Screen Refresh Rate Adjustment Method and Device, Electronic Device, Storage Medium

Abstract

Embodiments of the present disclosure relate to a method and apparatus for adjusting screen refresh rate, an electronic device, and a storage medium, and relate to the field of computer technologies. The method for adjusting screen refresh rate includes: obtaining application-related data; performing feature extraction on the application-related data to determine the predicted usage probability of a target application included in the application-related data, and determining the predicted interaction frequency of the target application; determining an application scenario based on the target application corresponding to the predicted usage probability, and obtaining an interaction frequency baseline value corresponding to the application scenario; and determining the screen refresh rate of the terminal by combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency. The technical solutions in the embodiments of the present disclosure can improve the effect of adjusting the screen refresh rate.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular, to a method and device for adjusting screen refresh rate, an electronic device, and a computer-readable storage medium. Background Art

[0002] During the use of a terminal, increasing the screen refresh rate can achieve a more smooth and coherent screen display effect.

[0003] In related technologies, the refresh rate can be adjusted according to the application scenario, reduced based on the user's active interaction frequency, and monitored according to the current state of the mobile phone to adjust the refresh rate. In the above methods, there is a lag in adjusting the screen refresh rate of the terminal, and it will cause additional power consumption, reducing the user experience.

[0004] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present disclosure is to provide a method and device for adjusting screen refresh rate, an electronic device, and a storage medium, so as to at least to some extent overcome the problem of lag in adjusting the screen refresh rate caused by the limitations and defects of related technologies.

[0006] Other features and advantages of the present disclosure will become apparent through the following detailed description, or be learned in part through the practice of the present disclosure.

[0007] According to a first aspect of the present disclosure, there is provided a method for adjusting screen refresh rate, including: obtaining application-related data; extracting features from the application-related data to determine the predicted usage probability of the target application included in the application-related data, and determining the predicted interaction frequency of the target application; determining an application scenario based on the target application corresponding to the predicted usage probability, and obtaining an interaction frequency baseline value corresponding to the application scenario; and determining the screen refresh rate of the terminal by combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency.

[0008] According to a second aspect of the present disclosure, there is provided a screen refresh rate adjustment device, including: a data acquisition module configured to acquire application-related data; an application prediction module configured to perform feature extraction on the application-related data, determine a predicted usage probability of a target application included in the application-related data, and determine a predicted interaction frequency of the target application; a baseline value determination module configured to determine an application scenario based on the target application corresponding to the predicted usage probability and acquire an interaction frequency baseline value corresponding to the application scenario; and a refresh rate adjustment module configured to determine a screen refresh rate of a terminal by combining a screen refresh rate reference value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency.

[0009] According to a third aspect of the present disclosure, there is provided an electronic device, including: a processor; and a memory configured to store executable instructions of the processor; wherein the processor is configured to execute the screen refresh rate adjustment method of the first aspect and its possible implementation manners via executing the executable instructions.

[0010] According to a fourth aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the screen refresh rate adjustment method of the first aspect and its possible implementation manners are implemented.

[0011] In the technical solutions provided in the embodiments of the present disclosure, on the one hand, by performing feature processing on application-related data, predicting the predicted usage probability of a target application to be clicked by a user and the predicted interaction frequency for the target application, and performing pre-dynamic adjustment of the refresh rate based on the predicted interaction frequency, the screen refresh rate reference value, and the interaction frequency baseline value, the hysteresis in adjusting the screen refresh rate in the related art is avoided, the screen refresh rate can be adjusted in a timely and accurate manner, and the accuracy and reliability are improved. On the other hand, by performing pre-dynamic adjustment of the refresh rate based on the predicted interaction frequency, the screen refresh rate reference value, and the interaction frequency baseline value, the problems caused by processing after the user perceives are avoided, the screen refresh rate can be adjusted without the user's perception, and the user experience is improved. On the further hand, since the screen refresh rate can be adjusted based on the predicted interaction frequency of the target application determined according to the user's application-related data, the adjusted screen refresh rate is more in line with the actual situation of the user, the pertinence can be improved, the matching degree between the screen refresh rate and the user is higher, personalized adjustment of the screen refresh rate is realized, and the power consumption of the terminal is also reduced.

[0012] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. Description of the Drawings

[0013] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0014] Figure 1 A schematic diagram showing an application scenario where the screen refresh rate adjustment method according to the embodiments of the present disclosure can be applied.

