Personality prediction method and device
By integrating the timing behavior data of multi-social media platforms, using Big Five personality labels and timing alignment technology, and combining pre-trained models for feature fusion, the problem of insufficient identification of personality traits in the existing technology is solved, and a comprehensive portrayal and high-accuracy prediction of individual personality traits is achieved.
Patent Information
- Application Number
- CN202510248491.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-04
- Publication Date
- 2025-06-24
AI Technical Summary
The prior art lacks the utilization of multiple media information and long-term feature data in personality trait recognition, resulting in insufficient accuracy and comprehensiveness of the recognition results.
By obtaining the user's multi-social media timing behavior data, the pre-constructed Big Five personality labels are used for annotation and feature extraction, combining dynamic time alignment algorithms and attention mechanisms for timing alignment, and finally using the pre-trained timing data analysis model and multi-output regression model for feature fusion and personality prediction.
It realizes a comprehensive, meticulous and accurate portrayal of user personality traits, and improves the generalization ability and accuracy of the prediction model.
Smart Images
Figure CN120197016A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the technical field of natural language processing, and particularly relates to a personality prediction method and apparatus. Background Art
[0002] With the continuous development of the information society, the Internet and social media have become increasingly popular. The public shares their lives on various social media platforms, governments and media use social platforms to release information and news, and politicians use social platforms to express their opinions and promote political views. These social networking platforms record a vast amount of text, image, and behavioral data, which contain the conscious and unconscious behaviors of the publishers during the behavior process, reflecting the most real thinking and inner activities of the publishers. The Big Five personality theory, also known as the Five-Factor Model, was proposed by American psychologists Paul Costa and Robert McCrae. The development of this theory can be traced back to the 1950s. After years of research and verification, the Big Five personality theory has gradually become one of the mainstream models for describing individual personality traits. It classifies human personality traits into five dimensions, namely neuroticism, extraversion, openness, agreeableness, and conscientiousness. Personality traits are considered to be the tendency to maintain relatively consistent behavior patterns at different times and in different situations. Therefore, personality traits can be identified through a series of social media behaviors.
[0003] In the past, personality trait assessment mainly relied on scales, which was time-consuming and laborious and the results were somewhat subjective. In recent years, the mainstream personality trait recognition methods have emerged to classify personality traits through constructing behavioral characteristics and machine learning methods. However, this method only focuses on the static characteristics of the target and the data is single. At present, there is still a lack of personality trait recognition methods under multiple media information and long-time series characteristic data. Summary of the Invention
[0004] The purpose of this application is to provide a personality prediction method and apparatus to overcome the defects of the prior art, aiming to accurately predict the user's personality.
[0005] The purpose of this application is achieved through the following technical solutions:
[0006] A personality prediction method, the method includes:
[0007] Obtain the multi-social media time-series behavior data of the user;
[0008] Annotate the multi-social media time-series behavior data with pre-constructed personality labels, establish an annotated data set, and extract features from the annotated data set to obtain a personality performance feature set;
[0009] Perform time-series alignment on the features in the same time period in the personality performance feature set to obtain aligned behavior time-series features for multiple time periods;
[0010] Merge and integrate the alignment behavior time series features in the same time period to obtain a fused time series feature sequence;
[0011] Obtain the time series features of the fused time series feature sequence using a pre-trained time series data analysis model;
[0012] Input the time series features of the fused time series feature sequence into a multi-output regression model to output the predicted personality of the user.
[0013] Furthermore, the obtaining of the multi-social media time series behavior data of the user includes:
[0014] Obtain the social media platforms and data types of the multi-social media time series behavior data to be collected;
[0015] According to the data type, call the API interface to access the social media platform to obtain the timestamped behavior data of the user's multiple social media.
[0016] Furthermore, the method for constructing the personality labels includes:
[0017] Obtain the score results of different dimensions collected by the Big Five Personality Inventory, and determine the grading criteria for each label according to the score results;
[0018] Obtain the predefined label meanings, and create corresponding label sets for each dimension according to the predefined label meanings and the grading criteria to obtain the pre-constructed Big Five Personality labels.
