Interface interaction design method and system

Abstract

The invention relates to the field of man-machine interaction, in particular to an interface interaction design method and system. The interface interaction design system comprises a data acquisition module, an element layout adjustment module and an element layout adjustment strategy output model updating module. According to the method, the interface state data set is obtained by constructing the design of the interface interaction into the layout design of the elements, the effect feedback, namely the interaction feedback data set, corresponding to the interface interaction is collected, and then the element layout adjustment strategy is output through the element layout adjustment strategy output model, so that the interface interaction design conforming to the operation habits of the user is obtained; and the element layout adjustment strategy output model is updated in a reinforcement learning mode, so that the effect of parameter optimization is achieved, the element layout adjustment strategy output by the element layout adjustment strategy output model is further enabled to better fit user habits, and the user experience of user interface interaction is improved.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction, and in particular to an interface interaction design method and system. Background Art

[0002] In the existing interface interaction design technology, most system interface designs rely on unified templates or preset interaction methods, but fail to fully consider the personalized needs and usage habits of users. Traditional interface interaction design usually provides users with standardized operation processes through fixed layouts and functional modules. Although this design can meet the basic needs of general users, it lacks flexibility and adaptability for users with different usage habits, resulting in insufficient user experience during use and difficulty in achieving optimal operation efficiency and satisfaction. Summary of the invention

[0003] The present invention constructs the design of interface interaction as the layout design of elements, obtains an interface state data set, collects effect feedback corresponding to the interface interaction, i.e., an interaction feedback data set, and then outputs the element layout adjustment strategy through an element layout adjustment strategy output model to obtain an interface interaction design that conforms to the user's operating habits; and updates the element layout adjustment strategy output model by means of reinforcement learning to achieve the effect of parameter optimization, further making the element layout adjustment strategy output by the element layout adjustment strategy output model more in line with the user's habits, thereby improving the user experience of user interface interaction; at the same time, in the process of interface interaction design, different element layout modes are introduced through mutation operations, thereby expanding the search space for interface interaction design optimization, and obtaining a better interface interaction experience.

[0004] The present invention provides an interface interaction design method, comprising: At the current interface design optimization time point, an interface state data set and an interaction feedback data set of the display interface are obtained. The interface state data set includes ambient light data and an interface layout data set. The interface layout data set includes M element layout data, where M is the total number of all elements loaded in the display interface. The element layout data includes element name, element display state, element coordinate area, and element display color. The interaction feedback data includes the average interaction time of a preset period before the current interface design optimization time point and the number of misoperations of a preset period before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset period. The average interaction time is the average value of all interaction operation times within the preset period. The number of misoperations is the number of misoperations when users interact within the preset period. Adopting the reinforcement learning method, the output model of the element layout adjustment strategy is updated according to the interface state data set and interaction feedback data set obtained at the current interface design optimization time point; The interface state data set obtained at the current interface design optimization time point is sent to the updated element layout adjustment strategy output model for processing, and an element layout adjustment strategy set is output. The element layout adjustment strategy set includes M element layout adjustment strategies, and the element layout adjustment strategies include element addition and deletion operations, element movement operations, element scaling operations, and element color adjustment operations. Based on the element layout adjustment strategy set, the interface layout data set is adjusted, a new interface layout data set is output, and a mutation operation is performed on the new interface layout data set. The mutation operation is used to introduce a new element layout, and the display interface is set through the interface layout data set after the mutation operation is performed, so as to realize the optimized design of interface interaction.

[0005] As a preferred aspect, a reinforcement learning method is used to update the element layout adjustment strategy output model according to the interface state data set and the interaction feedback data set obtained at the current interface design optimization time point, which specifically includes the following steps: The interface state data set obtained at the current interface design optimization time point is recorded as the following state set, and the interactive feedback data set obtained at the current interface design optimization time point is recorded as the following result set; the interface state data set obtained at the previous interface design optimization time point of the current interface design optimization time point is recorded as the aforementioned state set, and the element layout adjustment strategy set output at the previous interface design optimization time point of the current interface design optimization time point is recorded as the aforementioned action set; the aforementioned state set, the aforementioned action set, the following state set and the following result set are sequentially composed of real-time training samples, and the real-time training samples are stored in a real-time training sample library in chronological order; Obtain N real-time training samples from the current interface design optimization time point forward, select the nth real-time training sample in chronological order, n=1, 2, 3, ..., N, n is the sequence number of the real-time training sample, splice the aforementioned state set and the aforementioned action set in the selected nth real-time training sample and send them to the action evaluation network model for processing, and output the real-time action evaluation value QA n ; Send the selected n-th real-time training sample's subsequent state set to the target element layout adjustment strategy output model for processing, output the target simulation element layout adjustment strategy, then combine the selected n-th real-time training sample's subsequent state set and the target simulation element layout adjustment strategy to form the target layout adjustment effect analysis data, and send the target layout adjustment effect analysis data to the action evaluation network model for processing, and output the predicted action evaluation value B n , based on the real-time training samples, calculate the reward value R corresponding to the real-time training samples n , calculate the first target action evaluation value QB n =R n +γB n, where γ is the discount factor; until all real-time training samples are selected, the first target loss value of the action evaluation network model is calculated, and the parameters of the action evaluation network model are updated in real time using the gradient descent method by minimizing the first target loss value U of the action evaluation network model; The selected Nth real-time training sample contains the following state set, which is sent to the element layout adjustment strategy output model for processing, and the simulated element layout adjustment strategy is output. The following state set and the simulated element layout adjustment strategy in the selected Nth real-time training sample are then combined into layout adjustment effect analysis data, and the layout adjustment effect analysis data is sent to the updated action evaluation network model for processing, and the second target action evaluation value QC is output. n ; The second target action evaluation value QC n As the second target loss value Y of the element layout adjustment strategy output model, by maximizing the second target loss value Y of the element layout adjustment strategy output model, the parameters of the element layout adjustment strategy output model are updated using the gradient ascent method; It also includes updates to the target element layout adjustment strategy output model and the target action evaluation network model. The initial target element layout adjustment strategy output model is consistent with the element layout adjustment strategy output model, and the target action evaluation network model is consistent with the action evaluation network model.

