An identity generation system and method that fits social network group characteristics

Abstract

The application discloses a kind of identity generation system and method fitting social network group characteristics.The system includes acquisition module, classification module, processing module and generation module;Among them, acquisition module sends the group data set after pre-processing to classification module and the group data of same network address;Classification module sends group feature vector, group characteristic series, group response series and non-response series to processing module;Processing module sends target fitting feature set to generation module;Generation module determines comprehensive fitting value, determines user identity based on comprehensive fitting value.The present application is based on multi-modal data acquisition and pre-processing, natural language processing technology and grey level co-occurrence matrix feature extraction, random forest model analysis, dynamic feature fitting and statistical matching, realizes the dynamic generation of user identity that fits social environment, improves the comprehensiveness and reliability of identity generation.

Description

TECHNICAL FIELD

[0001] The present application relates to an identity generation system fitting social network group characteristics, and also relates to a corresponding identity generation method, and belongs to the technical field of data processing. BACKGROUND

[0002] In the digital era, social networks have become an indispensable part of human society, and user behavior analysis, group feature modeling and identity generation technology have wide application value in the fields of network security, public opinion monitoring, intelligent interaction, etc. With the rapid development of Internet technology, the behavior data of social network users has shown explosive growth, and the data form has gradually evolved from single text information to a complex collection containing text, image, video, audio and other multi-modal data, which puts higher requirements on the identity generation technology in social networks. Traditional identity generation methods often focus on user text information, analyze historical speech content, vocabulary usage habits, grammar structure and other text features, build user portraits and determine the corresponding user identity. This method has certain effectiveness in the early social network environment, as the data form is single and the processing logic is relatively simple, which can quickly model the user's text interaction behavior.

[0003] However, with the rise of short videos, live broadcasts and other multimedia forms, user data has shown significant characteristics of diversification and dynamization. The visual features, action sequences, scene information and other information contained in video data can more richly reflect users' interest preferences, social habits and behavior patterns. Relying solely on text features for identity generation makes it difficult to fully capture the complete behavior trajectory of users in different social scenarios, resulting in a significant deviation between the generated user identity and the real social environment. For example, in the video social scenario, users' behaviors such as liking, commenting and sharing video content, as well as the visual elements of the video content itself, all contain a lot of key information beyond text. The traditional method cannot effectively process these data, resulting in one-sidedness of the identity generation result.

[0004] At the same time, with the continuous improvement of network security and intelligent monitoring needs, there is an urgent need to improve the accuracy, dynamic adaptability and group fitting ability of social network identity generation technology. How to effectively integrate multi-modal data, capture group dynamic characteristics and improve the real-time and accuracy of identity generation has become a research difficulty and hotspot in the current field. SUMMARY

[0005] The primary technical problem to be solved by the present application is to provide an identity generation system fitting social network group characteristics.

[0006] Another technical problem to be solved by the present application is to provide an identity generation method fitting social network group characteristics.

[0007] To achieve the above technical purposes, the present application adopts the following technical solutions:

[0008] According to a first aspect of the embodiment of the present application, a fitting social network group feature identity generation system is provided, comprising a collection module, a classification module, a processing module and a generation module.

[0009] The collection module sends the pre-processed group data set and the group data of the same network address to the classification module; the classification module sends the group feature vector, the group feature sequence, the group response sequence and the non-response sequence to the processing module; the processing module sends the target fitting feature set to the generation module; and the generation module determines the comprehensive fitting value and determines the user identity based on the comprehensive fitting value.

[0010] Preferably, the collection module acquires all user data of the social network platform according to a preset collection time, pre-processes all user data to determine a group data set, and extracts all group data of the same network address in the group data set.

[0011] The classification module determines the group feature vector based on the group data, constructs a group feature sequence according to all group feature vectors, determines the vector parameter value of each group feature vector in the group feature sequence based on the vector model, divides the group feature sequence according to the vector parameter value, and constructs a group response sequence and a group non-response sequence.

[0012] The processing module extracts the same type of group feature vector in each group response sequence and establishes a fitting feature sequence, determines the fitting feature value based on the fitting feature sequence, determines the corresponding fitting feature value according to the remaining same type of group feature vector in the group response sequence, establishes a fitting feature set with all fitting feature values, compares the fitting feature set with historical data, judges whether to adjust the fitting feature set according to the comparison result, and determines the target fitting feature set.

[0013] The generation module performs feature fitting according to the target fitting feature set, determines the comprehensive fitting value according to the result of feature fitting, and determines the user identity based on the comprehensive fitting value.

[0014] Preferably, the pre-processing includes data noise reduction, duplicate data removal and missing value processing.

[0015] The missing value processing uses a preset interpolation algorithm to interpolate the values of adjacent data.

[0016] Preferably, the group feature vector includes a text feature vector and a video feature vector.