[0019] Furthermore, the feature extraction of the labeled data set to obtain the personality performance feature set includes:
[0020] Divide the data in the labeled data set into text data, image data, and video data;
[0021] Extract the features in the text data through natural language processing to obtain text features;
[0022] Perform feature encoding on the image data to obtain image features;
[0023] Perform dynamic feature capture on the video data to obtain video features;
[0024] Integrate the text features, image features, and video features to obtain the personality performance feature set.
[0025] 5. The personality prediction method according to claim 4, wherein the performing of the dynamic feature capture on the video data includes:
[0026] Extract frames from the video data, use a pre-trained deep neural network to perform feature encoding on each frame image, and use a three-dimensional convolutional neural network to capture the dynamic features in the video data.
[0027] Further, the temporal alignment of the features in the same time period in the personality performance feature set includes:
[0028] Align the different modality features in the personality performance feature set within a similar time range to obtain pre-aligned features;
[0029] Segment the pre-aligned features according to a preset time period length to obtain pre-aligned features for multiple time periods;
[0030] Use the dynamic time warping algorithm, combined with the attention mechanism, to calculate the distance between the different modality features of the pre-aligned features for each time period to find the optimal alignment path;
[0031] Align the different modality features of the pre-aligned features for each time period based on the optimal alignment path to obtain the aligned behavioral temporal features for multiple time periods.
[0032] Further, the use of the dynamic time warping algorithm, combined with the attention mechanism, to calculate the distance between the different modality features of the pre-aligned features for each time period to find the optimal alignment path includes:
[0033] Within each time period, use the attention mechanism to calculate the attention weights between the different modality features, and use the DTW algorithm to calculate the distance matrix between the modality features with attention weights. The DTW cumulative distance matrix D(i,j) based on attention is expressed as:
[0034] D(i, j) = a i，j ·d(x i , y j ) + min{a i-1,j ·D(i - 1, j), a i，j-1 ·D(i, j - 1), a i-1，j ·D(i - 1, j - 1)};
[0035]
[0036] score(x i , y j ) = dor(w q x i , w k y j );
[0037] In the formula, score(x i , y j ) represents the time point xi , y j The similarity score between, a i,j Indicates the attention weight, d(x i , y j ) represents the Euclidean distance between the features at two time points of TW, w q and w k Indicates the feature weight matrix.
[0038] Furthermore, the merging and integration of the aligned behavior time series features in the same time period to obtain the fused time series feature sequence includes:
[0039] Construct an ensemble learning model based on multiple meta-models, where the meta-models include decision tree models, support vector machines, and neural networks;
[0040] Train the ensemble learning model to obtain a fused model;
[0041] Use the fused model to extract the aligned behavior time series features within each time period, and arrange the extracted features in chronological order to obtain the combined features for each time period;
[0042] Fuse the combined features of all time periods to obtain a fused time series feature sequence.
[0043] Furthermore, inputting the time series features of the fused time series feature sequence into a multi-output regression model to output the predicted user personality includes:
[0044] Obtain a pre-constructed multi-output regression model, adjust the input layer of the pre-constructed multi-output regression model according to the time series features captured by the time series data analysis model, and adjust the output layer of the pre-constructed multi-output regression model according to the personality dimensions to obtain an adjusted multi-output regression model;
[0045] Collect historical time series features and corresponding personality prediction results as a sample set, and use the sample set to train and test the adjusted multi-output regression model to obtain a trained multi-output regression model;
[0046] Input the time series features captured by the time series data analysis model into the trained multi-output regression model, and obtain the predicted user personality output by the trained multi-output regression model.
[0047] On the other hand, the present application also provides a personality prediction device, and the device includes:
[0048] An acquisition module, which acquires the multi-social media time series behavior data of the user;
[0049] An extraction module that annotates the multi-social media time-series behavior data with pre-built personality tags, establishes an annotated data set, and extracts features from the annotated data set to obtain a personality performance feature set;
[0050] An alignment module that performs time-series alignment on the features in the same time period in the personality performance feature set to obtain aligned behavior time-series features for multiple time periods;
[0051] A merging module that merges and integrates the aligned behavior time-series features in the same time period to obtain a fused time-series feature sequence;
[0052] An analysis module that obtains the time-series features of the fused time-series feature sequence using a pre-trained time-series data analysis model;
[0053] A prediction module that inputs the time-series features of the fused time-series feature sequence into a multi-output regression model and outputs the predicted personality of the user.