[0006] As a preferred aspect, the reward value R corresponding to the real-time training sample is calculated based on the real-time training sample. n , specifically including the following steps: The average interaction time and number of erroneous operations corresponding to the following result set in the real-time training sample are recorded as T and C respectively, and the reward value R corresponding to the real-time training sample is calculated by the following formula n =α1T -1 +α2C, where α1 and α2 are the first weight coefficient and the second weight coefficient corresponding to the average interaction time and the number of incorrect operations respectively.

[0007] As a preferred aspect, the updating of the target element layout adjustment strategy output model and the target action evaluation network model specifically includes the following steps: At the target model replacement time point, the element layout adjustment strategy output model is directly used as the target element layout adjustment strategy output model to replace the previous target element layout adjustment strategy output model; the action evaluation network model is directly used as the target action evaluation network model to replace the previous target action evaluation network model; and the time points between two adjacent target model replacement time points include K interface design optimization time points.

[0008] As a preferred aspect, training is performed for the element layout adjustment strategy output model and the action evaluation network model, specifically including the following steps: Acquire a number of element layout adjustment strategy training samples, the element layout adjustment strategy training samples include an interface state data set and a target element layout adjustment strategy, compose an element layout adjustment strategy training set from all element layout adjustment strategy training samples, and send the element layout adjustment strategy training set to a parameter-initialized element layout adjustment strategy output model for training, calculate an element layout adjustment strategy loss value, and determine whether the element layout adjustment strategy loss value is within a first preset range. If the element layout adjustment strategy loss value is within the first preset range, output the trained element layout adjustment strategy output model; otherwise, continue to train the element layout adjustment strategy output model using the element layout adjustment strategy training set; Acquire several action evaluation network training samples, which include interface state data sets and element layout adjustment strategies, label the action evaluation network training samples by target evaluation values, form all labeled action evaluation network training samples into an action evaluation network training set, and send the action evaluation network training set to a parameter-initialized action evaluation network output model for training, calculate the action evaluation network loss value, and determine whether the action evaluation network loss value is within a second preset range. If the action evaluation network loss value is within the second preset range, output the trained action evaluation network output model; otherwise, continue to train the action evaluation network output model by using the action evaluation network training set.

[0009] As a preferred aspect, performing a mutation operation on the new interface layout dataset specifically includes the following steps: The interactive feedback dataset at the current interface design optimization time point and the interactive feedback dataset at the previous interface design optimization time point are differentially operated to construct an interactive feedback trend analysis dataset. The average interaction time difference and the error operation number difference in the interactive feedback trend analysis dataset are recorded as L1 and L2, respectively. The comprehensive change value μ=w1L1+w2L2 is calculated, where w1 and w2 are the first adjustment coefficient and the second adjustment coefficient, respectively. The mutation threshold P is calculated. c = (1 + μ) -1 , randomly generate a random number ζ between 0 and 1, and determine ζ>P c Is it true? If ζ>P c If ζ>P c Not true, no mutation operation is performed.

[0010] As a preferred aspect, constructing a variant selection data set specifically includes the following steps: For the interface state dataset obtained at the current interface design optimization time point, several simulated mutation selection datasets are randomly designed, all simulated mutation selection datasets are traversed, and mutation operation is performed on the interface state dataset obtained at the current interface design optimization time point based on the selected simulated mutation selection dataset to construct a simulated interface state dataset, and the interface state dataset and the simulated interface state dataset are respectively sent to the reward value output model for processing, and the corresponding reward value and simulated reward value are output, and the size of the reward value and the simulated reward value is judged. If the simulated reward value is greater than the reward value, the selected simulated mutation selection dataset is stored in the candidate library. If the simulated reward value is not greater than the reward value, no operation is performed; until all simulated mutation selection datasets are traversed, a simulated mutation selection dataset is randomly selected from the candidate library as the mutation selection dataset.