[0015] Figure 2 A schematic diagram schematically showing a screen refresh rate adjustment method according to an embodiment of the present disclosure.

[0016] Figure 3 A schematic diagram schematically showing obtaining application-related data in an embodiment of the present disclosure.

[0017] Figure 4 A schematic diagram schematically showing a model structure in an embodiment of the present disclosure.

[0018] Figure 5 A schematic flow diagram schematically showing obtaining a target vector in an embodiment of the present disclosure.

[0019] Figure 6 A schematic flow diagram schematically showing obtaining a sequence vector in an embodiment of the present disclosure.

[0020] Figure 7 A schematic flow diagram schematically showing a specific process of processing through an encoder in an embodiment of the present disclosure.

[0021] Figure 8 A schematic diagram schematically showing the structure of a classification predictor in an embodiment of the present disclosure.

[0022] Figure 9 A schematic flow diagram schematically showing a specific process of performing behavior prediction in an embodiment of the present disclosure.

[0023] Figure 10 A schematic flow diagram schematically showing the process of determining the screen refresh rate in an embodiment of the present disclosure.

[0024] Figure 11 A schematic flow diagram schematically showing the overall process of determining the screen refresh rate of a terminal in an embodiment of the present disclosure.

[0025] Figure 12 A schematic block diagram schematically showing a screen refresh rate adjustment device in an embodiment of the present disclosure.

[0026] Figure 13 A schematic block diagram schematically showing an electronic device in an embodiment of the present disclosure. Detailed implementation manners

[0027] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be employed. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0028] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0029] In the related art, the refresh rate is adjusted according to the application scenario, the refresh rate is reduced based on the active interaction frequency of the user, and the refresh rate is adjusted by monitoring the current state of the mobile phone. In the above methods, the first type of method is adjusted based on the application scenario. The definition setting of the application scenario is relatively simple, and the screen refresh rate is adjusted after identifying the current scenario. For example, four scenarios are divided according to whether YUV and GPU are in use, and then corresponding to four different screen refresh rate adjustment methods. Or only the game scenario is defined, and the current game scenario is identified based on the game instruction. The large scenario is single, and it is also a solution to adjust after switching the scenario. The second type of method adjusts the screen refresh rate according to the user interaction time interval. The refresh rate is reduced only after the user has not operated for a long time, and additional power consumption is generated during this period. The third type of method increases the refresh rate when the screen appears uneven. This way, the user has already noticed, so the user experience is poor.

[0030] To solve the technical problems in the related art, an embodiment of the present disclosure provides a method for adjusting the screen refresh rate, which can be applied to an application scenario of adjusting the screen refresh rate of a terminal during the use of an application on the terminal. Figure 1 The schematic diagram of the system architecture to which the method and device for adjusting the screen refresh rate according to the embodiment of the present disclosure can be applied is shown.

[0031] like Figure 1 As shown, the terminal 101 may be a smart device, such as a smart phone, a computer, a tablet computer, a smart speaker, a smart watch, a vehicle-mounted device, a wearable device, a monitoring device, etc. A smart device can run a variety of applications, and an application refers to an application program, which may be a third-party application program or an application program that comes with the terminal itself.

[0032] In the embodiment of the present disclosure, the user 102 can use the application 103 in the terminal 101. In the process of the terminal using the application in the terminal, the application 103 can be used as the target application 104, and the predicted use probability of the target application is determined based on the acquired user portrait 105, the target application 104, the context environment 106 of the target application, and the historical application sequence 107 and other application-related data, and the predicted interaction frequency of the target application is determined; based on the target application corresponding to the predicted use probability, the application scenario is determined, and the screen refresh rate 108 of the terminal is determined in combination with the screen refresh rate baseline value, the interaction frequency baseline value and the predicted interaction frequency of the application scenario.

[0033] It should be noted that the screen refresh rate adjustment method provided in the embodiment of the present disclosure may be executed by the terminal 101. The screen refresh rate adjustment method may also be set in the terminal.

[0034] Figure 2 The screen refresh rate adjustment method in the embodiment of the present disclosure is schematically shown in FIG. , which specifically includes the following steps:

[0035] Step S210, obtaining application related data;

[0036] Step S220, extracting features from the application-related data, determining a predicted usage probability of a target application included in the application-related data, and determining a predicted interaction frequency of the target application;

[0037] Step S230, determining an application scenario based on the target application corresponding to the predicted usage probability, and obtaining an interaction frequency baseline value corresponding to the application scenario;

[0038] Step S240, determining the screen refresh rate of the terminal in combination with the screen refresh rate reference value of the application scenario, the interaction frequency baseline value and the predicted interaction frequency.