[0017] The classification module analyzes all group data by using NLP technology, carries out word segmentation, removes stop words and stem extraction on text data in all group data, determines target text data, analyzes the relationship between words and extracts noun phrases;

[0018] The noun phrases meeting the syntactic structure are taken as keywords and the number of keywords is counted, the data size, data creation time and the number of keywords of each target text data are determined as text features, and each parameter in the text features is normalized to determine the text feature vector;

[0019] The video data in all group data is converted into corresponding gray images, and a gray level co-occurrence matrix is determined according to the gray level relationship between each pixel and other pixels in the gray image, and the video feature vector is determined based on the gray level co-occurrence matrix.

[0020] Preferably, the classification module obtains a vector data set, and divides the vector data set into a training set and a test set;

[0021] A random forest model is selected in advance, and the random forest model is iteratively trained according to the training set, and the iteratively trained random forest model is tested according to the test set;

[0022] If the test value of the current iteratively trained random forest model is greater than or equal to the test value of the previous iteratively trained random forest model, the iterative training is stopped, and the current iteratively trained random forest model is taken as the vector model, otherwise, a training union set is determined based on the test set and the training set, and the iterative training is continued according to the training union set until a preset iteration number is reached;

[0023] According to the vector model, the vector parameter value of each group feature vector is output, when the vector parameter value is greater than or equal to a preset vector parameter value, the group feature vector corresponding to the vector parameter value is divided into the group response sequence, otherwise, it is divided into the group non-response sequence.

[0024] Preferably, the classification module extracts samples from the test set and each training set respectively, extracts features from all samples according to a pre-trained recognition model, estimates the probability density according to the extracted features, and determines the feature distribution of the test set and each training set according to the result of the probability density estimation;

[0025] A loss function is constructed based on the feature distribution, and the extraction probability of each training set is determined according to the loss function, samples are extracted from the training set according to the extraction probability and composed into the training union set.

[0026] Preferably, the processing module determines a preset population feature vector range corresponding to the fitted feature sequence, wherein the preset population feature vector range comprises a first preset population feature vector and a second preset population feature vector, and the first preset population feature vector is greater than the second preset population feature vector.

[0027] The processing module divides population feature vectors of the same type greater than the first preset population feature vector in the fitted feature sequence into a first feature sequence.

[0028] The processing module divides population feature vectors of the same type less than or equal to the first preset population feature vector and greater than or equal to the second preset population feature vector in the fitted feature sequence into a second feature sequence.

[0029] The processing module divides population feature vectors of the same type less than the second preset population feature vector in the fitted feature sequence into a third feature sequence.

[0030] The processing module determines a fitted feature value according to the first feature sequence, the second feature sequence and the third feature sequence.

[0031] Preferably, the historical data comprises a plurality of historical fitted feature values.

[0032] When all fitted feature values in the fitted feature set exist in the historical data, the processing module determines not to adjust the fitted feature set, and determines the target fitted feature set according to the fitted feature set, otherwise, the processing module adjusts the fitted feature set, and determines the target fitted feature set according to the adjustment result.

[0033] Each data in the target fitted feature set is recorded as a to-be-fitted value.

[0034] Preferably, the processing module records a fitted feature value in the fitted feature set that does not exist in the historical data as a replacement feature value.

[0035] All historical data containing the replacement feature value is taken as a to-be-aggregated data set, the expected cluster number k is determined as 2, the parameters of Gaussian distribution are initialized, the probability of each data in the to-be-aggregated data set belonging to each Gaussian distribution is calculated, and a responsibility value is obtained.

[0036] According to the responsibility value, a cluster to which the replacement feature value belongs is determined, all historical fitted feature values contained in the cluster are taken as a similar feature set, and a historical fitted feature mean value in the similar set is obtained.

[0037] The historical fitted feature mean value is replaced with the replacement feature value in the fitted feature set, and the target fitted feature set is determined according to the replacement result.

[0038] According to a second aspect of the embodiments of the present application, a method for generating identity fitting social network group characteristics is provided, comprising the following steps:

[0039] S1: obtaining all user data of a social network platform according to a preset collection time, pre-processing all user data to determine a group data set, and extracting all group data of the same network address in the group data set through iteration;

[0040] S2: determining a group feature vector based on the group data, constructing a group feature sequence according to all group feature vectors, determining a vector parameter value of each group feature vector in the group feature sequence based on a vector model, dividing the group feature sequence according to the vector parameter value, and constructing a group response sequence and a group non-response sequence;

[0041] S3: extracting the same type of group feature vector in each group response sequence and constructing a fitting feature sequence, determining a fitting feature value based on the fitting feature sequence, determining a corresponding fitting feature value according to the remaining same type of group feature vector in the group response sequence, establishing a fitting feature set of all fitting feature values, comparing the fitting feature set with historical data, judging whether to adjust the fitting feature set according to the comparison result, and determining a target fitting feature set;

[0042] S4: fitting features according to the target fitting feature set, determining a comprehensive fitting value according to the result of the feature fitting, and determining a user identity based on the comprehensive fitting value.