[0011] As a preferred aspect, training is performed on the reward value output model, specifically including the following steps: A plurality of reward value output training samples are obtained, wherein the reward value output training samples include an interface state data set, and the interface state data set is labeled by the corresponding reward value, and all labeled reward value output training samples are combined into a reward value output training set, and the reward value output training set is sent to a parameter-initialized reward value output model for training, and a reward value output loss value is calculated to determine whether the reward value output loss value is within a third preset range. If the reward value output loss value is within the third preset range, the trained reward value output model is output; otherwise, the reward value output model is continuously trained by the reward value output training set.

[0012] The present invention also provides an interface interaction design system, comprising: A data acquisition module is used to obtain an interface state data set and an interaction feedback data set of the display interface at the current interface design optimization time point. The interface state data set includes ambient light data and an interface layout data set. The interface layout data set includes M element layout data, where M is the total number of all elements loaded in the display interface. The element layout data includes element name, element display state, element coordinate area, and element display color. The interaction feedback data includes the average interaction time of a preset period before the current interface design optimization time point and the number of misoperations of a preset period before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset period. The average interaction time is the average value of all interaction operation times within the preset period. The number of misoperations is the number of misoperations when users interact within the preset period. The element layout adjustment module is used to update the element layout adjustment strategy output model based on the interface state data set and interaction feedback data set obtained at the current interface design optimization time point by using reinforcement learning; The element layout adjustment strategy output model update module is used to send the interface status data set obtained at the current interface design optimization time point into the updated element layout adjustment strategy output model for processing, and output the element layout adjustment strategy set. The element layout adjustment strategy set includes M element layout adjustment strategies. The element layout adjustment strategy includes element addition and deletion operations, element movement operations, element scaling operations and element color adjustment operations; based on the element layout adjustment strategy set, the interface layout data set is adjusted, a new interface layout data set is output, and a mutation operation is performed on the new interface layout data set. The mutation operation is used to introduce a new element layout, and the display interface is set through the interface layout data set after the mutation operation is performed, so as to achieve the optimization design of the interface interaction.