[0039] In the embodiments of the present disclosure, application-related data can be used to represent all data related to an application. The application-related data includes a user profile, a target application, the context environment of the target application, and a historical application sequence. The target application can be the application program being used at the current moment. The context environment can be the hardware conditions when using the target application, for example, it can include, but is not limited to, one or more of network status (such as wifi or traffic data), Bluetooth, and battery information. The historical application sequence can be the application programs used at historical moments within a preset period before the current moment.

[0040] Feature extraction can be performed on the user profile, the target application, the context environment of the target application, and the historical application sequence for prediction to obtain the predicted usage probability of the target application and the predicted interaction frequency of the user with the target application. The predicted usage probability is used to describe the probability value of using the target application at the next moment, and can also be referred to as the predicted click-through rate of the target application.

[0041] Furthermore, the application scenario of the target application corresponding to the predicted usage probability can be determined, and the reference value of the screen refresh rate corresponding to the application scenario, as well as the baseline value of the interaction frequency corresponding to the application scenario, can be obtained. On this basis, taking the application scenario and the predicted interaction frequency as the basis, the reference value of the screen refresh rate and the baseline value of the interaction frequency corresponding to the application scenario are adjusted according to the predicted interaction frequency to obtain the final screen refresh rate of the terminal at the next moment. By predicting the click-through rate and the interaction frequency of the target application at the next moment and adjusting the screen refresh rate based on the prediction results, it is possible to achieve pre-adjustment of the screen refresh.

[0042] In the technical solution of the embodiments of the present disclosure, by performing feature processing on application-related data, the predicted usage probability of the target application that the user is about to click and the predicted interaction frequency for the target application are predicted. Based on the predicted interaction frequency, the screen refresh rate baseline value, and the interaction frequency baseline value, the pre-dynamic adjustment of the refresh rate is performed, avoiding the lag in adjusting the screen refresh rate in the related art, and enabling timely and accurate adjustment of the screen refresh rate, improving the accuracy and reliability. By performing pre-dynamic adjustment of the refresh rate based on the predicted interaction frequency, the screen refresh rate baseline value, and the interaction frequency baseline value, the problems caused by processing after the user perceives are avoided, and the screen refresh rate can be adjusted imperceptibly, improving the user experience. In addition, the predicted usage probability and predicted interaction frequency of the target application are obtained from the application-related data including the historical application sequence, and then the screen refresh rate of the terminal is jointly determined according to multiple dimensions such as the application scenario and the predicted interaction frequency. Since the screen refresh rate can be adjusted based on the predicted interaction frequency determined from the user's application-related data, the pertinence can be improved, making the adjusted screen refresh rate more in line with the actual situation of the user, with a higher matching degree, realizing personalized adjustment of the screen refresh rate, and also reducing the power consumption of the terminal, improving the pertinence and diversity.

[0043] Next, with reference to Figure 2 shown, each step in the screen refresh rate adjustment method in the embodiments of the present disclosure will be described in detail.

[0044] In step S210, application-related data is obtained.

[0045] In the embodiments of the present disclosure, the application-related data can be used to represent all data related to the application. Exemplarily, the application-related data can be obtained from multiple dimensions such as buried point data, positive and negative sample data, and historical application sequences. The application-related data includes user portraits, target applications, the context environment of the target application, and historical application sequences.

[0046] The buried point data can be the data generated during each interaction between each user and the application. The buried point data can be used to represent user behavior. Exemplarily, the application switching behavior of the user between different applications can be collected, such as switching from appA to appB, etc. At the same time, the context environment where the user is located during the application switching can be collected. The context environment can be used to represent the hardware situation when using the application, and can include, for example, but not limited to, one or more of the network status (such as wifi or traffic data), Bluetooth, and power information. In addition, the behavior data of the user during the application usage can be recorded simultaneously. The behavior data can include, for example, application information (such as the application name) and the average active interaction frequency of the user. It should be noted that the application app in the buried point data refers to the application used at the current moment and can be used to represent the target application. Based on this, each piece of buried point data can be composed of the context environment and the behavior data. For each user, since there may be multiple applications used, there can be multiple pieces of buried point data, and there is a one-to-one correspondence between the application and the buried point data. Exemplarily, a single piece of buried point data is organized into the following format, as shown in Table 1 specifically:

[0047] Table 1

[0048]

[0049]

[0050] Among them, app refers to the target application, and the average interaction frequency refers to the interaction frequency of the user with the target application.