[0043] Compared with the prior art, the present application innovatively realizes the dynamic generation of user identity fitting the social environment through multi-modal data collection and preprocessing, natural language processing technology, gray level co-occurrence matrix feature extraction, random forest model analysis, dynamic feature fitting and statistical matching, effectively integrates the common characteristics of the group and adjusts in real time according to the changes of the social environment, so that the generated user identity has dynamic adaptability and avoids the risk of being out of touch with reality. At the same time, by determining the comprehensive fitting value through feature fitting, the generated user identity can more accurately fit the real characteristics of the user, significantly improving the reliability of the social robot integrating into the group, greatly improving the comprehensiveness and accuracy of identity generation, and providing strong technical support for network security, public opinion monitoring, intelligent interaction and other fields. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 For the first embodiment of the present application, a structural schematic diagram of an identity generation system fitting social network group characteristics is provided.

[0045] Figure 2 For the second embodiment of the present application, a flowchart of a method for generating identity fitting social network group characteristics is provided. DETAILED DESCRIPTION

[0046] The technical content of the present application will be described in detail below in combination with the drawings and specific embodiments.

[0047] First embodiment

[0048] As shown in Figure 1 The first embodiment of the present application provides an identity generation system for fitting social network group characteristics, which includes a collection module, a classification module, a processing module and a generation module.

[0049] Among them, the collection module sends the pre-processed group data set and the group data of the same network address (including but not limited to IP address, etc.) to the classification module; the classification module sends the group feature vector, the group feature sequence, the group response sequence and the non-response sequence to the processing module; the processing module sends the target fitting feature set to the generation module; the generation module determines the comprehensive fitting value, and determines the user identity based on the comprehensive fitting value.

[0050] In an embodiment of the present application, the collection module acquires all user data of the social network platform according to the preset collection time, determines the group data set by preprocessing all user data, traverses in the group data set, and extracts all group data of the same network address.

[0051] Among them, when all user data is preprocessed to determine the group data set, it includes: preprocessing including data noise reduction, removing duplicate data and processing missing values, and the missing values are interpolated by using a preset interpolation algorithm to the numerical values of adjacent data.

[0052] In practical applications, user data is often affected by noise interference, such as errors in network transmission and abnormal data records caused by internet failures. Group data, on the other hand, is preprocessed user data, while a group dataset is the collection of all group data. Data denoising can eliminate the influence of this noise, allowing the group data in the group dataset to accurately reflect the true behavioral characteristics of users, thereby improving the overall quality of the group dataset. The presence of duplicate data can introduce biases during data analysis. For example, if a user likes a video five times, the video like data will appear five times. During preprocessing, only the video like data from the single like is retained. Furthermore, duplicate data may lead to the overemphasis of certain features or patterns, affecting the accuracy of the analysis results. Removing duplicate data ensures that each data sample contributes independently to the analysis results, avoiding bias caused by data duplication, effectively reducing data redundancy, and improving data processing efficiency. Preset interpolation algorithms include linear interpolation, polynomial interpolation, and spline interpolation. The specific algorithm chosen depends on the actual amount of user data and is not limited here. In user data, missing values ​​may appear due to various reasons (such as equipment failure, data collection errors, etc.). Directly deleting data records containing missing values ​​may result in the loss of useful information. By interpolating the values ​​of adjacent data based on a preset interpolation algorithm, missing values ​​can be reasonably filled without deleting data records, thereby preserving data integrity and improving the overall data quality of the population dataset.

[0053] In one embodiment of the present invention, the classification module determines the group feature vector based on the group data, constructs the group feature sequence based on all the group feature vectors, determines the vector parameter value of each group feature vector in the group feature sequence based on the vector model, divides the group feature sequence according to the vector parameter value, and constructs the group response sequence and the group non-response sequence.

[0054] When determining the group feature vector based on group data, the group feature vector includes: text feature vector and video feature vector.

[0055] The classification module analyzes all group data according to natural language processing (NLP) technology, carries out word segmentation, removes stop words and stem extraction on text data in all group data, determines target text data, analyzes and extracts noun phrases between words, takes noun phrases conforming to the syntactic structure as keywords and counts the number of keywords, determines the data size, data creation time and keyword number of each target text data as text features, carries out normalization processing on each parameter in the text features, determines a text feature vector, converts video data in all group data into corresponding gray images, and determines a gray level relationship between each pixel and other pixels in the gray images to determine a gray level co-occurrence matrix, and determines a video feature vector based on the gray level co-occurrence matrix.

