A multi-dimensional user feature-driven interface interaction design method and system
By predicting the trajectory endpoint and amplifying the intention control, the problem of unstable user interface interaction in dynamic environments is solved, improving the accuracy and convenience of user operation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU HUASHANG UNIV
- Filing Date
- 2025-07-11
- Publication Date
- 2026-06-30
AI Technical Summary
In dynamic environments, such as subways or buses, users find it difficult to keep the device stable, resulting in a high touch error rate, decreased operation accuracy, and a poor user experience.
By predicting the endpoint of a user's sliding trajectory in an unstable state, candidate controls are identified, and based on multi-dimensional user characteristics such as historical usage frequency and functional logical relevance, the most likely intended control is amplified to improve the convenience of interaction.
In unstable conditions, the accuracy and convenience of user interface interaction are improved, thus enhancing the user experience.
Smart Images
Figure CN120803411B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of interface interaction design technology, and specifically to a multi-dimensional user feature-driven interface interaction design method and system. Background Technology
[0002] Current mobile terminal interface interaction design still primarily focuses on optimization for static, stable environments, neglecting user needs and behavioral characteristics in dynamic scenarios. Especially on public transportation such as subways and buses, the vibrations, acceleration, deceleration, and steering of the vehicle make it difficult for users to maintain their balance, leading to instability in handheld devices. In this unstable environment, touch error rates increase, and user operation accuracy decreases. This results in frequent operational errors, task interruptions, and repetitive operations, leading to a poor user experience. Summary of the Invention
[0003] This invention executes a mobile phone unstable interface interaction mode on user interaction operations in an unstable environment. It can predict the sliding trajectory generated by user interaction operations in an unstable environment and obtain the endpoint of the predicted trajectory. The endpoint of the predicted trajectory is the position where the user will interact with the mobile phone at the next monitoring time point. Based on the endpoint of the predicted trajectory, candidate controls are determined. Combining multi-dimensional user characteristics such as the historical usage frequency ratio between the user and the candidate controls and the strength of functional logic correlation, the most likely intention control for the user to interact next is determined. The intention control is then magnified, making the interaction between the user and the intention control more convenient and improving the user interface interaction experience in an unstable environment.
[0004] This invention provides a multi-dimensional user feature-driven interface interaction design method, comprising:
[0005] Determine if the phone is in an unstable state. If the phone is in an unstable state, activate the unstable interface interaction mode.
[0006] In the unstable interface interaction mode of the mobile phone, the system responds to the user's touch operation with the mobile phone interface and obtains the user interaction location data at the current monitoring time point. The user interaction location data is the X and Y values corresponding to the position touched by the user on the mobile phone interface. The user interaction location data at the current monitoring time point and the user interaction location data corresponding to the previous N-1 monitoring time points are combined to form a user interaction location time series set. The user interaction location time series set is sent to the trajectory prediction model for processing and outputs the predicted trajectory endpoint corresponding to the next monitoring time. Based on the predicted trajectory endpoint, a candidate control set is obtained. The candidate control set includes several candidate controls. The candidate controls are controls on the current mobile phone interface that are within a preset distance from the predicted trajectory endpoint. Based on the distance correlation value, historical usage frequency ratio and functional logic correlation strength of the candidate controls, the intention score corresponding to all candidate controls in the candidate control set is calculated. All candidate controls in the candidate control set are arranged from largest to smallest according to the corresponding intention score. The top N candidate controls are selected as intention controls and a magnification operation is set for the intention controls.
[0007] The trajectory prediction model is based on the LSTM model. Compared with the commonly used LSTM model, the trajectory prediction model adds enhanced updates to the forget gate weight matrix and input gate weight matrix in the LSTM model.
[0008] As a preferred aspect, the enhanced update of the forget gate weight matrix and input gate weight matrix in the trajectory prediction model specifically includes the following steps:
[0009] The trajectory type data vector is determined based on the user interaction location time series set. The trajectory type data vector includes the standard deviation of velocity, the standard deviation of acceleration, and the standard deviation of the rate of change of direction.
[0010] A self-attention mechanism is executed on the forget gate weight matrix based on the trajectory type data vector. The corresponding value vector and key vector are constructed based on the forget gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vector.
[0011] A self-attention mechanism is performed on the input gate weight matrix based on the trajectory type data vector. The corresponding value vector and key vector are constructed based on the input gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vector.