[0051] After obtaining the above-mentioned buried point data, in order to improve the accuracy, the buried point data can be further processed. Referring to Figure 3 as shown, in step S310, data preprocessing can be performed on the buried point data, and the data preprocessing can be a cleaning operation. Exemplarily, the null values and empty strings of each field in the buried point data can be filtered out, and the applications without icons and the applications that are not actively interacted by the user can be filtered out to implement the cleaning of the buried point behavior table. The applications that are not actively interacted by the user refer to the applications that can be automatically executed, such as wireless settings or one-key screen locking, etc.

[0052] After cleaning the buried point data, the application type table of the application can be obtained. For example, the second-level type second_category of the application can be taken as the application scenario corresponding to the application, and it can be concatenated with the user behavior table. The second-level type of the application can be, for example, video type or image type, etc. Then the user behavior table is concatenated with the user installation table. The user installed application table refers to the set of installed applications and is used for subsequent negative sampling and the candidate app set required for prediction. That is, prediction can be performed within the corresponding range of the user installed application table, and prediction cannot be performed outside its range. Further, each feature in the buried point data can be encoded. Discrete features are directly encoded, and continuous features such as battery power and interaction frequency are binned into discrete values and then encoded for subsequent model processing. For example, the application name is semantic, and it can be encoded into a digital form and input into the model for processing. It should be noted that the buried point data can be positive sample data.

[0053] For the accuracy and comprehensiveness of prediction, the application association data can also include positive sample data, negative sample data, and historical application sequences, and the positive sample data can include the target application. Exemplarily, in step S320, positive and negative sample sampling can be performed, specifically including sampling rate calculation, positive sample sampling, and negative sample sampling. When obtaining positive sample data, the sampling probability of popular applications can be reduced to improve the reliability of the data. The number of popular applications can be one or more, and it can be the applications with the top N user numbers, which are specifically updated and determined according to actual needs. For example, a certain chat application or a video application, etc. In some embodiments, the sampling probability of popular applications in the buried point data is adjusted, and positive sample data is obtained according to the adjusted sampling probability. The sampling probability can be as shown in formula (1):

[0054]

[0055] where app i is the number of users of the i-th app, z(app i ) is the proportion of the number of users of the i-th app, and a is an adjustable parameter. After reducing the sampling probability, positive sample data can be obtained according to the reduced sampling probability.

[0056] Further, the user installed application table can be obtained from the database to determine negative sample data. The user installed application table can be the set of applications installed on the user's terminal. All applications in the set can be used as negative sample data.

[0057] In addition, multiple historical applications recently used by the user can be collected to form a historical application sequence of the user. The historical application sequence is the app behavior sequence before the current moment. When obtaining the historical application sequence in step S330, specifically, multiple historical applications used by the user within a preset period can be obtained. The preset period can be, for example, within one month, etc. To improve accuracy, the multiple obtained historical applications can be filtered. Applications with a residence time greater than a first preset duration or less than a second preset duration can be filtered out, where the first preset duration is greater than the second preset duration. For example, applications with too short a retention time in the sequence (such applications may be due to the user's accidental operation) are removed, and app behaviors with too long a time interval (more than one and a half hours) in the sequence are also filtered out. Based on this, other applications can be arranged in chronological order to form a historical application sequence.

[0058] It should be noted that since the application association data can include buried point data and the historical application sequence, the application association data can include multiple features. For example, it can include but is not limited to the user portrait, target application, and context environment of the target application in the buried point data, and can also include the historical application sequence. The user portrait can be a user identifier, the target application refers to the name or identifier of the currently used or to-be-predicted application, the context environment can be one or more of the network state (such as wifi or traffic data), Bluetooth, and battery information when using the target application, and the historical application sequence refers to the sequence of used historical applications arranged in a certain usage order. The historical application sequence can include multiple features, and the multiple features can include but are not limited to the application identifier app_id, application type second category, and the user's interaction frequency freq.

[0059] In step S220, feature extraction is performed on the application association data to determine the predicted usage probability of the target application included in the application association data and determine the predicted interaction frequency of the target application.