[0056] The natural language processing technology removes stop words (such as “of”, “has been”, “in” and other words without actual semantics or with weak semantics), reduces the amount of data while highlighting the key information, and stem extraction can unify different forms of words (such as different tenses of verbs and singular and plural forms of nouns) into stem forms (such as “beautiful” into “beautiful”) to reduce semantic redundancy. According to the dependency syntax analysis model of NLP technology, the grammatical and semantic relationship between words is analyzed, and noun phrases are extracted therefrom. Dependency syntax analysis can accurately identify noun phrases (such as “local user” and “event guest”), which reflect the theme of the target text data and are the key basis for keyword extraction. The target text data processed by the natural language processing technology can effectively extract features, determine the data size, data creation time and keyword number of each target text data as text features, quantitatively describe the target text data from multiple dimensions, the data size can reflect the length of the text, the data creation time represents the current time after the natural language processing technology processing, and the keyword number reflects the information content and the importance of the theme of the text. The text features comprehensively describe the characteristics of the target text data, and then each parameter in the text features is normalized to finally determine the text feature vector. The determination process of the text feature vector is lengthy and mature, and will not be described here.

[0057] Converting video data into a gray image reduces the dimension and complexity of the data while retaining important visual information in the video. Gray images only contain brightness information, and compared with color images, gray images focus more on the structure and texture information of the image. For image structure analysis (calculation of the gray level co-occurrence matrix), the gray image can remove color and other interference factors to facilitate the extraction of the video feature vector of the video data. The process of calculating the gray level co-occurrence matrix and determining the corresponding video feature vector is lengthy and mature, and will not be described here. By extracting the text feature vector and the video feature vector, the feature dimension of the user data is more extensive, the reliability of the social robot integrating into the group is improved, and the stability of the user identity generation is ensured.

[0058] In the determination of the vector parameter value of each group feature vector in the group feature sequence based on the vector model, the division of the group feature sequence according to the vector parameter value, the construction of the group response sequence and the group non-response sequence, the classification module obtains a vector data set containing the attributes of the group feature vector (such as the sentence length corresponding to the text feature vector, the number of paragraphs, the color distribution corresponding to the video feature vector, the optical flow feature, etc.), and divides the vector data set into a training set and a test set. A random forest model is selected in advance, and the random forest model is iteratively trained according to the training set. The iteratively trained random forest model is tested according to the test set. If the test value of the current iteratively trained random forest model is greater than or equal to the test value of the random forest model of the previous iteration, the iteration training is stopped, and the current iteratively trained random forest model is taken as the vector model. Otherwise, a training union is determined based on the test set and the training set, and the iteration training is continued according to the training union until a preset iteration number is reached. The vector parameter value of each group feature vector is output according to the vector model. When the vector parameter value is greater than or equal to a preset vector parameter value, the group feature vector corresponding to the vector parameter value is divided into the group response sequence, otherwise it is divided into the group non-response sequence.

[0059] Among them, because the group feature vector includes the text feature vector and the video feature vector, by constructing all the group feature vectors into the group feature sequence, the group feature sequence can be analyzed for features. The vector data set contains the sentence length, the number of paragraphs, the video color, the video optical flow feature, etc. of the group feature vector. By dividing the vector data set into a training set and a test set, 10-20% is usually taken as a training set, divided into 3-4 training sets, and the rest is taken as a test set to improve the generalization effect of the model. The training set is used to train the random forest model, and the test set is used to evaluate the performance of the trained model. The test indicators include accuracy, loss function value, recall rate, etc. In each iteration, the model tries to learn the patterns and relationships in the data to improve its prediction or classification ability. If the test value of the current iteratively trained random forest model is greater than or equal to the test value of the random forest model of the previous iteration, it means that the model performance has improved or remained stable, and the model has reached a satisfactory performance level. Then stop the iteration training, and take the current iteratively trained random forest model as the vector model. Otherwise, determine the training union based on the test set and the training set and continue the iteration training. In the process of continuing the iteration training, if the test value is still less than the test value of the random forest model of the previous iteration, continue to determine the training union based on the test set and the training set to ensure that the model has enough training process to learn the complex patterns and features in the data.

[0060] Meanwhile, according to the vector model, a vector parameter value of each group feature vector is output, when the vector parameter value is greater than or equal to a preset vector parameter value, the group feature vector corresponding to the vector parameter value is divided into a group response sequence, otherwise, it is divided into a group non-response sequence, for example, if the group feature vector of the color feature of a very small area has no representativeness and is irrelevant to the theme, plot and other elements, the corresponding vector parameter value is low, and the group feature vector is divided into the group non-response sequence, and the important group feature vector is divided into the group response sequence, the group response sequence is processed in a targeted manner, the efficiency of data processing is improved, the real identity of the user is approached, and the generated user identity has dynamic adaptability.