[0012] As a preferred aspect, the intention score for all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. This specifically includes the following steps:
[0013] For each candidate control, perform the following operations: calculate the distance between the center of the candidate control and the endpoint of the predicted trajectory, denoted as the distance correlation value; calculate the ratio of the number of times the candidate control is used to the total number of times all candidate controls in the candidate control set are used, denoted as the historical usage frequency ratio; determine the component corresponding to the last completed interaction operation through the operation log, denoted as the target component; form a function pair (target component, candidate control) with the target component and the candidate control; match the function pair with the logical function library; if the match is successful, output the functional logical correlation strength as 1; if the match fails, output the functional logical correlation strength as 0. The logical function library includes several logical function pairs; form an intention analysis vector from the distance correlation value, historical usage frequency ratio, and functional logical correlation strength of the candidate control; then feed the intention analysis vector into the intention analysis network for processing, and output the intention score corresponding to the candidate control.
[0014] As a preferred aspect, this also includes real-time training of the intention analysis network, with the following specific steps:
[0015] After a user performs an interaction with the mobile phone interface, the component that performs the interaction is recorded as the training component. The intention analysis vector corresponding to the training component at the previous monitoring time point is obtained and recorded as the training intention analysis vector. The training intention analysis vector is fed into the current intention analysis network for training. During training, the training intention analysis vector is used as the input of the intention analysis network to obtain the predicted data volume output by the intention analysis network. The real-time loss value is calculated based on the difference between the predicted data volume and 1. Based on the real-time loss value, the parameters of the intention analysis network are adjusted using the gradient descent method to achieve real-time training of the intention analysis network.
[0016] As a preferred aspect, the intention control is set to be magnified, which specifically includes the following steps: the magnification operation setting refers to magnifying the intention control to (1+A) times its original size, where A is the intention score corresponding to the intention control.
[0017] As a preferred approach, training the trajectory prediction model includes the following steps:
[0018] Obtain several trajectory prediction training samples, including N+1 user interaction location data. Combine all trajectory prediction training samples into a trajectory prediction training set. Train the trajectory prediction model using the trajectory prediction training set. During training, use the first N user interaction location data from the trajectory prediction training samples as the input to the trajectory prediction model, and use the last user interaction location data from the trajectory prediction training samples as the target output of the trajectory prediction model. Calculate the trajectory prediction loss value and determine whether the trajectory prediction loss value is within a first preset range. If the trajectory prediction loss value is within the first preset range, output the trained trajectory prediction model; otherwise, continue training the trajectory prediction model using the trajectory prediction training set.
[0019] As a preferred approach, training the intention analysis network includes the following steps:
[0020] Obtain several intention analysis training samples, each containing an intention analysis vector. Label the intention analysis training samples with intention scores, each set having a score of 1. Combine all labeled intention analysis training samples into an intention analysis training set. Train the intention analysis network using this training set. During training, use the intention analysis training samples as input to the network and the labeled intention scores as the target output. Calculate the intention analysis loss value and determine if it falls within a second preset range. If it does, output the trained intention analysis network; otherwise, continue training the network using the training set.
[0021] This invention also provides a multi-dimensional user feature-driven interface interaction design system, comprising:
[0022] The mobile phone instability judgment module is used to determine whether the mobile phone is in an unstable state. If the mobile phone is in an unstable state, the mobile phone instability interface interaction mode is activated.
[0023] The trajectory prediction module is used to respond to the user's touch operation with the mobile phone interface and obtain the user interaction location data at the current monitoring time point. The user interaction location data is the X and Y values corresponding to the position touched by the user on the mobile phone interface. The user interaction location data at the current monitoring time point and the user interaction location data corresponding to the previous N-1 monitoring time points are combined to form a user interaction location time series set. The user interaction location time series set is sent to the trajectory prediction model for processing and outputs the predicted trajectory endpoint corresponding to the next monitoring time.
[0024] The intention control determination module is used to obtain a set of candidate controls based on the predicted trajectory endpoint. The set of candidate controls includes several candidate controls. The candidate controls are controls on the current mobile phone screen that are within a preset distance from the predicted trajectory endpoint. The intention score of all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. All candidate controls in the candidate control set are arranged in descending order of their corresponding intention scores, and the top N candidate controls are selected as intention controls.
[0025] The intention control magnification module is used to set magnification operations for intention controls.
