected with the recommending module through a wireless network.

Description

Broadband service dynamic compliance user portrait construction and recommendation system Technical Field The invention relates to the technical field of data analysis and management, in particular to a broadband service dynamic compliance user portrait construction and recommendation system. Background With the rapid development of digital economy, broadband services are used as a core component of digital infrastructure, the scale of users is continuously expanded, service scenes are continuously enriched, and the demands of users on personalized services are increasingly highlighted. The invention patent application with the application number 201910951783.4 discloses a user portrait construction and content recommendation method, a device and equipment, and aims to solve the problems that as each child is an independent individual, interesting things and books are different, own reading preference exists when the books are read, some children are interested in habit development, some children are interested in family, reading interest of each child needs to be known, and targeted reading recommendation is provided for each child. However, in the prior art, the problem of insufficient validity of multi-mode feature fusion exists, information redundancy or loss easily occurs when features are fused due to the isomerism of different mode data, so that accuracy and integrity of user portraits are affected. Therefore, we propose a broadband business dynamic compliance user portrayal construction and recommendation system. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a broadband service dynamic compliance user portrait construction and recommendation system which can effectively solve the problems in the prior art. In order to achieve the above object, the present invention is achieved by the following technical scheme; The invention discloses a broadband service dynamic compliance user portrait construction and recommendation system, which comprises: The system comprises an acquisition module, a preprocessing module, a pre-warning module, a fusion module, a recommendation module, a target service recommendation module and a user service recommendation module, wherein the acquisition module is used for acquiring multi-source heterogeneous modal data of a user in a broadband service scene and performing desensitization processing on the acquired data, the preprocessing module is used for performing missing value filling and outlier detection on the desensitized multi-source heterogeneous modal data, completing standardization and cross-modal dimension adaptation based on semantic attributes of the multi-source heterogeneous modal data, outputting semantically preprocessed data of adaptation feature fusion, the fusion module is used for completing cross-modal feature association and deep fusion of semantically preprocessed data based on a transitor code and attention weight distribution, and generating a unified-dimension high-dimensional fusion feature set; The acquisition module is interactively connected with the preprocessing module through a wireless network, the preprocessing module is interactively connected with the fusion module through the wireless network, the fusion module is interactively connected with the construction module through the wireless network, the construction module is interactively connected with the early warning module through the wireless network, and the early warning module is interactively connected with the recommendation module through the wireless network; The multi-source heterogeneous modal data comprises user structured attribute data, network behavior log data, service use track data and terminal association interaction data. Still further, the desensitizing treatment in the acquisition module includes: for a core sensitive field in the user structured attribute data, a desensitization mode of combining a feature mask and semantic replacement is used; and for the network behavior log data, the service use track data and the terminal association interaction data, a dynamic disturbance desensitization mode is used, and the disturbance amplitude is adaptively adjusted according to the mode type of the data. Still further, the missing value filling of the preprocessing module obeys: ; Wherein: filling the result for the missing value of the jth modal data of the ith user; total number of heterogeneous modes for multiple sources; the cross-modal semantic similarity between the jth modality and the kth modality is obtained; confidence weight for the kth modality; the valid data value for the kth modality for the ith user. Furthermore, the preprocessing module operates the stage to construct a cross-mode semantic mapping matrix based on semantic attributes of each mode dataMatrix elementsSemantic association strength of the q-th feature of the p-th mode and the unified dimension feature of the target is represented, and th