CN-122020333-A - Personalized comprehensive endowment service recommendation method and system based on artificial intelligence big data

CN122020333ACN 122020333 ACN122020333 ACN 122020333ACN-122020333-A

Abstract

The application provides a personalized comprehensive endowment service recommendation method and system based on artificial intelligence big data, which relate to the technical field of computer data processing and artificial intelligence, and the application collects travel data of a plurality of target users and browsing data of community service APP, wherein the travel data comprises travel time stamps and in-out directions, and the browsing data comprises a plurality of browsing records, browsing frequency and residence time; according to the method, a behavior feature vector is built according to the record quantity of the travel data in and out directions in a preset time period, an interest feature vector is built according to browsing data, the behavior feature vector and the interest feature vector are respectively input into a neural network based on depth typical correlation analysis to generate a fusion feature vector of each target user, all the target users are divided into a plurality of social groups based on the fusion feature vector, and community activity information is sent to edge node objects in the same social group based on a low-dimensional manifold space, so that accurate social recommendation for the isolated old is realized.

Inventors

HUANG CHAO
XU ZHIBIN
ZHANG JINXIANG

Assignees

天津茵诺科技集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (10)

1. The personalized comprehensive endowment service recommendation method based on the artificial intelligence big data is characterized by comprising the following steps of: Collecting travel data of a plurality of target users and browsing data of community service APP, wherein the travel data comprise travel time stamps and in-out directions, and the browsing data comprise a plurality of browsing records, browsing frequency and stay time; According to the record quantity of the travel data in and out directions in a preset time period, a behavior feature vector is constructed, and according to the browsing data, an interest feature vector is constructed; Respectively inputting the behavior feature vector and the interest feature vector into a neural network based on depth canonical correlation analysis, and generating a fusion feature vector of each target user by jointly optimizing an objective function of the neural network; Based on the fusion feature vector, calculating Gaussian similarity among the target users to construct a Laplace matrix, mapping the Laplace matrix to a low-dimensional manifold space, and dividing all the target users into a plurality of social groups by using a clustering algorithm; and determining an edge node object and a center node object in the social group based on the low-dimensional manifold space, extracting community activity information in which the center node object participates in a high frequency, and sending the community activity information to the edge node object in the same social group.
2. The method according to claim 1, wherein the method further comprises: Acquiring the generation time of each browsing record in the browsing data, and calculating the time weight coefficient of each browsing record based on the difference value between the generation time and the current time; the constructing the interest feature vector according to the browsing data comprises the following steps: and constructing an interest feature vector according to the browsing data by utilizing the time weight coefficient of each browsing record.
3. The method of claim 2, wherein constructing a feature vector of interest from the browsing data using the temporal weight coefficients of each browsing record comprises: converting content tags of browsing records in browsing data of each target user into semantic embedded vectors, and calculating basic attention of each browsing record according to browsing frequency and stay time in the browsing data; multiplying the basic attention degree with a corresponding time weight coefficient to obtain a comprehensive weight value of each browsing record; Multiplying the semantic embedded vector of each browsing record by the corresponding comprehensive weight value to obtain a weighted semantic vector, and accumulating all weighted semantic vectors of the same target user to obtain an interest feature vector.
4. The method of claim 1, wherein separately inputting the behavioral feature vector and the interest feature vector into a neural network based on a depth canonical correlation analysis, generating a fused feature vector for each target user by jointly optimizing an objective function of the neural network, comprises: Constructing a neural network based on depth canonical correlation analysis, wherein the neural network comprises a first mapping channel with a first network weight parameter and a second mapping channel with a second network weight parameter; Inputting the behavior feature vector into the first mapping channel to obtain a first feature vector, and inputting the interest feature vector into the second mapping channel to obtain a second feature vector; Calculating covariance of the first feature vector and the second feature vector, and calculating a pearson correlation coefficient based on the covariance; Constructing an objective function taking the opposite number of the pearson correlation coefficient as a loss value, and performing joint optimization on the objective function by utilizing a gradient descent algorithm until the lifting amplitude of the pearson correlation coefficient is within a preset threshold value, and determining that the first network weight parameter and the second network weight parameter are converged; the behavior feature vector and the interest feature vector are mapped again by utilizing the converged first network weight parameter and the converged second network weight parameter to obtain a target first feature vector and a target second feature vector; And splicing the target first feature vector and the target second feature vector to obtain a fusion feature vector of each target user.