[0026] The present invention has the following advantages:
[0027] This invention executes a mobile phone unstable interface interaction mode on user interaction operations in an unstable environment. It can predict the sliding trajectory generated by user interaction operations in an unstable environment and obtain the endpoint of the predicted trajectory. The endpoint of the predicted trajectory is the position where the user will interact with the mobile phone at the next monitoring time point. Based on the endpoint of the predicted trajectory, candidate controls are determined. Combining multi-dimensional user characteristics such as the historical usage frequency ratio between the user and the candidate controls and the strength of functional logic correlation, the most likely intention control for the user to interact next is determined. The intention control is then magnified, making the interaction between the user and the intention control more convenient and improving the user interface interaction experience in an unstable environment. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the structure of the multi-dimensional user feature-driven interface interaction design system used in an embodiment of the present invention. Detailed Implementation
[0029] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.
[0030] Example 1: A multi-dimensional user feature-driven interface interaction design method, comprising:
[0031] The phone's state data is collected using its built-in gyroscope and accelerometer. This data includes three-axis acceleration and three-axis angular velocity, which characterize the phone's motion in the user's hand and determine if the phone is unstable. Instability can occur when the user is on a subway or standing on a bus, where their center of gravity is unstable. In such a state, the user's interaction with the phone interface is affected, making accurate interaction impossible. The phone state data is processed by a phone state analysis model, which outputs phone state labels, including stable and unstable labels. If the phone state label is stable, the phone is not unstable, and the data collection continues for the next time stamp. If the phone state label is unstable, the phone is unstable, and an unstable interface interaction mode is activated. The phone state analysis model is based on a backpropagation (BP) neural network.
[0032] In the unstable interface interaction mode, the system responds to user touch operations on the phone screen and acquires user interaction position data at the current monitoring time point. This user interaction position data consists of the X and Y values corresponding to the user's touch position on the phone screen. The coordinate system for these X and Y values is generally based on the bottom left corner of the phone's vertical screen as the origin, with the X-axis pointing horizontally to the right and the Y-axis pointing vertically upwards. The user interaction position data at the current monitoring time point, along with the user interaction position data from the previous N-1 monitoring time points, forms a user interaction position time series. If the user wants to interact with the phone, in the case of phone instability, the user's thumb will slide on the phone. The sliding trajectory can be used to predict the next position the user will interact with the phone. The user interaction position time series is fed into a trajectory prediction model for processing, outputting the predicted trajectory endpoint for the next monitoring time point. The predicted trajectory endpoint is output in the form of X and Y values. Based on the predicted trajectory endpoint, a candidate control set is obtained. The candidate control set includes several candidate controls, which are located on the current phone screen at a preset distance from the predicted trajectory endpoint. The distance to the control is set by the developer. The control is a component on the mobile interface that can interact to perform functions, such as copy, paste, and forward in the chat interface, and add to cart and view cart in the purchase interface. The intention score of all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. The historical usage frequency ratio and functional logic correlation strength are constructed based on the user's interaction behavior in the history record, reflecting the multi-dimensional user characteristics corresponding to the user's interaction with the mobile interface. The intention score represents the probability that the user may interact with the control next. All candidate controls in the candidate control set are arranged from largest to smallest according to their corresponding intention scores. The top N candidate controls are selected as intention controls. The intention controls are the controls that the user is most likely to interact with next. A zoom-in operation setting is set for intention controls. In order to make it easier for users to interact with intention controls, a zoom-in operation setting can be used to make the interaction between users and intention controls more convenient.
[0033] The trajectory prediction model is based on the LSTM model. Compared with the commonly used LSTM model, the trajectory prediction model adds enhanced updates to the forget gate weight matrix and input gate weight matrix in the LSTM model. The LSTM model generally consists of LSTM units as the basis. Each time series analysis involves feeding the user interaction location data in the user interaction location time series set into the LSTM unit for processing, specifically including forget gate processing, input gate processing, and output gate processing. Forget gate processing determines which information is discarded through the forget gate weight matrix, input gate processing determines which information is retained, and output gate processing determines which information is output. The specific forget gate processing, input gate processing, and output gate processing operations refer to existing LSTM models and will not be elaborated here. The trajectory prediction model in this application enhances and updates the forget gate weight matrix and input gate weight matrix in the commonly used LSTM model, highlighting the weight ratio of different trajectory prediction methods in trajectory prediction.