[0013] The present invention has the following advantages: The present invention constructs the design of interface interaction as the layout design of elements, obtains an interface state data set, collects effect feedback corresponding to the interface interaction, i.e., an interaction feedback data set, and then outputs the element layout adjustment strategy through an element layout adjustment strategy output model to obtain an interface interaction design that conforms to the user's operating habits; and updates the element layout adjustment strategy output model by means of reinforcement learning to achieve the effect of parameter optimization, further making the element layout adjustment strategy output by the element layout adjustment strategy output model more in line with the user's habits, thereby improving the user experience of user interface interaction; at the same time, in the process of interface interaction design, different element layout modes are introduced through mutation operations, thereby expanding the search space for interface interaction design optimization, and obtaining a better interface interaction experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a structural schematic diagram of the interface interaction design system adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] Embodiment 1, an interface interaction design method, comprising: At the current interface design optimization time point, the interface state dataset and interaction feedback dataset of the display interface are obtained. The display interface here can be a mobile phone display screen and a car display screen, etc. The interface state dataset includes ambient light data and interface layout dataset, wherein the ambient light data is obtained through the lighting component on the display screen corresponding to the interface, and what is obtained is the brightness value corresponding to the current ambient light. The interface layout dataset includes M element layout data, where M is the total number of all elements loaded in the display interface. It should be noted that the element here refers to various icons that can be presented on the display interface for realizing human-computer interaction, and the M value represents the total number of elements, which does not mean that all M elements must be displayed in the display interface, but that some elements will be selected from the M elements to be displayed in the display interface. The element layout data includes element name, element display state, element coordinate area and element display color, wherein the element name can be the APP software name and the editing status bar, etc., which are generally stored using one-hot encoding; element display state representation Whether the corresponding element is displayed on the display interface, generally including display and non-display, generally stored in Boolean values ​​1 and 0 respectively; the element coordinate area represents the position of the corresponding element on the display interface, generally stored in the form of center coordinates, width and height of a rectangular frame coordinate; the element display color represents the color of the corresponding element display, generally stored in RGB color code; the interactive feedback data includes the average interaction time of the preset cycle before the current interface design optimization time point and the number of misoperations in the preset cycle before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset cycle. The average interaction time is the average value of all interaction operation times within the preset cycle, which is used to describe whether the element layout on the display interface is reasonable and effective. The number of misoperations is the number of misoperations when users interact within the preset cycle, such as APP click errors, accidental file deletion, etc.; the average interaction time and the number of misoperations can characterize the pros and cons of the current interface interaction design and provide a reference for subsequent element layout design; Adopting the reinforcement learning method, the element layout adjustment strategy output model is updated according to the interface status data set and interaction feedback data set obtained at the current interface design optimization time point, so that the element layout adjustment strategy set output by the element layout adjustment strategy output model can better meet the actual needs of users with the help of average interaction time and number of misoperations, thus improving the user experience of interface interaction. The interface state data set obtained at the current interface design optimization time point is sent to the updated element layout adjustment strategy output model for processing, and the element layout adjustment strategy set is output. The element layout adjustment strategy set includes M element layout adjustment strategies. The element layout adjustment strategies include element addition and deletion operations, element movement operations, element scaling operations, and element color adjustment operations. The element addition and deletion operations are adjustments to the element display state, generally storing +1 or -1; the element movement operations and element scaling operations are adjustments to the element coordinate area. The element movement operations are generally addition and subtraction operations on the center coordinates, and the element scaling operations are generally multiplication operations on the width and height; the element color adjustment operation is an adjustment to the element display state. Adjust the display color; adjust the interface layout data set based on the element layout adjustment strategy set, output a new interface layout data set, and perform a mutation operation on the new interface layout data set. The mutation operation is used to introduce a new element layout, and the display interface is set through the interface layout data set after the mutation operation, so as to achieve the optimization design of the interface interaction; since the process of updating the element layout adjustment strategy output model is essentially a process of learning user habits by searching for element layout adjustment strategies, in order to avoid the search falling into a local optimal solution, different element layout methods can be introduced by performing a mutation operation on the interface layout data set, thereby expanding the search space for interface interaction design optimization; Using reinforcement learning, the element layout adjustment strategy output model is updated according to the interface state dataset and interaction feedback dataset obtained at the current interface design optimization time point, which specifically includes the following steps: The interface state data set obtained at the current interface design optimization time point is recorded as the subsequent state set, and the interaction feedback data set obtained at the current interface design optimization time point is recorded as the subsequent result set; the interface state data set obtained at the previous interface design optimization time point of the current interface design optimization time point is recorded as the aforementioned state set, and the element layout adjustment strategy set output at the previous interface design optimization time point of the current interface design optimization time point is recorded as the aforementioned action set; the aforementioned state set, the aforementioned action set, the subsequent state set and the subsequent result set are sequentially composed of real-time training samples, and the real-time training samples are stored in the real-time training sample library in chronological order. It should be noted that each time the design optimization of the interface interaction is performed, a real-time training sample will be output. These real-time training samples describe the user experience results after executing the element layout adjustment strategy set, and are used to provide a reference for updating the element layout adjustment strategy output model; Obtain N real-time training samples from the current interface design optimization time point forward, select the nth real-time training sample in chronological order, n=1, 2, 3, ..., N, n is the sequence number of the real-time training sample, splice the aforementioned state set and the aforementioned action set in the selected nth real-time training sample and send them to the action evaluation network model for processing, and output the real-time action evaluation value QA n, the real-time action evaluation value can reflect the effect of the element layout adjustment strategy adopted at the previous interface design optimization time point, whether it is better or worse in the change trend; the subsequent state set in the selected n-th real-time training sample is sent to the target element layout adjustment strategy output model for processing, and the target simulation element layout adjustment strategy is output. The target simulation element layout adjustment strategy is the element layout adjustment strategy that will be selected based on the interface state data set obtained at the current interface design optimization time point. Then, the subsequent state set in the selected n-th real-time training sample and the target simulation element layout adjustment strategy are combined to form the target layout adjustment effect analysis data, and the target layout adjustment effect analysis data is sent to the action evaluation network model for processing, and the predicted action evaluation value B is output. n The predicted action evaluation value reflects the effect that will be achieved by adopting the predicted element layout adjustment strategy, and can reflect the changing trend of the element layout adjustment strategy effect. Considering the changing trend of the element layout adjustment strategy, it can further meet the user's needs for interface interaction. Based on the real-time training samples, the reward value R corresponding to the real-time training sample is calculated. n , calculate the first target action evaluation value QB n =R n +γB n , where γ is a discount factor, which is used to characterize the emphasis on reward value and predicted action evaluation value, and is determined by the developer; until all real-time training samples are selected, the first target loss value of the action evaluation network model is calculated. The first target loss value calculated with the reward value as the target can reflect the gap between the effect of the action evaluation network model evaluation and the actual user experience. By minimizing the first target loss value U of the action evaluation network model, the parameters of the action evaluation network model are updated in real time using the gradient descent method, and the gradient descent method can use the Adam optimizer; it should be noted that the update of the action evaluation network model here specifies a direction to achieve a higher reward value for the subsequent update of the element layout adjustment strategy output model; The subsequent state set in the selected Nth real-time training sample is sent to the element layout adjustment strategy output model for processing, and the simulated element layout adjustment strategy is output. It should be noted that the previous target simulated element layout adjustment strategy is output by the target element layout adjustment strategy output model, which is different from the simulated element layout adjustment strategy here. Then, the subsequent state set and the simulated element layout adjustment strategy in the selected Nth real-time training sample are combined into layout adjustment effect analysis data, and the layout adjustment effect analysis data is sent to the updated action evaluation network model for processing, and the second target action evaluation value QC is output. n ; The second target action evaluation value QC nAs the second target loss value Y of the element layout adjustment strategy output model, the parameters of the element layout adjustment strategy output model are updated using the gradient ascent method by maximizing the second target loss value Y of the element layout adjustment strategy output model; it should be noted that the optimization operation of the parameters in the element layout adjustment strategy output model in a manner of maximizing the second target loss value Y of the element layout adjustment strategy output model can provide users with a better interface interaction experience in the process of updating the element layout adjustment strategy output model; It also includes updates to the target element layout adjustment strategy output model and the target action evaluation network model. The initial target element layout adjustment strategy output model is consistent with the element layout adjustment strategy output model, and the target action evaluation network model is consistent with the action evaluation network model.