[0054] The beneficial effects of this application are as follows:
[0055] (1) By integrating the time-series behavior data of users across social media platforms and using pre-built Big Five personality tags for annotation, this application ensures the richness and pertinence of the data. Moreover, it uses the dynamic time warping algorithm combined with the attention mechanism to effectively solve the time-series alignment problem of behavior data in different time periods, thus being able to capture the key changes and patterns in user behavior more precisely.
[0056] (2) By fusing multi-time period time-series feature sequences and using a pre-trained time-series data analysis model to extract deep-level time-series features, this application significantly improves the generalization ability and accuracy of the prediction model.
[0057] (3) This application uses the trained multi-output regression model to output the predicted personality of the user, achieving a comprehensive, detailed, and accurate portrayal of the individual's personality characteristics. Description of the Drawings
[0058] Figure 1 is a flowchart of the personality prediction method according to an embodiment of this application;
[0059] Figure 2 is a schematic structural diagram of the personality prediction device according to an embodiment of this application. Detailed Embodiments
[0060] The following uses specific specific examples to illustrate the implementation manners of the present application. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.
[0061] Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope protected by the present application.
[0062] In the past, personality trait assessment mainly relied on scales, which was time-consuming and laborious and the results were somewhat subjective. In recent years, in the mainstream personality trait recognition methods, there has emerged a method of classifying personality traits by constructing behavioral characteristics and machine learning methods. However, this method only focuses on the static characteristics of the target and the data is single. Currently, there is still a lack of a personality trait recognition method under multiple media information and long-time series characteristic data.
[0063] In order to solve the above technical problems, the following various embodiments of the personality prediction method and device of the present applicant are proposed.
[0064] Refer to Figure 1 , as Figure 1 shown is the flowchart of the personality prediction method of the embodiment of the present application. The method includes the following steps:
[0065] S101. Obtain the multi-social media time-series behavior data of the user;
[0066] S102. Based on the pre-constructed Big Five personality labels, annotate the multi-social media time-series behavior data of the user, establish an annotated data set, and perform feature extraction on the annotated data set to obtain personality features;
[0067] S103. Based on the dynamic time warping algorithm introducing an attention mechanism, perform time-series alignment on the personality features to obtain aligned behavior time-series features;
[0068] S104. Merge the aligned behavior time-series features in the same time period, integrate multiple merged aligned behavior time-series features in the same time period to obtain a fused time-series feature sequence;
[0069] S105. Use the pre-trained time-series data analysis model to process the fused time-series feature sequence, and obtain the time-series features captured by the pre-trained time-series data analysis model;
[0070] S106. Input the time series features captured by the time series data analysis model into the trained multi-output regression model, and obtain the predicted user personality output by the trained multi-output regression model.
[0071] In this embodiment, the integration method based on stacking generalization combines the aligned behavior time series features in the same time period to obtain a fused time series feature sequence composed of multiple time periods.
[0072] In this embodiment, time series data analysis models such as LSTM and RNN are used to process its time series features to capture the time dependence and changes in the data.
[0073] The personality prediction method provided in this embodiment integrates the time series behavior data of users across social media platforms and uses pre-constructed Big Five personality tags for annotation, ensuring the richness and pertinence of the data. Moreover, it uses the dynamic time warping algorithm combined with the attention mechanism to effectively solve the time series alignment problem of behavior data in different time periods, so as to more accurately capture the key changes and patterns in user behavior. Further, by fusing the time series feature sequences of multiple time periods and using a pre-trained time series data analysis model to extract deep-level time series features, the generalization ability and accuracy of the prediction model are significantly improved. Finally, the predicted user personality is output by using the trained multi-output regression model, realizing a comprehensive, detailed and accurate description of individual personality characteristics.
[0074] In this embodiment, in step S101, the multi-social media time series behavior data of the user is obtained, including: obtaining the social media platforms that need to collect data and the data types that need to be collected. The data types that need to be collected include time-stamped avatars, interaction information, and user information; according to the data types that need to be collected, the API interface is called to access the social media platforms that need to collect data to obtain the time-stamped behavior data of the user in multiple social media, and the obtained behavior data is used as the multi-social media time series behavior data of the user.
[0075] Specifically, the social media platforms include Facebook, Twitter, Instagram, Weibo, WeChat, etc.
[0076] Specifically, the time-stamped interaction information includes pictures, videos, and expressions posted by the user, and the user information includes photos and videos about the user.