[0034] This application executes a mobile phone unstable interface interaction mode on user interaction operations in an unstable environment. It can predict the sliding trajectory generated by user interaction operations in an unstable environment and obtain the endpoint of the predicted trajectory. The endpoint of the predicted trajectory is the position where the user will interact with the mobile phone at the next monitoring time point. Based on the endpoint of the predicted trajectory, candidate controls are determined. Combining multi-dimensional user characteristics such as the historical usage frequency ratio between the user and the candidate controls and the strength of functional logic correlation, the most likely intention control for the user to interact next is determined. The intention control is then magnified, making the interaction between the user and the intention control more convenient and improving the user interface interaction experience in an unstable environment.
[0035] In the trajectory prediction model, the reinforcement update of the forget gate weight matrix and input gate weight matrix in the LSTM model specifically includes the following steps:
[0036] The trajectory type data vector is determined based on the user interaction location time series set. This data vector includes the standard deviations of velocity, acceleration, and direction change rate. The standard deviation of velocity is determined by dividing the distance between user interaction location data points corresponding to adjacent monitoring time points by the time difference to obtain the tangential velocity at each monitoring time point. The standard deviation of velocity is then calculated based on the tangential velocities at all monitoring time points. Similarly, the standard deviation of acceleration is determined by dividing the difference between tangential velocities at adjacent monitoring time points by the time difference to obtain the acceleration at each monitoring time point. The standard deviation of acceleration is then calculated based on the accelerations at all monitoring time points. Finally, the standard deviation of the direction change rate is determined by dividing the vector between the user interaction location data at the monitoring time point and the user interaction location data at the next monitoring time point by the corresponding time difference. The direction change rate is then determined by dividing the tangential direction at adjacent monitoring time points by the corresponding time difference. The standard deviation of all direction change rates is then calculated.
[0037] A self-attention mechanism is applied to the forget gate weight matrix based on trajectory type data vectors. The corresponding value vector and key vector are constructed based on the forget gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vectors. The trajectory type represented by the trajectory type data vectors guides which information is discarded. For example, if the standard deviations of velocity, acceleration, and direction change rate in the trajectory type data vectors are relatively low, it indicates that the sliding trajectory corresponding to the trajectory type data vectors is a linear and stable trajectory. Therefore, the probability of discarding information from nonlinear trajectory predictions will increase.
[0038] A self-attention mechanism is executed on the input gate weight matrix based on the trajectory type data vector. The corresponding value vector and key vector are constructed based on the input gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vector. The trajectory type represented by the trajectory type data vector guides which information is retained. For example, if the standard deviations of velocity, acceleration, and direction change rate in the trajectory type data vector are relatively low, it indicates that the sliding trajectory corresponding to the trajectory type data vector is a linear stable trajectory. Therefore, the probability of retaining the information of linear trajectory prediction will increase.
[0039] The self-attention mechanism, referencing the Transformer model setup, typically involves multiplying the input vector with the value weight matrix, key weight matrix, and query weight matrix respectively to construct the corresponding value vector V, key vector K, and query vector Q. The self-attention mechanism is then implemented using the following formula: H = softmax(QK). T / D 0.5)V, where H is the vector output by the self-attention mechanism, T is the matrix transpose operation, and D is the dimension of the key vector K; the value weight matrix, key weight matrix, and query weight matrix in the trajectory prediction model of this application are trained end-to-end with the trajectory prediction model;
[0040] The intention score for all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. The specific steps include the following:
[0041] For each candidate control, the following operations are performed: Calculate the distance between the center of the candidate control and the endpoint of the predicted trajectory, denoted as the distance correlation value; calculate the ratio of the number of times the candidate control is used to the total number of times all candidate controls in the candidate control set are used, denoted as the historical usage frequency percentage. It should be noted that the usage count of each control is statistically recorded. Identify the component corresponding to the last completed interaction operation through the operation log, denoted as the target component. Form a function pair (target component, candidate control) between the target component and the candidate control. Match the function pair with the logical function library. If the match is successful, output a functional logical correlation strength of 1; if the match fails, output a functional logical correlation strength of 0. The logical function library includes several logical function pairs, which are (component, component). The logical function library can be set by the operator or constructed using the corresponding operation logs on the mobile phone. Combine the distance correlation value, historical usage frequency percentage, and functional logical correlation strength of the candidate control into an intention analysis vector. Then, feed the intention analysis vector into an intention analysis network for processing, outputting the intention score corresponding to the candidate control. The intention analysis network is based on a BP neural network.