[0061] In the determination of the training set based on the test set and the training set, the classification module extracts samples from the test set and each training set respectively, performs feature extraction on all samples according to the pre-trained recognition model, and performs probability density estimation according to the extracted features, determines the feature distribution of the test set and each training set according to the result of the probability density estimation, constructs a loss function based on the feature distribution, and determines the extraction probability of each training set according to the loss function, extracts samples from the training set according to the extraction probability and forms the training set.

[0062] Specifically, samples are extracted from the test set and the training set, features are extracted and probability density estimation is performed by using the recognition model, the feature distribution of the test set and the training set can be accurately mastered, and data deviation of the model in the training process is avoided. The loss function is constructed based on the feature distribution, and the extraction probability of the training set is determined accordingly, the extraction proportion of different training set samples can be flexibly adjusted, the model focuses on the learning of key feature data, the capture ability of complex data features is improved, the training effect is optimized, the training set formed by the training set and the test set is considered comprehensively, the phenomenon of model overfitting is effectively reduced, the model is exposed to representative and diverse data in the training process, and therefore the accuracy and reliability of the test are improved.

[0063] In an embodiment of the present application, the processing module extracts the same type of group feature vector in each group response sequence and establishes a fitting feature sequence, determines a fitting feature value based on the fitting feature sequence, and determines a corresponding fitting feature value according to the remaining same type of group feature vector in the group response sequence, establishes a fitting feature set of all fitting feature values, compares the fitting feature set with historical data, judges whether to adjust the fitting feature set according to the comparison result, and determines a target fitting feature set.

[0064] The processing module determines a preset group feature vector range corresponding to the fitting feature sequence, the preset group feature vector range includes a first preset group feature vector and a second preset group feature vector, the first preset group feature vector is greater than the second preset group feature vector, the processing module divides the same type group feature vectors greater than the first preset group feature vector in the fitting feature sequence into a first feature sequence, the processing module divides the same type group feature vectors less than or equal to the first preset group feature vector and greater than or equal to the second preset group feature vector in the fitting feature sequence into a second feature sequence, the processing module divides the same type group feature vectors less than the second preset group feature vector in the fitting feature sequence into a third feature sequence, and the fitting feature value is determined according to the first feature sequence, the second feature sequence and the third feature sequence.

[0065] Specifically, the same type group feature vector represents a type parameter with a consistent video feature vector and a type parameter with a consistent text feature vector, such as a punctuation mark, a character, a scene frame, an action frame, etc. For example, a text feature vector containing characters or a video feature vector containing action frames, a fitting feature sequence is established, and the preset group feature vector range corresponding to the fitting feature sequence is represented as the range of the number of characters or the number of action frames. The accuracy of determining the fitting feature value is improved by dividing the first feature sequence, the second feature sequence and the third feature sequence, the first number of the first feature sequence is counted, the second number of the second feature sequence is counted, the third number of the third feature sequence is counted, and the fitting feature value is determined according to the following formula:

[0066]

[0067] Wherein, P is the fitting feature value; n is the first number; Hmax is the maximum value of the vector module length of the same type group feature vector in the first feature sequence; Hi is the vector module length of the i th same type group feature vector in the first feature sequence; m is the second number; Gmax is the maximum value of the vector module length of the same type group feature vector in the second feature sequence; Gj is the vector module length of the j th same type group feature vector in the second feature sequence; z is the third number; Lmax is the maximum value of the vector module length of the same type group feature vector in the third feature sequence; Lu is the vector module length of the u th same type group feature vector in the third feature sequence. It should be noted that the vector module length in the present application is the length of the same type group feature vector in its direction.

[0068] In the comparison of the fitting feature set and the historical data, the historical data includes a plurality of historical fitting feature values, when all the fitting feature values in the fitting feature set exist in the historical data, the processing module determines not to adjust the fitting feature set, and determines the target fitting feature set according to the fitting feature set, otherwise, adjusts the fitting feature set, and determines the target fitting feature set according to the adjustment result, and records each data in the target fitting feature set as a to-be-fitted value.

[0069] When all the fitting feature values in the fitting feature set exist in the historical data, it is determined that the fitting feature set is not adjusted, which avoids unnecessary data adjustment. If there is a fitting feature value that does not exist in the historical data, it may be due to calculation or other reasons, resulting in a certain deviation from the historical data. Comparing the fitting feature set with the historical data and dynamically adjusting the fitting feature set can capture the changes in the social environment in real time while maintaining a certain synchronization with the historical social data, effectively compensating for the feature mismatch problem, ensuring that the target fitting feature set can accurately reflect the data rule, thereby improving the accuracy of the fitting result, thereby approaching the user's real identity, and ensuring the reliability and stability of the generated user identity.