[0017] The present application constructs the design of interface interaction as the layout design of elements to obtain an interface state data set, and collects effect feedback corresponding to the interface interaction, namely, the interaction feedback data set, and then outputs the element layout adjustment strategy through the element layout adjustment strategy output model to obtain an interface interaction design that conforms to the user's operating habits; and updates the element layout adjustment strategy output model by means of reinforcement learning to achieve the effect of parameter optimization, further making the element layout adjustment strategy output by the element layout adjustment strategy output model more in line with user habits, thereby improving the user experience of user interface interaction; at the same time, in the process of interface interaction design, different element layout methods are introduced through mutation operations to expand the search space for interface interaction design optimization, thereby obtaining a better interface interaction experience.

[0018] Calculate the reward value R corresponding to the real-time training sample based on the real-time training sample n , specifically including the following steps: The average interaction time and number of erroneous operations corresponding to the following result set in the real-time training sample are recorded as T and C respectively, and the reward value R corresponding to the real-time training sample is calculated by the following formula n =α1T -1 +α2C, where α1 and α2 are the first weight coefficient and the second weight coefficient corresponding to the average interaction time and the number of erroneous operations respectively. It should be noted that the first weight coefficient and the second weight coefficient respectively characterize the emphasis corresponding to the average interaction time and the number of erroneous operations, which are determined according to the purpose of the display interface. Generally, the developer sets them directly. In order to make the output model of the element layout adjustment strategy better, the first weight coefficient and the second weight coefficient can also be set through the group optimization algorithm.

[0019] The update of the target element layout adjustment strategy output model and the target action evaluation network model specifically includes the following steps: At the target model replacement time point, the element layout adjustment strategy output model is directly used as the target element layout adjustment strategy output model to replace the previous target element layout adjustment strategy output model; the action evaluation network model is directly used as the target action evaluation network model to replace the previous target action evaluation network model; and the time points between two adjacent target model replacement time points include K interface design optimization time points. The K value is determined by the developer. Generally, when the model is updated frequently in the early stage, the K value is smaller, and when the model is updated slowly in the later stage, the K value is larger.

[0020] The training of the element layout adjustment strategy output model and the action evaluation network model includes the following steps: Obtain several element layout adjustment strategy training samples, the element layout adjustment strategy training samples include an interface state data set and a target element layout adjustment strategy. It should be noted that the target element layout adjustment strategy is an adjustment plan provided by experts to the corresponding user based on the interface interaction design corresponding to the interface state data set, and it has been highly evaluated by the user. All element layout adjustment strategy training samples form an element layout adjustment strategy training set, and the element layout adjustment strategy training set is sent to the parameter-initialized element layout adjustment strategy output model for training, the element layout adjustment strategy loss value is calculated, and it is determined whether the element layout adjustment strategy loss value is within a first preset range, which is artificially set. If the element layout adjustment strategy loss value is within the first preset range, the trained element layout adjustment strategy output model is output; otherwise, the element layout adjustment strategy output model is continuously trained through the element layout adjustment strategy training set; Obtain several action evaluation network training samples, the action evaluation network training samples include an interface state data set and an element layout adjustment strategy. It should be noted that the interface state data set and the element layout adjustment strategy are collected during the actual interface design operation process. The action evaluation network training samples are labeled by target evaluation values. The target evaluation value is generally 0 or 1. 0 indicates that the element layout adjustment strategy is poorly effective in the user's opinion, while 1 indicates that the element layout adjustment strategy is very effective in the user's opinion. All labeled action evaluation network training samples form an action evaluation network training set, and the action evaluation network training set is sent to the parameter-initialized action evaluation network output model for training. The action evaluation network loss value is calculated to determine whether the action evaluation network loss value is within a second preset range. The second preset range is artificially set. If the action evaluation network loss value is within the second preset range, the trained action evaluation network output model is output; otherwise, the action evaluation network output model continues to be trained through the action evaluation network training set; Perform mutation operation on the new interface layout dataset, including the following steps: Perform a differential operation on the interaction feedback dataset at the current interface design optimization time point and the interaction feedback dataset at the previous interface design optimization time point to construct an interaction feedback trend analysis dataset, and record the average interaction time difference and the error operation number difference in the interaction feedback trend analysis dataset as L1 and L2 respectively, and calculate the comprehensive change value μ=w1L1+w2L2, where w1 and w2 are the first adjustment coefficient and the second adjustment coefficient respectively, which are used to characterize the change trend proportion of the average interaction time difference and the error operation number difference, which is determined by the developer, and calculate the mutation threshold P c = (1 + μ) -1 , randomly generate a random number ζ between 0 and 1, and determine ζ>P c Is it true? If ζ>P c If the mutation selection data set is established, the new interface layout data set is mutated according to the mutation selection data set. The mutation selection data set includes two element names. The mutation operation is to swap the two element names in the interface layout data set. The specific operation is to replace the two element names corresponding to the new interface layout data set with those in the mutation selection data set to introduce a new element layout. If ζ>P c Not true, no mutation operation is performed; Constructing a mutation selection dataset includes the following steps: For the interface state data set obtained at the current interface design optimization time point, several simulated mutation selection data sets are randomly designed. It should be noted that the simulated mutation selection data set stores two randomly selected element names, and the two element names are different. All simulated mutation selection data sets are traversed, and mutation operations are performed on the interface state data set obtained at the current interface design optimization time point based on the selected simulated mutation selection data set to construct a simulated interface state data set. The interface state data set and the simulated interface state data set are respectively sent to the reward value output model for processing, and the corresponding reward value and simulated reward value are output. The size of the reward value and the simulated reward value is judged. If the simulated reward value is greater than the reward value, the selected simulated mutation selection data set is stored in the candidate library. If the simulated reward value is not greater than the reward value, no operation is performed. Until all simulated mutation selection data sets are traversed, a simulated mutation selection data set is randomly selected from the candidate library as the mutation selection data set.