[0060] In the embodiments of the present disclosure, a deep learning model can be used to perform feature extraction on the application association data to obtain a target vector, and further perform feature fitting on the target vector to realize the prediction of the usage probability and interaction frequency. The deep learning model can be any type of model. For obtaining the target vector, feature processing can be performed through a Transfomer model.

[0061] Figure 4 The schematic diagram of the model structure is shown in reference Figure 4As shown in [figure], it mainly includes a feature vector layer, i.e., an embedding layer 410, a behavior sequence information extractor, namely an encoder (Transformer encoder) 420, a first classification predictor MLP 450, and a second classification predictor MLP 460. In addition, it may also include an average pooling layer 430 and a concatenation layer 440.

[0062] Among them, the feature vector layer, i.e., the embedding layer, is used to convert each feature into a vector. The behavior sequence information extractor is used to further extract features from the historical behavior sequence. The first classification predictor is used to predict the click-through rate of the app at the next moment, and the first classification predictor is used to predict the user interaction frequency at the next moment. Based on this, the entire prediction process can be as follows: The feature vector layer processes the target app and the historical app sequence in the app-related data, inputs the result into the encoder for processing, and performs average pooling on the sequence vectors of the historical app sequence output by the encoder. It is concatenated with the app vector of the target app, the context vector output by the context image through the feature vector layer, and the user portrait vector output by the user portrait through the feature vector layer to obtain the target vector. Further, the target vector is input into the first classification predictor to obtain the predicted usage probability, and the target vector is input into the second classification predictor to obtain the predicted interaction frequency.

[0063] Figure 5 Figure [number] schematically shows a flowchart for obtaining the target vector. Refer to Figure 5 As shown in [figure], it mainly includes the following steps:

[0064] In step S510, feature extraction is performed on the historical app sequence to obtain multiple sequence vectors, and feature extraction is performed on the target app to obtain the app vector.

[0065] In the embodiments of the present disclosure, the sequence vector refers to the vector of the historical app sequence output by the encoder. The number of sequence vectors is the same as the number of apps included in the historical app sequence. For example, if there are N apps, then N sequence vectors are output.

[0066] When obtaining the sequence vector, the feature vector layer can be used to process the historical app sequence in the app-related data to obtain an intermediate sequence vector, and the intermediate sequence vector is input into the encoder for encoding to obtain the sequence vector output by the encoder.

[0067] Figure 6 Figure [number] schematically shows a flowchart for obtaining the sequence vector. Refer to Figure 6 As shown in [figure], it mainly includes the following steps:

[0068] In step S610, based on the embedding matrix, vectors corresponding to multiple features included in each app in the historical app sequence are obtained;

[0069] In step S620, fuse each of the vectors to obtain an intermediate sequence vector;

[0070] In step S630, encode the intermediate sequence vector through an encoder to obtain the sequence vector.

[0071] In the embodiments of the present disclosure, when processing a historical application sequence, first, multiple features included in each application in the historical application sequence can be transformed through an embedding matrix. The multiple features may include, but are not limited to, an application identifier app_id, an application type second category, and a user's interaction frequency freq. Specifically, the embedding matrix can correspond to each feature one by one, that is, each feature has a corresponding embedding matrix to transform the feature. The embedding matrix can be constructed according to the feature encoding and a preset dimension, where the feature encoding can be the rows of the embedding matrix, and the preset dimension can be the columns of the embedding matrix. In the embodiments of the present disclosure, the preset dimensions of different features can be the same. For example, they can all be d, and the value of d can be 32. Of course, it can also be other values, which are specifically limited according to actual requirements.

[0072] When constructing the embedding matrix, first, it is necessary to determine the feature encoding corresponding to each feature. The feature encoding is used to represent the category corresponding to each feature, and different methods can be selected to determine it according to the continuous state of the feature. In some embodiments, if the category of the feature is a discrete feature, encode the discrete feature in the order of arrangement to determine the feature encoding; if the feature is a continuous feature, convert the continuous feature into a discrete feature and encode the discrete feature to determine the feature encoding. That is, if the feature is a discrete feature, add a preset value to the number of features to determine the feature encoding of the feature; if the feature is a continuous feature, the continuous feature can be binned to convert the continuous feature into a discrete feature, and add the discrete feature to a preset value to determine the feature encoding. The preset value can be 1 or other values, which are specifically limited according to actual requirements. It should be noted that the value of the feature encoding can be greater than or equal to 1, that is, the encoding starts from 1. The binning method can be equal-interval binning or other binning methods, such as model binning, etc., which are not specifically limited here.