[0077] In this embodiment, before step S102, where the multi-social media time-series behavior data of the user is labeled based on the pre-constructed Big Five personality tags to establish a labeled data set and feature extraction is performed on the labeled data set, it further includes: obtaining the score results of five dimensions of openness, conscientiousness, extraversion, agreeableness, and neuroticism collected based on the Big Five personality inventory, and determining the grading criteria for each tag according to the score results; obtaining the predefined tag meanings, and creating corresponding tag sets for each dimension according to the predefined tag meanings and the grading criteria for each tag, so as to obtain the pre-constructed Big Five personality tags.
[0078] In this embodiment, the five-personality inventory, namely the Big Five personality inventory or the NEO personality inventory, is a personality test tool based on the Big Five personality theory. The Big Five personality inventory evaluates an individual's personality traits through the following five dimensions:
[0079] Openness to Experience: Describes the degree to which an individual is open and exploratory towards experiences.
[0080] Conscientiousness: Reflects traits such as whether an individual is responsible or unreliable, perseverant or gives up halfway, firm and steady or changeable.
[0081] Extraversion: One end is extreme extraversion and the other end is extreme introversion.
[0082] Agreeableness: Describes an individual's tendency to be pleasant and inclusive in social situations.
[0083] Neuroticism (emotional stability): Reflects the degree of emotional stability of an individual.
[0084] In this embodiment, information can be collected through the Big Five personality inventory to obtain the score results of five dimensions of openness, conscientiousness, extraversion, agreeableness, and neuroticism.
[0085] In this embodiment, according to the grading criteria and the predefined tag meanings, corresponding tag sets can be created for each dimension. For example, the tag set for openness includes conservative type, balanced type, and open type; the tag set for conscientiousness includes casual type, steady type, and conscientious type; the tag set for extraversion includes introverted type, balanced type, and extraverted type; the tag set for agreeableness includes hostile type, neutral type, and friendly type; the tag set for neuroticism includes neurotic type, balanced type, and stable type.
[0086] In this embodiment, in step S102, based on the pre-constructed Big Five personality tags, the multi-social media time-series behavior data of the user is annotated to establish an annotated data set, and feature extraction is performed on the annotated data set to obtain a personality performance feature set, including: annotating the multi-social media time-series behavior data of the user based on the pre-constructed Big Five personality tags to establish an annotated data set; dividing the data in the annotated data set into text data, image data, and video data; using natural language processing techniques to extract features from the text data to obtain text features; using a pre-trained deep neural network to perform feature encoding on the image data to obtain image features; performing frame extraction on the video data, using a pre-trained deep neural network to perform feature encoding on each frame of the image, and using a three-dimensional convolutional neural network to capture the dynamic features in the video data to obtain video features; integrating the text features, image features, and video features to obtain a personality performance feature set.
[0087] Specifically, natural language processing techniques (such as TF-IDF, Word2Vec, BERT, etc.) are used to perform vector representation on the text data to extract text features.
[0088] Specifically, a pre-trained deep neural network (such as VGG, ResNet, etc.) is used to perform feature encoding on the image data to obtain image features.
[0089] Specifically, a three-dimensional convolutional neural network (such as C3D, I3D, etc.) is used to capture the spatio-temporal dynamic features in the video.
[0090] In this embodiment, the integration of text features, image features, and video features can be achieved through simple feature splicing, weighted summation, or using more complex fusion methods (such as attention mechanisms).
[0091] In this embodiment, by comprehensively integrating multi-modal social media time-series behavior data and combining advanced technologies such as natural language processing, deep learning, and three-dimensional convolutional neural networks, the personality performance features of the user are comprehensively and accurately extracted, realizing an in-depth understanding and characterization of the user's Big Five personality traits, and providing rich and efficient data support for fields such as personalized recommendation, mental health assessment, and academic research.
[0092] In this embodiment, in step S103, the dynamic time warping algorithm is used to perform temporal alignment on the personality performance feature set in combination with the attention mechanism to obtain aligned behavioral temporal features, including: performing feature alignment on different modality features within a similar time in the personality performance feature set to obtain pre-aligned features; segmenting the pre-aligned features according to a preset time period length to obtain pre-aligned features for multiple time periods; using the dynamic time warping algorithm to calculate the distances between different modality features of the pre-aligned features for each time period in combination with the attention mechanism to find the optimal alignment path; and aligning the different modality features of the pre-aligned features for each time period based on the optimal alignment path to obtain aligned behavioral temporal features for multiple time periods.