[0021] Training the reward value output model includes the following steps: A number of reward value output training samples are obtained, the reward value output training samples include an interface state data set, and the interface state data set is annotated by the corresponding reward value. It should be noted that the interface state data set and the reward value are obtained at the same interface design optimization time point. All annotated reward value output training samples form a reward value output training set, and the reward value output training set is sent to the reward value output model with parameter initialization for training, and the reward value output loss value is calculated to determine whether the reward value output loss value is within a third preset range. The third preset range is set manually. If the reward value output loss value is within the third preset range, the trained reward value output model is output; otherwise, the reward value output model continues to be trained through the reward value output training set.

[0022] Embodiment 2, an interface interaction design system, such as Figure 1 As shown, including: The data acquisition module is used to obtain the interface status data set and interactive feedback data set of the display interface at the current interface design optimization time point. The display interface here can be a mobile phone display screen and a car display screen, etc. The interface status data set includes ambient light data and interface layout data set, wherein the ambient light data is obtained through the lighting component on the display screen corresponding to the interface, and the brightness value corresponding to the current ambient light is obtained. The interface layout data set includes M element layout data, where M is the total number of all elements loaded in the display interface. It should be noted that the element here refers to various icons that can be presented on the display interface for realizing human-computer interaction, and the M value represents the total number of elements, which does not mean that all M elements must be displayed in the display interface, but that some elements will be selected from the M elements to be displayed in the display interface. The element layout data includes element name, element display status, element coordinate area and element display color, wherein the element name can be the APP software name and the editing status bar, etc., which are generally stored using one-hot encoding; element display The state represents whether the corresponding element is displayed on the display interface, generally including display and non-display, and is generally stored in Boolean values ​​1 and 0 respectively; the element coordinate area represents the position of the corresponding element on the display interface, and is generally stored in the form of center coordinates, width and height rectangular frame coordinates; the element display color represents the color of the corresponding element display, and is generally stored in RGB color code; the interactive feedback data includes the average interaction time of a preset cycle before the current interface design optimization time point and the number of misoperations in a preset cycle before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset cycle. The average interaction time is the average value of all interaction operation times within the preset cycle, which is used to describe whether the element layout on the display interface is reasonable and effective. The number of misoperations is the number of misoperations when users interact within the preset cycle, such as APP click errors, accidental file deletion, etc.; the average interaction time and the number of misoperations can be used to characterize the pros and cons of the current interface interaction design and provide a reference for subsequent element layout design; The element layout adjustment module is used to send the interface status data set obtained at the current interface design optimization time point into the updated element layout adjustment strategy output model for processing, and output the element layout adjustment strategy set. The element layout adjustment strategy set includes M element layout adjustment strategies. The element layout adjustment strategy includes element addition and deletion operations, element movement operations, element scaling operations and element color adjustment operations. Among them, the element addition and deletion operations are adjustments to the element display status, generally storing +1 or -1; the element movement operations and element scaling operations are adjustments to the element coordinate area. The element movement operations are generally addition and subtraction operations on the center coordinates, and the element scaling operations are generally multiplication operations on the width and height; the element color adjustment operation is the adjustment of the element display color. ; Based on the element layout adjustment strategy set, the interface layout data set is adjusted, a new interface layout data set is output, and a mutation operation is performed on the new interface layout data set. The mutation operation is to swap the names of two elements in the interface layout data set, and the display interface is set through the interface layout data set after the mutation operation is performed to achieve the optimization design of the interface interaction; Since the process of updating the element layout adjustment strategy output model is essentially a process of learning user habits by searching for element layout adjustment strategies, in order to avoid the search falling into a local optimal solution, different element layout methods can be introduced by performing a mutation operation on the interface layout data set, thereby expanding the search space for interface interaction design optimization and obtaining a better interface interaction experience; The element layout adjustment strategy output model update module is used to update the element layout adjustment strategy output model by using reinforcement learning, based on the interface status dataset and interaction feedback dataset obtained at the current interface design optimization time point, so that the element layout adjustment strategy set output by the element layout adjustment strategy output model can better meet the user's actual usage needs with the assistance of the average interaction time and the number of incorrect operations, thereby improving the user's experience of interface interaction.