[0042] To improve the accuracy of intention scoring analysis, real-time training of the intention analysis network is also included, with the following specific steps:
[0043] After a user performs an interaction with the mobile phone interface, the component that performs the interaction is recorded as the training component. The intention analysis vector corresponding to the training component at the previous monitoring time point is obtained and recorded as the training intention analysis vector. The training intention analysis vector is fed into the current intention analysis network for training. During training, the training intention analysis vector is used as the input of the intention analysis network to obtain the predicted data volume output by the intention analysis network. The real-time loss value is calculated based on the difference between the predicted data volume and 1. Based on the real-time loss value, the parameters of the intention analysis network are adjusted using the gradient descent method to achieve real-time training of the intention analysis network. It should be noted that using the component that the user actually interacts with as feedback to train the intention analysis network in real time can make the analysis of the score more in line with the user's actual needs and further improve the user's interface interaction experience in unstable state environments.
[0044] The zoom-in operation setting for the intention control includes the following steps: The zoom-in operation setting refers to zooming the intention control to (1+A) times its original size, where A is the intention score corresponding to the intention control;
[0045] Training the mobile phone status analysis model involves the following steps:
[0046] Several training samples for mobile phone status analysis are obtained. These training samples include mobile phone status analysis vectors, which are collected by developers on the actual mobile phones. The training samples are labeled with mobile phone status tags. All labeled training samples are combined into a training set. The mobile phone status analysis model is trained using this training set. During training, the training samples are used as input to the model, and the labeled mobile phone status tags are used as the target output. The accuracy of the model is then evaluated. If the accuracy meets expectations, the trained model is output; otherwise, the model is trained again using the training set.
[0047] Training the trajectory prediction model involves the following steps:
[0048] Acquire several trajectory prediction training samples, including N+1 user interaction location data points. These N+1 user interaction location data points are actual data collected under simulated instability conditions. Combine all trajectory prediction training samples into a trajectory prediction training set. Train the trajectory prediction model using this training set. During training, use the first N user interaction location data points from the trajectory prediction training samples as input to the trajectory prediction model, and the last user interaction location data point as the target output. Calculate the trajectory prediction loss value and determine if it falls within a first preset range (set by the developers). If the trajectory prediction loss value falls within the first preset range, output the trained trajectory prediction model; otherwise, continue training the trajectory prediction model using the trajectory prediction training set.
[0049] Training the intention analysis network involves the following steps:
[0050] Several intention analysis training samples are obtained, including intention analysis vectors. These vectors are constructed by developers under simulated instability conditions. After each interactive operation, the component that performed the operation is marked as a labeled component. The intention analysis vector corresponding to the labeled component at the previous monitoring time point is obtained and marked as an intention analysis training sample. The intention analysis training samples are labeled with intention scores, all of which are 1. All labeled intention analysis training samples are combined into an intention analysis training set. The intention analysis network is trained using this training set. During training, the intention analysis training samples are used as input to the network, and the intention scores labeled on the training samples are used as the target output. The intention analysis loss value is calculated, and it is determined whether the loss value is within a second preset range (set by the developers). If the loss value is within the second preset range, the trained intention analysis network is output; otherwise, the network is trained again using the training set.
[0051] Example 2: A multi-dimensional user feature-driven interface interaction design system, see [link to example]. Figure 1 ,include:
[0052] The mobile phone instability judgment module collects mobile phone state data corresponding to the phone's current state using the phone's built-in gyroscope and accelerometer. This data includes three-axis acceleration and three-axis angular velocity, which characterize the phone's motion in the user's hand and determine whether the phone is in an unstable state. Instability can occur when the user is on a subway or standing on a bus, where the center of gravity is unstable. In this state, the user's interaction with the phone interface is affected, making accurate interaction impossible. The mobile phone state data is sent to the mobile phone state analysis model for processing, outputting mobile phone state labels, including stable and unstable state labels. If the mobile phone state label is stable, the phone is not in an unstable state, and the data collection continues for the next timestamp. If the mobile phone state label is unstable, the phone is in an unstable state, and the unstable interface interaction mode is activated.