[0073] In some embodiments, the embedding matrices of each feature in the application association data can be calculated separately. For example, the total number of users, the total number of apps, and the number of second-level app categories (i.e., the number of app types) are calculated. A preset value (e.g., 1) is added to each total number to form the feature codes in their respective corresponding embedding matrices. Among them, the feature codes all start from 1, and 0 corresponds to the embedding for subsequent complementation. The dimensions of all embedding matrices are the preset dimension d. For example, Wi-Fi and Bluetooth have only two categories: connected and unconnected, and the sizes of the formed embedding matrices are both (2, d). The battery level is a continuous feature. When encoding, the battery level from 0 to 100 is binned at intervals of 10 to discretize the battery level to obtain discrete features. Therefore, the values after discretization, i.e., the discrete features, are {0, 1, 2, …, 10}, and the size of the formed embedding matrix is (11, d). The interaction frequency is also a continuous feature, and it is similarly converted into discrete features according to the principle of equal-interval binning to form an embedding matrix.

[0074] After obtaining the embedding matrices, the specific feature values in the application association data can be used to retrieve the d-dimensional vectors corresponding to each feature from the corresponding embedding matrices. Each application and the target application in the historical application sequence have three features: the application identifier app_id, the application type second category, and the user's interaction frequency freq. Therefore, three embedding vectors of the features can be obtained through the embedding matrices, that is, the vectors of each feature. Exemplarily, the application identifier of each application in the historical application sequence is input into the embedding matrix corresponding to the application identifier to obtain a vector of the preset dimension as the first vector; the application type of each application is input into the embedding matrix corresponding to the application type to obtain a vector of the preset dimension as the second vector; the interaction frequency of each application is input into the embedding matrix corresponding to the interaction frequency to obtain a vector of the preset dimension as the third vector.

[0075] Furthermore, the vectors corresponding to the features in the historical application sequence, that is, the first vector, the second vector, and the third vector, can be concatenated to obtain a concatenated vector, and the concatenated vector is fused with the position information vector to obtain the intermediate sequence vector output by the embedding layer. Among them, the position information vector includes at least one of the position encoding and the time interval encoding. The position encoding is used to represent the absolute position of each application, and the time interval encoding is used to represent the time interval between adjacent applications in the historical application sequence.

[0076] In the embodiments of the present disclosure, to indicate the sequential relationship among various applications app in a historical application sequence of length N, different position encodings pos need to be configured at each position in the sequence. The position encoding is used to represent the sequential usage relationship between each application in the historical application sequence. The dimension of the position encoding matrix is (N + 1, d), where N represents the positions of N historical applications, and the last position is the position of the target app (the target application at the next moment). The position encoding matrix is trained in a manner of model learning, which is not specifically limited herein. Based on this, the concatenated vector and the position encoding can be fused to obtain the intermediate sequence vector output by the embedding layer. The fusion here can be an addition operation, that is, adding the concatenated vector and the position encoding to construct a 3d-dimensional vector as the intermediate sequence vector.

[0077] In addition, when considering the user's interaction history as an ordered sequence and using position encoding to represent the sequential usage relationship between apps in the sequence, the time interval between applications in the historical application sequence can also be considered. Specifically, the time interval can be encoded to obtain the time interval encoding, and further, the position encoding and the time interval encoding are fused, and then fused with the concatenated vector to obtain the intermediate sequence vector, so as to implement an attention module that can perceive the time interval, improving the accuracy and comprehensiveness.

[0078] After obtaining the intermediate sequence vector, the intermediate sequence vector can be input into the encoder, and the encoder encodes the intermediate sequence vector to obtain the sequence vector output by the encoder.

[0079] For the input target application, the same method can be used for processing. Exemplarily, the types of features of the target application are the same as those of the applications included in the historical application sequence, and can also include but are not limited to the application identifier app_id, the application type second category, and the user's interaction frequency freq. The feature vector of each feature can be determined according to the embedding matrix of the features of the target application, and the feature vector of each feature is added to the position encoding to obtain the intermediate application vector. Further, the intermediate application vector is encoded to obtain the application vector. Since the processing method of the target application is the same as that of each application in the historical application sequence, it will not be elaborated here.