[0070] In the determination of adjusting the fitting feature set and determining the target fitting feature set according to the adjustment result, the processing module records the fitting feature value in the fitting feature set that does not exist in the historical data as a replacement feature value, takes the replacement feature value and the historical data as a to-be-aggregated data set, determines the expected cluster number k as 2, initializes the parameters of the Gaussian distribution, calculates the probability of each data in the to-be-aggregated data set belonging to each Gaussian distribution, and obtains the responsibility value. According to the responsibility value, it is determined to which Gaussian distribution the replacement feature value is assigned to the cluster corresponding to the Gaussian distribution, all historical fitting feature values contained in the cluster are taken as a similar feature set, and the arithmetic mean (historical fitting feature mean) of the historical fitting feature values in the similar feature set is calculated. The historical fitting feature mean is replaced with the replacement feature value in the fitting feature set, and the target fitting feature set is determined according to the replacement result.

[0071] By analyzing the historical data through the clustering algorithm, the closest data set to the current replacement feature value is found, the accuracy of replacing the replacement feature value is improved, and the reliability of the social robot integrating into the group is improved through data-driven automatic adjustment. Moreover, by comprehensively utilizing a large amount of historical data, the process of replacing the replacement feature value provides rich reference information, which can continuously accumulate from the historical data, and guarantees the reliability and stability of the user identity generation.

[0072] In one embodiment of the present application, the generating module performs feature fitting according to the target fitting feature set, determines a comprehensive fitting value according to the result of the feature fitting, and determines the user identity based on the comprehensive fitting value.

[0073] In the process of performing feature fitting according to the target fitting feature set and determining the comprehensive fitting value according to the result of the feature fitting, the generating module determines the median and the average of the target fitting feature set, extracts the to-be-fitted values greater than the median from the target fitting feature set and constructs a first fitting set, extracts the to-be-fitted values greater than the average from the target fitting feature set and constructs a second fitting set, judges whether there is an intersection between the first fitting set and the second fitting set, if yes, constructs a fitting data set according to the intersection values and takes the average of the fitting data set as the comprehensive fitting value, and if no, takes the average of the target fitting feature set as the comprehensive fitting value, and determines the user identity by traversing the database based on the comprehensive fitting value.

[0074] Whether or not the replacement operation is performed (i.e. whether or not there is a replacement feature value), the processing module finally outputs a target fitting feature set, each data item in the set being recorded as a to-be-fitted value. By determining the median and the average of the target fitting feature set and constructing different fitting sets according to the median and the average, fitting features between data can be mined from multiple perspectives, thereby obtaining a comprehensive fitting value, effectively distinguishing different distribution states in the original data (video data, text data), thereby fitting the original data in a targeted manner, and the data greater than the median and the average can reflect relatively prominent or representative features in the data. By screening and analyzing these data, the fitting result can be made to fit the actual situation of the user. If there is an intersection, it indicates that these data are typical under different measurement standards. The intersection values are constructed into a fitting data set and the average is taken as the comprehensive fitting value, which can enhance the representativeness of the fitting result. If there is no intersection, the average of the target fitting feature set is taken as the comprehensive fitting value, which ensures the integrity of the fitting process and provides an explicit and unified judgment basis for subsequent determination of the user identity in the database, improves the efficiency and accuracy of data processing, and guarantees the reliability and stability of the user identity generation.

[0075] It should be noted that in the database, each user identity information is associated with a series of feature data, which is stored in the form of feature vectors, numerical sets, etc. The comprehensive fitting value as the key feature value to be matched is the basis for judging the user identity. This value needs to be logically consistent or related to the user feature data stored in the database (usually organized in the form of feature vectors or numerical sets). The generation module finds the user feature data that best matches the comprehensive fitting value by traversing the database, thereby determining the corresponding user identity. The database has a certain mapping relationship between the user identity in the database and the comprehensive fitting value determined by the database management system (DBMS), which enables the generated user identity to fit the real features of the user.

[0076] The present application collects massive user data of social network platform in preset period, and preferentially extracts all data under the same network address, forms a group data set after data denoising, deduplication and missing value interpolation preprocessing, ensures to capture user behavior completely in the same network environment and avoids omission; the classification module extracts text feature vector and video feature vector from the preprocessed group data, uses NLP technology to perform word segmentation, stop word filtering, stem extraction, noun phrase recognition and keyword counting on the text, and performs grayscale and gray level co-occurrence matrix processing on the video data, then calculates the vector parameter value of each group feature vector by using the random forest model as the vector model, and divides the group response sequence and the group non-response sequence according to the vector parameter value; the processing module extracts the same type of group feature vector from the group response sequence to establish a fitting feature sequence, further divides it into first, second and third feature sequences according to the preset two-level threshold, calculates the fitting feature value according to the predetermined formula and constructs a fitting feature set, and compares the set with the historical fitting feature value, if there is a replacement feature value that has never appeared, finds the similar historical mean value to replace it through Gaussian mixture clustering to form a target fitting feature set; the generation module calculates the median and average of the target fitting feature set to construct the first and second fitting sets respectively, takes the intersection or the full set average of the two as the comprehensive fitting value, and finally traverses and matches in the database to determine the user identity, so as to realize multi-modal data fusion, real-time updating of group common features and dynamic generation of identity, ensure that the output identity fits the real features of the user and improve the reliability of the social robot integrating into the group.