[0023] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. An interface interaction design method, characterized in that: include: At the current interface design optimization time point, an interface state data set and an interaction feedback data set of the display interface are obtained. The interface state data set includes ambient light data and an interface layout data set. The interface layout data set includes M element layout data, where M is the total number of all elements loaded in the display interface. The element layout data includes element name, element display state, element coordinate area, and element display color. The interaction feedback data includes the average interaction time of a preset period before the current interface design optimization time point and the number of misoperations of a preset period before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset period. The average interaction time is the average value of all interaction operation times within the preset period. The number of misoperations is the number of misoperations when users interact within the preset period. Adopting the reinforcement learning method, the output model of the element layout adjustment strategy is updated according to the interface state data set and interaction feedback data set obtained at the current interface design optimization time point; The interface state data set obtained at the current interface design optimization time point is sent to the updated element layout adjustment strategy output model for processing, and an element layout adjustment strategy set is output. The element layout adjustment strategy set includes M element layout adjustment strategies, and the element layout adjustment strategies include element addition and deletion operations, element movement operations, element scaling operations, and element color adjustment operations. Based on the element layout adjustment strategy set, the interface layout data set is adjusted, a new interface layout data set is output, and a mutation operation is performed on the new interface layout data set. The mutation operation is used to introduce a new element layout, and the display interface is set through the interface layout data set after the mutation operation is performed, so as to realize the optimized design of interface interaction.

2. The interface interaction design method according to claim 1, characterized in that: Using reinforcement learning, the element layout adjustment strategy output model is updated according to the interface state dataset and interaction feedback dataset obtained at the current interface design optimization time point, which specifically includes the following steps: The interface state data set obtained at the current interface design optimization time point is recorded as the state set described below, and the interaction feedback data set obtained at the current interface design optimization time point is recorded as the result set described below; Recording the interface state data set obtained at the previous interface design optimization time point of the current interface design optimization time point as the aforementioned state set, and recording the element layout adjustment strategy set output at the previous interface design optimization time point of the current interface design optimization time point as the aforementioned action set; the aforementioned state set, the aforementioned action set, the following state set, and the following result set are sequentially composed of real-time training samples, and the real-time training samples are stored in a real-time training sample library in chronological order; Obtain N real-time training samples from the current interface design optimization time point forward, select the nth real-time training sample in chronological order, n=1, 2, 3, ..., N, n is the sequence number of the real-time training sample, splice the aforementioned state set and the aforementioned action set in the selected nth real-time training sample and send them to the action evaluation network model for processing, and output the real-time action evaluation value QA n ; Send the selected n-th real-time training sample's subsequent state set to the target element layout adjustment strategy output model for processing, output the target simulation element layout adjustment strategy, then combine the selected n-th real-time training sample's subsequent state set and the target simulation element layout adjustment strategy to form the target layout adjustment effect analysis data, and send the target layout adjustment effect analysis data to the action evaluation network model for processing, and output the predicted action evaluation value B n , based on the real-time training samples, calculate the reward value R corresponding to the real-time training samples n , calculate the first target action evaluation value QB n =R n +γB n , where γ is the discount factor; until all real-time training samples are selected, the first target loss value of the action evaluation network model is calculated, and the parameters of the action evaluation network model are updated in real time using the gradient descent method by minimizing the first target loss value U of the action evaluation network model; The selected Nth real-time training sample contains the following state set, which is sent to the element layout adjustment strategy output model for processing, and the simulated element layout adjustment strategy is output. The following state set and the simulated element layout adjustment strategy in the selected Nth real-time training sample are then combined into layout adjustment effect analysis data, and the layout adjustment effect analysis data is sent to the updated action evaluation network model for processing, and the second target action evaluation value QC is output. n ; The second target action evaluation value QC n As the second target loss value Y of the element layout adjustment strategy output model, by maximizing the second target loss value Y of the element layout adjustment strategy output model, the parameters of the element layout adjustment strategy output model are updated using the gradient ascent method; It also includes updates to the target element layout adjustment strategy output model and the target action evaluation network model. The initial target element layout adjustment strategy output model is consistent with the element layout adjustment strategy output model, and the target action evaluation network model is consistent with the action evaluation network model.

3. The interface interaction design method according to claim 2, characterized in that: Calculate the reward value R corresponding to the real-time training sample based on the real-time training sample n , specifically including the following steps: The average interaction time and number of erroneous operations corresponding to the following result set in the real-time training sample are recorded as T and C respectively, and the reward value R corresponding to the real-time training sample is calculated by the following formula n =α1T -1 +α2C, where α1 and α2 are the first weight coefficient and the second weight coefficient corresponding to the average interaction time and the number of incorrect operations respectively.

4. The interface interaction design method according to claim 3, characterized in that: The update of the target element layout adjustment strategy output model and the target action evaluation network model specifically includes the following steps: At the target model replacement time point, the element layout adjustment strategy output model is directly used as the target element layout adjustment strategy output model to replace the previous target element layout adjustment strategy output model; the action evaluation network model is directly used as the target action evaluation network model to replace the previous target action evaluation network model; and the time points between two adjacent target model replacement time points include K interface design optimization time points.