[0093] Specifically, since the data sampling rates of different modalities may be different, it is necessary to synchronize them in time first. The modality data can be adjusted to a unified time resolution by means of interpolation or resampling. Then, the continuous pre-aligned features are segmented according to the defined time period length, and the features within each time period form a pre-aligned feature group. Next, within each time period, the attention mechanism is used to calculate the attention weights between different modality features, and the DTW algorithm is used to calculate the distance matrix between the modality features with attention weights. Finally, according to the optimal alignment path, the time indices of the modality features are adjusted to achieve precise alignment of different modalities in time, that is, the modality features aligned according to the optimal path within each time period are merged to form the aligned behavioral temporal features for that time period.
[0094] In this implementation, the DTW cumulative distance matrix D(i,j) based on attention is expressed as:
[0095] D(i,j) = a i,j ·d(x i ,y j ) + min{a i-1,j ·D(i - 1,j), a i,j-1 ·D(i,j - 1), a i-1,j ·D(i - 1,j - 1)};
[0096]
[0097] score(x i ,y j ) = dot(w q x i , w k y j );
[0098] In the formula, score(x i ,y j ) represents the time points x i ,y jThe similarity score between them, a i,j represents the attention weight, d(x i , y j ) represents the Euclidean distance between the features of two time points of TW, w q and w k represent the feature weight matrix.
[0099] In this embodiment, by combining the dynamic time warping algorithm with the attention mechanism, the precise alignment of multi-modal data in the personality performance feature set in time series is realized. It can flexibly handle the stretching and bending of different modal data on the time axis, and at the same time use the attention mechanism to emphasize key features, so as to effectively extract more accurate and consistent aligned behavior time series features.
[0100] In this embodiment, in step S104, the aligned behavior time series features in the same time period are merged, and multiple merged aligned behavior time series features in the same time period are integrated to obtain a fused time series feature sequence, including: constructing an ensemble learning model based on multiple meta-models, where the meta-models include decision tree models, support vector machines, and neural networks; training the ensemble learning model to obtain a fused model; using the fused model to extract the aligned behavior time series features in each time period, and arranging the extracted features in chronological order to obtain the merged features of each time period; fusing the merged features of all time periods to obtain a fused time series feature sequence.
[0101] In this embodiment, the ensemble learning model includes a decision tree model, a support vector machine, and a neural network.
[0102] Specifically, during the training process, hyperparameters such as the weights and learning rates of each meta-model can be adjusted to optimize the performance of the model.
[0103] Specifically, the merged features of all time periods can be fused using the weighted average method to obtain a fused time series feature sequence.
[0104] Further, before merging the aligned behavior time series features in the same time period to obtain a fused time series feature sequence composed of multiple time periods, it also includes: normalizing and denoising the aligned behavior time series features of multiple time periods.
[0105] Specifically, methods such as Z-score normalization and Min-Max normalization can be used to scale each feature value to the same numerical range.
[0106] Specifically, a filter (such as a low-pass filter, median filter, etc.) can be used for smoothing, or statistical methods (such as the IQR method to detect and remove outliers) can be used for denoising.
[0107] In this solution, by leveraging the advantages of different meta-models, the generalization ability and robustness of the overall model can be improved.
[0108] In this embodiment, in step S106, the temporal features captured by the temporal data analysis model are input into the trained multi-output regression model, and the predicted user personality output by the trained multi-output regression model is obtained, including: obtaining a pre-constructed multi-output regression model, adjusting the input layer of the pre-constructed multi-output regression model according to the temporal features captured by the temporal data analysis model, and adjusting the output layer of the pre-constructed multi-output regression model according to the personality dimensions to obtain an adjusted multi-output regression model; collecting historical temporal features and corresponding personality prediction results as a sample set, and using the sample set to train and test the adjusted multi-output regression model to obtain a trained multi-output regression model; inputting the temporal features captured by the temporal data analysis model into the trained multi-output regression model, and obtaining the predicted user personality output by the trained multi-output regression model.
[0109] In this embodiment, the temporal features captured by the temporal data analysis model include features such as trends, seasonality, and periodicity in the fused temporal feature sequence.