[0077] Second embodiment

[0078] As Figure 2 shown, on the basis of the above-mentioned identity generation system, the second embodiment of the present application provides an identity generation method for fitting social network group features, at least including the following steps:

[0079] S1: acquiring all user data of a social network platform according to a preset collection time, pre-processing all the user data to determine a group data set, and traversing the group data set to extract all group data of the same network address.

[0080] S2: determining a group feature vector based on the group data, constructing a group feature sequence according to all the group feature vectors, determining a vector parameter value of each group feature vector in the group feature sequence based on a vector model, dividing the group feature sequence according to the vector parameter value, and constructing a group response sequence and a group non-response sequence.

[0081] S3: extracting the same type of group feature vector in each group response sequence and constructing a fitting feature sequence, determining a fitting feature value based on the fitting feature sequence, determining a corresponding fitting feature value according to the remaining same type of group feature vector in the group response sequence, constructing a fitting feature set with all the fitting feature values, comparing the fitting feature set with historical data, judging whether to adjust the fitting feature set according to the comparison result, and determining a target fitting feature set.

[0082] S4: performing feature fitting according to the target fitting feature set, determining a comprehensive fitting value according to the result of the feature fitting, and determining a user identity based on the comprehensive fitting value.

[0083] It should be noted that the above multiple embodiments are only illustrative. The technical solutions of various embodiments can be combined, and all are within the protection scope of the present application.

[0084] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise explicitly specified.

[0085] The identity generation system and method for fitting social network group features provided by the present application are described in detail above. Any obvious modification made by a person skilled in the art without departing from the essential content of the present application will constitute an infringement of the patent right of the present application and will bear the corresponding legal responsibility.

Claims

1. An identity generation system that fits the characteristics of social network groups, characterized in that... The system includes a data acquisition module, a classification module, a processing module, and a generation module. The data acquisition module sends a preprocessed group dataset and group data from the same network address to the classification module. The classification module sends group feature vectors, group feature sequences, group response sequences, and non-response sequences to the processing module. The processing module sends a target fitting feature set to the generation module. The generation module determines a comprehensive fitting value and identifies the user based on this value. The acquisition module acquires all user data from the social network platform according to a preset acquisition time, preprocesses all user data to determine a group dataset, and traverses the group dataset to extract all group data from the same network address. The classification module determines the group feature vector based on the group data, constructs the group feature sequence based on all the group feature vectors, determines the vector parameter value of each group feature vector in the group feature sequence based on the vector model, and divides the group feature sequence according to the vector parameter value to construct the group response sequence and the group non-response sequence. The processing module extracts the same type of group feature vectors from each group response sequence and establishes a fitting feature sequence. Based on the fitting feature sequence, it determines the fitting feature value and determines the corresponding fitting feature value according to the remaining same type of group feature vectors in the group response sequence. It establishes a fitting feature set for all the fitting feature values, compares the fitting feature set with historical data, and determines whether to adjust the fitting feature set based on the comparison result, and determines the target fitting feature set. The generation module performs feature fitting based on the target fitting feature set, determines a comprehensive fitting value based on the feature fitting result, and determines the user identity based on the comprehensive fitting value.

2. The identity generation system as described in claim 1, characterized in that... The preprocessing includes data noise reduction, removal of duplicate data, and handling of missing values; The missing value processing uses a preset interpolation algorithm to interpolate the values ​​of adjacent data.

3. The identity generation system as described in claim 2, characterized in that... The group feature vector includes text feature vectors and video feature vectors; The classification module uses NLP technology to analyze all group data, performs word segmentation, stop word removal and stemming on the text data in all group data, determines the target text data, analyzes the relationship between words and extracts noun phrases; Noun phrases that conform to syntactic structure are used as keywords and the number of keywords is counted. The data size, data creation time and the number of keywords of each target text data are determined as text features. Each parameter in the text features is normalized to determine the text feature vector. The video data in all group data is converted into corresponding grayscale images, and a grayscale co-occurrence matrix is ​​determined based on the grayscale relationship between each pixel and other pixels in the grayscale image. The video feature vector is then determined based on the grayscale co-occurrence matrix.