[0053] The trajectory prediction module is used to respond to touch operations between the user and the mobile phone interface to obtain the user interaction location data at the current monitoring time point. The user interaction location data here is the X and Y values corresponding to the position touched by the user on the mobile phone interface. The coordinate system corresponding to the X and Y values is generally with the lower left corner of the vertical interface of the mobile phone as the origin, the horizontal rightward direction as the X-axis, and the vertical upward direction as the Y-axis. The user interaction location data at the current monitoring time point and the user interaction location data corresponding to the previous N-1 monitoring time points are combined to form the user interaction location time series set. If the user wants to interact with the mobile phone, in the case of the mobile phone being unstable, the user's thumb will slide with the mobile phone. The sliding trajectory can be used to predict the position where the user will interact with the mobile phone next. The user interaction location time series set is fed into the trajectory prediction model for processing and outputs the predicted trajectory endpoint corresponding to the next monitoring time. The predicted trajectory endpoint is output in the form of X and Y values.
[0054] The intention control determination module is used to predict the endpoint of the trajectory and obtain a set of candidate controls. The set of candidate controls includes several candidate controls. Candidate controls are controls on the current mobile phone interface that are within a preset distance from the predicted trajectory endpoint. The preset distance is set by the developers. Controls are interactive components on the mobile phone interface that can perform functions, such as copy, paste, and forward in a chat interface, and add to cart and view cart in a purchase interface. Based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls, the intention score corresponding to all candidate controls in the candidate control set is calculated. The historical usage frequency ratio and functional logic correlation strength are constructed based on the user's interactive behavior in the historical records, reflecting the multi-dimensional user characteristics corresponding to the user's interactive operation with the mobile phone interface. The intention score represents the probability that the user may interact with the control next. All candidate controls in the candidate control set are arranged from largest to smallest according to their corresponding intention scores. The top N candidate controls are selected as intention controls. The intention controls are the controls that the user is most likely to interact with next.
[0055] The intention control zoom module is used to set zoom operation for intention controls. In order to make it easier for users to interact with intention controls, zoom operation settings can be used to make the interaction between users and intention controls more convenient.
[0056] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.
Claims
1. A multi-dimensional user feature-driven interface interaction design method, characterized in that, include: Determine if the phone is in an unstable state. If the phone is in an unstable state, activate the unstable interface interaction mode. In the unstable interface interaction mode of the mobile phone, the system responds to the user's touch operation with the mobile phone interface and obtains the user interaction location data at the current monitoring time point. The user interaction location data is the X and Y values corresponding to the position touched by the user on the mobile phone interface. The user interaction location data at the current monitoring time point and the user interaction location data corresponding to the previous N-1 monitoring time points are combined to form a user interaction location time series set. The user interaction location time series set is sent to the trajectory prediction model for processing and outputs the predicted trajectory endpoint corresponding to the next monitoring time. Based on the predicted trajectory endpoint, a candidate control set is obtained. The candidate control set includes several candidate controls. The candidate controls are controls on the current mobile phone interface that are within a preset distance from the predicted trajectory endpoint. Based on the distance correlation value, historical usage frequency ratio and functional logic correlation strength of the candidate controls, the intention score corresponding to all candidate controls in the candidate control set is calculated. All candidate controls in the candidate control set are arranged from largest to smallest according to the corresponding intention score. The top N candidate controls are selected as intention controls and a magnification operation is set for the intention controls. The trajectory prediction model is based on the LSTM model. Compared with the commonly used LSTM model, the trajectory prediction model adds enhanced updates to the forget gate weight matrix and input gate weight matrix in the LSTM model. The intention score for all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. The specific steps include the following: For each candidate control, perform the following operations: calculate the distance between the center of the candidate control and the endpoint of the predicted trajectory, denoted as the distance correlation value; calculate the ratio of the number of times the candidate control is used to the total number of times all candidate controls in the candidate control set are used, denoted as the historical usage frequency ratio; determine the component corresponding to the last completed interaction operation through the operation log, denoted as the target component; pair the target component and the candidate control into a function pair; match the function pair with the logical function library; if the match is successful, output the functional logical correlation strength as 1; if the match fails, output the functional logical correlation strength as 0. The logical function library includes several logical function pairs; combine the distance correlation value, historical usage frequency ratio, and functional logical correlation strength of the candidate control into an intention analysis vector; then feed the intention analysis vector into the intention analysis network for processing, and output the intention score corresponding to the candidate control.