[0110] It can be understood that the multi-output regression model is a machine learning model used to handle regression problems involving predicting two or more target variables.
[0111] Exemplarily, the pre-constructed multi-output regression model includes an input layer, a hidden layer, and an output layer. The input layer of the pre-constructed multi-output regression model is adjusted according to the temporal features captured by the temporal data analysis model, and the output layer of the pre-constructed multi-output regression model is adjusted according to the personality dimensions to obtain an adjusted multi-output regression model. That is, the number of features in the input layer depends on the specific dimensions of the data, and the number of neurons in the output layer should match the number of personality dimensions to be predicted. For example, if five main personality dimensions of a user (such as openness, conscientiousness, extraversion, agreeableness, neuroticism) need to be predicted, then there should be five neurons in the output layer. The multi-output regression model includes one or more hidden layers, and each layer includes a certain number of neurons.
[0112] The personality prediction method in the embodiment of the present invention is described above. Next, the device in the embodiment of the present invention is described. Refer to Figure 2 , such as Figure 2 shown is a schematic structural diagram of the personality prediction device in the embodiment of the present application. The personality prediction device specifically includes the following structures:
[0113] An acquisition module 201, configured to acquire multi-social-media temporal behavior data of a user;
[0114] The extraction module 202 is used to annotate the multi-social media time-series behavior data of users based on pre-built Big Five personality tags, establish an annotated data set, and extract features from the annotated data set to obtain a personality performance feature set;
[0115] The alignment module 203 is used to use the dynamic time warping algorithm and combine the attention mechanism to perform time-series alignment on the features in the same time period in the personality performance feature set to obtain aligned behavior time-series features for multiple time periods;
[0116] The merging module 204 is used to merge the aligned behavior time-series features in the same time period, integrate multiple merged aligned behavior time-series features in the same time period to obtain a fused time-series feature sequence;
[0117] The analysis module 205 is used to process the fused time-series feature sequence using a pre-trained time-series data analysis model and obtain the time-series features captured by the pre-trained time-series data analysis model;
[0118] The prediction module 206 is used to input the time-series features captured by the time-series data analysis model into a trained multi-output regression model and obtain the predicted personality of the user output by the trained multi-output regression model.
[0119] In the embodiments of the present application, by integrating the time-series behavior data of users across social media platforms and using pre-built Big Five personality tags for annotation, the richness and pertinence of the data are ensured. Moreover, it uses the dynamic time warping algorithm combined with the attention mechanism to effectively solve the time-series alignment problem of behavior data in different time periods, so as to be able to capture the key changes and patterns in user behavior more accurately. Further, by fusing multi-time period time-series feature sequences and using a pre-trained time-series data analysis model to extract deep-level time-series features, the generalization ability and accuracy of the prediction model are significantly improved. Finally, using the trained multi-output regression model to output the predicted personality of the user realizes a comprehensive, detailed and accurate description of the individual's personality characteristics.
[0120] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A personality prediction method, characterized in that: The method comprises: Obtain users’ multi-social media time-series behavior data; Annotating the multi-social media time-series behavior data with pre-constructed personality tags to establish an annotated data set, and performing feature extraction on the annotated data set to obtain a personality expression feature set; Performing time series alignment on the features of the personality expression feature set in the same time period to obtain aligned behavior time series features of multiple time periods; The time series features of the aligned behaviors in the same period are merged and integrated to obtain a fused time series feature sequence; Acquire the time series features of the fused time series feature sequence using a pre-trained time series data analysis model; The time series features of the fused time series feature sequence are input into a multi-output regression model to output the user's predicted personality.
2. The personality prediction method according to claim 1, characterized in that: The obtaining of the user's multi-social media time series behavior data includes: Obtain the social media platforms and data types of the multi-social media time-series behavior data to be collected; According to the data type, an API interface is called to access the social media platform to obtain the user's behavioral data with timestamps in multiple social media.
3. The personality prediction method according to claim 1, characterized in that: The method for constructing the personality label includes: Obtain the score results of different dimensions collected by the Big Five Personality Scale, and determine the grading standard of each label according to the score results; The predefined label meanings are obtained, and a corresponding label set is created for each dimension according to the predefined label meanings and the grading standard to obtain a pre-constructed Big Five personality label.