4. The identity generation system as described in claim 3, characterized in that: The classification module acquires a vector dataset and divides the vector dataset into a training set and a test set. A random forest model is pre-selected, and iterative training is performed on the random forest model based on the training set. The iteratively trained random forest model is then tested based on the test set. If the test value of the random forest model after the current iteration is greater than or equal to the test value of the random forest model after the previous iteration, the iterative training is stopped, and the random forest model after the current iteration is used as the vector model. Otherwise, a training set is determined based on the test set and the training set, and iterative training continues based on the training set until the preset number of iterations is reached. The vector model outputs the vector parameter value of each group feature vector. When the vector parameter value is greater than or equal to the preset vector parameter value, the group feature vector corresponding to the vector parameter value is assigned to the group response sequence; otherwise, it is assigned to the group non-response sequence.

5. The identity generation system as described in claim 4, characterized in that... The classification module extracts samples from the test set and each training set respectively, extracts features from all samples according to the pre-trained recognition model, estimates the probability density based on the extracted features, and determines the feature distribution of the test set and each training set based on the probability density estimation results. A loss function is constructed based on the feature distribution, and the extraction probability of each training set is determined according to the loss function. Samples are extracted from the training set according to the extraction probability and formed into the training set.

6. The identity generation system as described in claim 5, characterized in that... The processing module determines a preset group feature vector range corresponding to the fitted feature sequence. The preset group feature vector range includes a first preset group feature vector and a second preset group feature vector, wherein the first preset group feature vector is greater than the second preset group feature vector. The processing module divides the same type of group feature vectors that are greater than the first preset group feature vector into the first feature sequence; The processing module divides the same type of group feature vectors in the fitted feature sequence that are less than or equal to the first preset group feature vector and greater than or equal to the second preset group feature vector into the second feature sequence; The processing module divides the same type of group feature vectors in the fitted feature sequence that are smaller than the second preset group feature vector into a third feature sequence; The fitted feature values ​​are determined based on the first feature sequence, the second feature sequence, and the third feature sequence.

7. The identity generation system as described in claim 6, characterized in that... The historical data includes multiple historical fitted feature values; When all fitting feature values ​​in the fitting feature set exist in the historical data, the processing module determines not to adjust the fitting feature set and determines the target fitting feature set based on the fitting feature set; otherwise, the fitting feature set is adjusted and the target fitting feature set is determined based on the adjustment result. Each data point in the target fitting feature set is denoted as the value to be fitted.

8. The identity generation system as described in claim 7, characterized in that... The processing module records the fitted feature values ​​in the fitted feature set that do not exist in the historical data as replacement feature values; Take all historical data containing the replacement feature values ​​as the dataset to be aggregated, determine the expected number of clusters k as 2, initialize the parameters of the Gaussian distribution, calculate the probability that each data in the dataset to be aggregated belongs to each Gaussian distribution, and obtain the responsibility value. Based on the responsibility value, determine the cluster to which the replacement feature value belongs, and take all historical fitting feature values ​​contained in the cluster as a similar feature set, and obtain the mean value of historical fitting features in the similar set; The historical fitting feature mean is replaced with the replacement feature value in the fitting feature set, and the target fitting feature set is determined based on the replacement result.

9. The identity generation system as described in claim 8, characterized in that... The generation module determines the median and mean of the target fitted feature set; Extract the values ​​to be fitted that are greater than the median from the target fitting feature set and construct a first fitting set; Extract the values ​​to be fitted from the target fitting feature set that are greater than the average value and construct a second fitting set; The generation module determines whether there is an intersection between the first fitting set and the second fitting set; If yes, then construct a fitted dataset based on the intersection values ​​and use the mean of the fitted dataset as the comprehensive fitted value; if no, then use the mean of the target fitted feature set as the comprehensive fitted value. The user's identity is determined by iterating through the database based on the comprehensive fitted values.

10. A method for generating identities by fitting the characteristics of social network groups, implemented based on the identity generation system described in any one of claims 1 to 9, characterized in that... Includes the following steps: S1: Obtain all user data from the social network platform according to the preset collection time, preprocess all user data to determine the group dataset, traverse the group dataset, and extract all group data from the same network address; S2: Determine the population feature vector based on the population data, construct the population feature sequence based on all population feature vectors, determine the vector parameter value of each population feature vector in the population feature sequence based on the vector model, divide the population feature sequence according to the vector parameter value, and construct the population response sequence and the population non-response sequence; S3: Extract the feature vectors of the same type of group in each group response sequence and establish a fitting feature sequence. Determine the fitting feature value based on the fitting feature sequence, and determine the corresponding fitting feature value according to the remaining feature vectors of the same type of group in the group response sequence. Establish a fitting feature set for all fitting feature values, compare the fitting feature set with historical data, and determine whether to adjust the fitting feature set based on the comparison results. Determine the target fitting feature set. S4: Perform feature fitting based on the target fitting feature set, determine the comprehensive fitting value based on the feature fitting result, and determine the user identity based on the comprehensive fitting value.

Images