5. The method of claim 4, wherein constructing an objective function having an inverse of the pearson correlation coefficient as a loss value, jointly optimizing the objective function using a gradient descent algorithm until a magnitude of elevation of the pearson correlation coefficient is within a preset threshold, determining that the first network weight parameter and the second network weight parameter converge, comprises: randomly selecting a preset number of training users, and taking the behavior feature vector and the interest feature vector of each training user as a training sample; Stacking the first feature vectors and the second feature vectors corresponding to all training samples according to a preset arrangement mode to obtain a first feature matrix and a second feature matrix; Performing decentration and standardization on the first feature matrix and the second feature matrix respectively to generate a typical correlation matrix; Adding all main diagonal elements of the typical correlation matrix to obtain a pearson correlation coefficient sum; Taking the inverse number of the pearson correlation coefficient sum as a loss value to construct an objective function; Calculating a first partial derivative of the objective function with respect to a first network weight parameter and a second partial derivative of the objective function with respect to a second network weight parameter, respectively, using a gradient descent algorithm; And subtracting and updating the first network weight parameter and the second network weight parameter according to the first partial derivative and the second partial derivative, calculating the total sum of the updated pearson correlation coefficients until the lifting amplitude of the pearson correlation coefficients before and after updating is smaller than a preset threshold value, and determining that the first network weight parameter and the second network weight parameter are converged.
6. The method of claim 1, wherein computing gaussian similarity between the target users based on the fused feature vectors to construct a laplacian matrix, by mapping the laplacian matrix to a low-dimensional manifold space, and dividing all target users into a plurality of social groups using a clustering algorithm, comprises: Calculating Euclidean distance between any two fusion feature vectors, substituting each Euclidean distance into a Gaussian kernel function to obtain Gaussian similarity between the target users so as to construct a Gaussian similarity matrix; Constructing a diagonal matrix according to the row sum of the Gaussian similarity matrix, and performing matrix subtraction operation on the diagonal matrix and the Gaussian similarity matrix to obtain a Laplace matrix; Performing eigenvalue decomposition on the Laplace matrix to obtain a plurality of eigenvalues, performing ascending arrangement on all eigenvalues to obtain an eigenvalue sequence, selecting non-zero eigenvalues with the number of the preset dimensions from the eigenvalue sequence, and constructing an eigenvector matrix according to eigenvectors corresponding to the non-zero eigenvalues; And taking each row of data in the feature vector matrix as one coordinate point in a low-dimensional manifold space, and clustering all coordinate points in the low-dimensional manifold space to obtain a plurality of social groups, wherein target users in each social group have the same clustering label.
7. The method of claim 1, wherein determining edge node objects and center node objects in the social group based on the low-dimensional manifold space, and extracting community activity information in which the center node objects participate at high frequency, and transmitting the community activity information to edge node objects within the same social group, comprises: based on the low-dimensional manifold space, carrying out arithmetic average operation on coordinate points in each social group to obtain the center coordinate of each social group; Calculating a space Euclidean distance between a coordinate point and a central coordinate of each target user in the same social group, taking a target user with the space Euclidean distance smaller than or equal to a preset distance threshold as a central node object, and taking a target user with the space Euclidean distance larger than the preset distance threshold as an edge node object; counting the participation times of the central node object in each community activity in a preset time period, and arranging the participation times of all the community activities in a descending order to obtain a participation heat sequence of each social group; And selecting a preset number of community activities from the participation heat sequence as community activity information, acquiring terminal equipment identifiers of edge node objects in each social group, and transmitting the community activity information to the edge node objects in the same social group through the terminal equipment identifiers.
8. The personalized comprehensive pension service recommendation system based on the artificial intelligence big data is characterized by comprising: the system comprises an acquisition module, a storage module and a storage module, wherein the acquisition module is used for acquiring travel data of a plurality of target users and browsing data of community service APP, the travel data comprise travel time stamps and in-out directions, and the browsing data comprise a plurality of browsing records, browsing frequency and residence time; The construction module is used for constructing a behavior feature vector according to the record quantity of the travel data in and out directions in a preset time period and constructing an interest feature vector according to the browsing data; The optimization module is used for respectively inputting the behavior feature vector and the interest feature vector into a neural network based on depth typical correlation analysis, and generating a fusion feature vector of each target user by jointly optimizing an objective function of the neural network; the construction module is further used for calculating Gaussian similarity among the target users based on the fusion feature vectors to construct a Laplace matrix, mapping the Laplace matrix to a low-dimensional manifold space, and dividing all the target users into a plurality of social groups by using a clustering algorithm; and the determining module is used for determining an edge node object and a center node object in the social group based on the low-dimensional manifold space, extracting community activity information of the center node object participating in high frequency, and sending the community activity information to the edge node object in the same social group.
9. An electronic device, comprising: A memory for storing a computer program; A processor for implementing the steps of the artificial intelligence big data based personalized integrated endowment service recommendation method according to any of claims 1 to 7 when executing the computer program.
10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and when the computer program is executed by a processor, the computer program can implement the personalized comprehensive pension service recommendation method based on artificial intelligence big data according to any one of claims 1 to 7.