5. The interface interaction design method according to claim 4, characterized in that: The training of the element layout adjustment strategy output model and the action evaluation network model includes the following steps: Acquire a number of element layout adjustment strategy training samples, the element layout adjustment strategy training samples include an interface state data set and a target element layout adjustment strategy, compose an element layout adjustment strategy training set from all element layout adjustment strategy training samples, and send the element layout adjustment strategy training set to a parameter-initialized element layout adjustment strategy output model for training, calculate an element layout adjustment strategy loss value, and determine whether the element layout adjustment strategy loss value is within a first preset range. If the element layout adjustment strategy loss value is within the first preset range, output the trained element layout adjustment strategy output model; otherwise, continue to train the element layout adjustment strategy output model using the element layout adjustment strategy training set; Acquire several action evaluation network training samples, which include interface state data sets and element layout adjustment strategies, label the action evaluation network training samples by target evaluation values, form all labeled action evaluation network training samples into an action evaluation network training set, and send the action evaluation network training set to a parameter-initialized action evaluation network output model for training, calculate the action evaluation network loss value, and determine whether the action evaluation network loss value is within a second preset range. If the action evaluation network loss value is within the second preset range, output the trained action evaluation network output model; otherwise, continue to train the action evaluation network output model by using the action evaluation network training set.

6. The interface interaction design method according to claim 5, characterized in that: Perform mutation operation on the new interface layout dataset, including the following steps: The interactive feedback dataset at the current interface design optimization time point and the interactive feedback dataset at the previous interface design optimization time point are differentially operated to construct an interactive feedback trend analysis dataset. The average interaction time difference and the error operation number difference in the interactive feedback trend analysis dataset are recorded as L1 and L2, respectively. The comprehensive change value μ=w1L1+w2L2 is calculated, where w1 and w2 are the first adjustment coefficient and the second adjustment coefficient, respectively. The mutation threshold P is calculated. c = (1 + μ) -1 , randomly generate a random number ζ between 0 and 1, and determine ζ>P c Is it true? If ζ>P c Established, the new interface layout dataset is mutated according to the mutation selection dataset, the mutation selection dataset includes two element names, and the mutation operation is to swap the two element names in the interface layout dataset to introduce a new element layout; if ζ>P c Not true, no mutation operation is performed.

7. The interface interaction design method according to claim 6, characterized in that: Constructing a mutation selection dataset includes the following steps: For the interface state dataset obtained at the current interface design optimization time point, several simulated mutation selection datasets are randomly designed, all simulated mutation selection datasets are traversed, and mutation operation is performed on the interface state dataset obtained at the current interface design optimization time point based on the selected simulated mutation selection dataset to construct a simulated interface state dataset, and the interface state dataset and the simulated interface state dataset are respectively sent to the reward value output model for processing, and the corresponding reward value and simulated reward value are output, and the size of the reward value and the simulated reward value is judged. If the simulated reward value is greater than the reward value, the selected simulated mutation selection dataset is stored in the candidate library. If the simulated reward value is not greater than the reward value, no operation is performed; until all simulated mutation selection datasets are traversed, a simulated mutation selection dataset is randomly selected from the candidate library as the mutation selection dataset.

8. The interface interaction design method according to claim 7, characterized in that: Training the reward value output model includes the following steps: A plurality of reward value output training samples are obtained, wherein the reward value output training samples include an interface state data set, and the interface state data set is labeled by the corresponding reward value, and all labeled reward value output training samples are combined into a reward value output training set, and the reward value output training set is sent to a parameter-initialized reward value output model for training, and a reward value output loss value is calculated to determine whether the reward value output loss value is within a third preset range. If the reward value output loss value is within the third preset range, the trained reward value output model is output; otherwise, the reward value output model is continuously trained by the reward value output training set.

9. An interface interaction design system, characterized in that: The system applies an interface interaction design method as described in claims 1 to 7 above, including: A data acquisition module is used to obtain an interface state data set and an interaction feedback data set of the display interface at the current interface design optimization time point. The interface state data set includes ambient light data and an interface layout data set. The interface layout data set includes M element layout data, where M is the total number of all elements loaded in the display interface. The element layout data includes element name, element display state, element coordinate area, and element display color. The interaction feedback data includes the average interaction time of a preset period before the current interface design optimization time point and the number of misoperations of a preset period before the current interface design optimization time point. The time period between two adjacent interface design optimization time points is the preset period. The average interaction time is the average value of all interaction operation times within the preset period. The number of misoperations is the number of misoperations when users interact within the preset period. The element layout adjustment module is used to update the element layout adjustment strategy output model based on the interface state data set and interaction feedback data set obtained at the current interface design optimization time point by using reinforcement learning; The element layout adjustment strategy output model update module is used to send the interface status data set obtained at the current interface design optimization time point into the updated element layout adjustment strategy output model for processing, and output the element layout adjustment strategy set. The element layout adjustment strategy set includes M element layout adjustment strategies. The element layout adjustment strategy includes element addition and deletion operations, element movement operations, element scaling operations and element color adjustment operations; based on the element layout adjustment strategy set, the interface layout data set is adjusted, a new interface layout data set is output, and a mutation operation is performed on the new interface layout data set. The mutation operation is used to introduce a new element layout, and the display interface is set through the interface layout data set after the mutation operation is performed, so as to achieve the optimization design of the interface interaction.

Images