2. The multi-dimensional user feature-driven interface interaction design method according to claim 1, characterized in that, In the trajectory prediction model, the reinforcement update of the forget gate weight matrix and input gate weight matrix in the LSTM model specifically includes the following steps: The trajectory type data vector is determined based on the user interaction location time series set. The trajectory type data vector includes the standard deviation of velocity, the standard deviation of acceleration, and the standard deviation of the rate of change of direction. A self-attention mechanism is executed on the forget gate weight matrix based on the trajectory type data vector. The corresponding value vector and key vector are constructed based on the forget gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vector. A self-attention mechanism is performed on the input gate weight matrix based on the trajectory type data vector. The corresponding value vector and key vector are constructed based on the input gate weight matrix, and the corresponding query vector is constructed based on the trajectory type data vector.
3. The multi-dimensional user feature-driven interface interaction design method according to claim 2, characterized in that, It also includes real-time training of the intention analysis network, with the following specific steps: After a user performs an interaction with the mobile phone interface, the component that performs the interaction is recorded as the training component. The intention analysis vector corresponding to the training component at the previous monitoring time point is obtained and recorded as the training intention analysis vector. The training intention analysis vector is fed into the current intention analysis network for training. During training, the training intention analysis vector is used as the input of the intention analysis network to obtain the predicted data volume output by the intention analysis network. The real-time loss value is calculated based on the difference between the predicted data volume and 1. Based on the real-time loss value, the parameters of the intention analysis network are adjusted using the gradient descent method to achieve real-time training of the intention analysis network.
4. The multi-dimensional user feature-driven interface interaction design method according to claim 3, characterized in that, The zoom-in setting for the intention control includes the following steps: The zoom-in setting refers to zooming in on the intention control to 1+A times its original size, where A is the intention score corresponding to the intention control.
5. The multi-dimensional user feature-driven interface interaction design method according to claim 4, characterized in that, Training the trajectory prediction model involves the following steps: Obtain several trajectory prediction training samples, including N+1 user interaction location data. Combine all trajectory prediction training samples into a trajectory prediction training set. Train the trajectory prediction model using the trajectory prediction training set. During training, use the first N user interaction location data from the trajectory prediction training samples as the input to the trajectory prediction model, and use the last user interaction location data from the trajectory prediction training samples as the target output of the trajectory prediction model. Calculate the trajectory prediction loss value and determine whether the trajectory prediction loss value is within a first preset range. If the trajectory prediction loss value is within the first preset range, output the trained trajectory prediction model; otherwise, continue training the trajectory prediction model using the trajectory prediction training set.
6. The multi-dimensional user feature-driven interface interaction design method according to claim 5, characterized in that, Training the intention analysis network involves the following steps: Obtain several intention analysis training samples, each containing an intention analysis vector. Label the intention analysis training samples with intention scores, each set having a score of 1. Combine all labeled intention analysis training samples into an intention analysis training set. Train the intention analysis network using this training set. During training, use the intention analysis training samples as input to the network and the labeled intention scores as the target output. Calculate the intention analysis loss value and determine if it falls within a second preset range. If it does, output the trained intention analysis network; otherwise, continue training the network using the training set.
7. A multi-dimensional user feature-driven interface interaction design system, characterized in that, The system applies a multi-dimensional user feature-driven interface interaction design method according to any one of claims 1-6, including: The mobile phone instability judgment module is used to determine whether the mobile phone is in an unstable state. If the mobile phone is in an unstable state, the mobile phone instability interface interaction mode is activated. The trajectory prediction module is used to respond to the user's touch operation with the mobile phone interface and obtain the user interaction location data at the current monitoring time point. The user interaction location data is the X and Y values corresponding to the position touched by the user on the mobile phone interface. The user interaction location data at the current monitoring time point and the user interaction location data corresponding to the previous N-1 monitoring time points are combined to form a user interaction location time series set. The user interaction location time series set is sent to the trajectory prediction model for processing and outputs the predicted trajectory endpoint corresponding to the next monitoring time. The intention control determination module is used to obtain a set of candidate controls based on the predicted trajectory endpoint. The set of candidate controls includes several candidate controls. The candidate controls are controls on the current mobile phone screen that are within a preset distance from the predicted trajectory endpoint. The intention score of all candidate controls in the candidate control set is calculated based on the distance correlation value, historical usage frequency ratio, and functional logic correlation strength of the candidate controls. All candidate controls in the candidate control set are arranged in descending order of their corresponding intention scores, and the top N candidate controls are selected as intention controls. The intention control magnification module is used to set magnification operations for intention controls.