Description

Personalized comprehensive endowment service recommendation method and system based on artificial intelligence big data Technical Field The application relates to the technical field of computer data processing and artificial intelligence, in particular to a personalized comprehensive pension service recommendation method and system based on artificial intelligence big data. Background With the continuous deepening of the aging degree of population, the intelligent community aged people become a key field for improving the life quality of the aged, wherein the personalized service recommendation technology based on big data has a wide application prospect in the aspect of promoting the social participation of the aged people. The living habit and interest preference of the old are mined through the intelligent algorithm, and the method has important social value for relieving the increasingly serious social disjointing and solitary problems of the old. The existing community pension service recommendation method generally depends on basic demographic tags of the old or historical browsing records with single dimension, and mainly adopts traditional collaborative filtering or rule-based matching algorithm to push information. In the prior art, on-line digital reading behaviors and off-line physical travel rules of users are mostly regarded as split data islands, and mechanical matching is only carried out according to simple click heat or static labels, so that deep correlation analysis on multi-source heterogeneous behavior data is lacking. However, this approach of relying on only single view data has difficulty capturing the deep inherent consistency between what the elderly want and what he is going to, resulting in recommended activities that may exceed the actual mobility of the elderly or be inconsistent with their actual potential social intent. In addition, the existing method lacks effective identification of potential social network topological structures in communities, and cannot utilize the same-frequency influence of core active groups in communities to drive isolated old people, so that recommendation services are difficult to truly convert into offline social behaviors. Therefore, the technical problems of low recommendation accuracy and poor social integration effect caused by the lack of effective integration of multidimensional behavior features and social structure analysis exist in the prior art. Disclosure of Invention The application aims to provide a personalized comprehensive endowment service recommendation method and system based on artificial intelligence big data, which are used for solving the technical problems of low recommendation accuracy and poor social fusion effect caused by lack of effective fusion of multidimensional behavior characteristics and social structure analysis in the prior art. In a first aspect, the present application provides a personalized comprehensive pension service recommendation method based on artificial intelligence big data, including: Collecting travel data of a plurality of target users and browsing data of community service APP, wherein the travel data comprise travel time stamps and in-out directions, and the browsing data comprise a plurality of browsing records, browsing frequency and residence time; According to the record quantity of the travel data in and out directions in a preset time period, constructing a behavior feature vector, and according to browsing data, constructing an interest feature vector; respectively inputting the behavior feature vector and the interest feature vector into a neural network based on depth typical correlation analysis, and generating a fusion feature vector of each target user through the objective function of the joint optimization neural network; based on the fusion feature vector, calculating Gaussian similarity among target users to construct a Laplace matrix, mapping the Laplace matrix to a low-dimensional manifold space, and dividing all the target users into a plurality of social groups by using a clustering algorithm; based on the low-dimensional manifold space, determining edge node objects and center node objects in the social group, extracting community activity information of high-frequency participation of the center node objects, and sending the community activity information to the edge node objects in the same social group. Optionally, the method further comprises: Acquiring the generation time of each browsing record in the browsing data, and calculating the time weight coefficient of each browsing record based on the difference value between the generation time and the current time; constructing an interest feature vector according to browsing data, including: And according to the browsing data, constructing an interest feature vector by utilizing the time weight coefficient of each browsing record. Optionally, constructing the interest feature vector according to the browsing data by u