4. The personality prediction method according to claim 1, characterized in that: The feature extraction of the labeled data set to obtain a personality expression feature set includes: Dividing the data in the annotated data set into text data, image data and video data; Extract features from the text data through natural language processing to obtain text features; Performing feature encoding on the image data to obtain image features; Capturing dynamic features of the video data to obtain video features; By integrating text features, image features and video features, a personality expression feature set is obtained.
5. The personality prediction method according to claim 4, characterized in that: The capturing of dynamic features of the video data comprises: The video data is frame extracted, a pre-trained deep neural network is used to perform feature encoding on each frame image, and a three-dimensional convolutional neural network is used to capture dynamic features in the video data.
6. The personality prediction method according to claim 1, characterized in that: The time sequence alignment of the features of the personality feature set in the same period includes: Performing feature alignment on different modal features within a similar time period of the personality expression feature set to obtain pre-aligned features; The pre-alignment features are divided according to the length of the preset time period to obtain the pre-alignment features of multiple time periods; The dynamic time warping algorithm is used in combination with the attention mechanism to calculate the distance between the different modal features of the pre-aligned features in each period to find the optimal alignment path; Based on the optimal alignment path, different modal features of the pre-aligned features of each time period are aligned to obtain the alignment behavior time series features of multiple time periods.
7. The personality prediction method according to claim 6, characterized in that: The method of using the dynamic time warping algorithm and the attention mechanism to calculate the distance between different modal features of the pre-aligned features of each time period to find the optimal alignment path includes: In each time period, the attention mechanism is used to calculate the attention weights between different modal features, and the DTW algorithm is used to calculate the distance matrix between the modal features with attention weights. The attention-based DTW cumulative distance matrix D(i, j) is expressed as: D(i,j)=a i，j ·d(x i ,y j )+min{a i-1，j ·D(i-1,j),a i，j-1 ·D(i,j-1),a i-1，j ·D(i-1,j-1)}; score(x i ,y j )=dot(w q x i ,w k y j ); In the formula, score(x i ,y j ) represents the time point x i ,y j The similarity score between i,j represents the attention weight, d(x i ,y j ) represents the Euclidean distance between the features at two time points TW, w q and w k represents the feature weight matrix.
8. The personality prediction method according to claim 1, characterized in that: The step of merging and integrating the time series features of the aligned behaviors in the same period to obtain a fused time series feature sequence includes: Construct an integrated learning model based on multiple meta-models, including decision tree model, support vector machine and neural network; Train the integrated learning model to obtain a fusion model; The fusion model is used to extract the time series features of the alignment behavior in each time period, and the extracted features are arranged in chronological order to obtain the merged features of each time period; The combined features of all time periods are fused to obtain a fused time series feature sequence.
9. The personality prediction method according to claim 1, characterized in that: The step of inputting the time series features of the fused time series feature sequence into a multi-output regression model and outputting the user's predicted personality comprises: Obtain a pre-constructed multi-output regression model, adjust an input layer of the pre-constructed multi-output regression model according to the time series features captured by the time series data analysis model, and adjust an output layer of the pre-constructed multi-output regression model according to the personality dimension to obtain an adjusted multi-output regression model; Collect historical time series features and corresponding personality prediction results as a sample set, and use the sample set to train and test the adjusted multi-output regression model to obtain a trained multi-output regression model; The time series features captured by the time series data analysis model are input into the trained multi-output regression model, and the user predicted personality output by the trained multi-output regression model is obtained.
10. A personality prediction device, characterized in that: The device comprises: An acquisition module, wherein the acquisition module acquires the user's multi-social media time-series behavior data; An extraction module, wherein the extraction module annotates the multi-social media time series behavior data with a pre-constructed personality tag to establish an annotated data set, and performs feature extraction on the annotated data set to obtain a personality expression feature set; An alignment module, wherein the alignment module performs time series alignment on the features of the personality expression feature set in the same time period to obtain aligned behavior time series features of multiple time periods; A merging module, wherein the merging module merges and integrates the time series features of the aligned behaviors in the same period to obtain a fused time series feature sequence; An analysis module, wherein the analysis module obtains the time series features of the fused time series feature sequence using a pre-trained time series data analysis model; A prediction module, wherein the prediction module inputs the time series features of the fused time series feature sequence into a multi-output regression model and outputs the user's predicted personality.