[0080] Figure 7 schematically shows a flowchart for obtaining the sequence vector, and Figure 7 is a specific implementation manner of step S630. Referring to Figure 7 shown in, it mainly includes the following steps:

[0081] In step S710, the intermediate sequence vector is subjected to convolution processing through the multi-head self-attention layer in the encoder to obtain the multi-head self-attention vector;

[0082] In step S720, perform a fully connected process on the multi-head self-attention vector to obtain the sequence vector corresponding to the intermediate sequence vector; the sequence vector includes the representation vectors of each application in the historical application sequence.

[0083] In the embodiments of the present disclosure, the encoder may include a multi-head attention module and a feed-forward neural network module. The input data of the encoder will first be input into the multi-head attention module, i.e., the Self-Attention layer, which allows the encoder to use the information of other vectors in the sequence when encoding a certain vector. Then, the output of the multi-head attention module will flow into the feed-forward neural network module.

[0084] The multi-head attention module may include a multi-head attention layer; in addition, it may also include an Add&Norm layer. Among them, Add represents residual connection, which is used to prevent network degradation; Norm is used to normalize the activation values of each layer; in addition, it also includes a linear fully connected layer. The multi-head attention layer contains multiple parallel self-attention layers, and each multi-head attention layer is called a head. For each head, before performing attention calculation, the vector will be mapped by three linear layers, and the output of this attention head will be concatenated and then input into the last linear layer for integration.

[0085] The encoder adopts a multi-head attention mechanism, and the dimension d of the embedding matrix is set to 32. The input vector in the Transformer is 96-dimensional, and the vector dimension is not high. After testing, the number of heads of the adopted multi-head attention layer is 3, and the number of encoding layers is 1. In each head, the 96-dimensional matrix is mapped to 32 dimensions.

[0086] Based on this, the process of processing through the multi-head attention module can be: concatenate the attention outputs of each head to obtain a concatenated result, input the concatenated result into a linear fully connected layer for processing to obtain the final multi-head self-attention vector. Further, the multi-head self-attention vector can be subjected to residual processing and normalization processing to obtain a normalized result for input into the FFN layer for subsequent processing.

[0087] Among them, the process of obtaining the multi-head self-attention vector by performing convolution processing on the intermediate sequence vector through the multi-head self-attention layer in the encoder may include the following steps:

[0088] Map the intermediate sequence vectors at each position to corresponding reference matrices through multiple linear transformation matrices;

[0089] Based on scaled dot-product attention, perform logical processing on the reference matrices to determine the weighted sum at each position to determine the output of each head;

[0090] Concatenate the outputs of each head to obtain a concatenated result, and synthesize the concatenated result to obtain a multi-head self-attention vector.

[0091] Among them, the linear transformation matrix can be W Q , W K and W V . The linear transformation matrix can be linearly mapped with the intermediate sequence vectors at each position of the input to obtain the corresponding reference matrices Q, K, and V. That is, the intermediate sequence vector at each position is mapped to three reference matrices. Further, logical processing can be performed on the reference matrices based on scaled dot-product attention to determine the weighted sum at each position, and the weighted sum at each position can be used as the output of each head. The weighted sum at each position can be calculated using formula (2):

[0092]

[0093] where d k is the key dimension in a single head, for example, 32.

[0094] Further, after concatenating the attention outputs of each head, a concatenated result can be obtained, and the concatenated result can be input into a linear fully connected layer for synthesis to implement fully connected processing, obtaining the final output of the multi-head self-force layer, that is, the multi-head self-attention vector. Next, the output multi-head self-attention vector can also be subjected to residual processing and normalization processing using the normalization layer in the multi-head self-attention module to obtain a normalized result. Further, the normalized result can be input into a FFN (Feed-forward network) for processing.

[0095] The feed-forward neural network module mainly can include two layers of fully connected layers and a normalization layer Add&Norm. Among them, the number of hidden layer neurons in the feed-forward neural network module is 4 times the output dimension of the attention layer of the multi-head self-attention module. In the feed-forward neural network module, the fully connected layer is used to integrate all features, that is, perform two-layer linear mapping on the output of the previous layer and activate it with a non-linear activation function in the middle. The fully connected layer plays the role of mapping the learned distributed feature representation to the sample label space. In actual use, the fully connected layer can be implemented by convolution operations. The normalization layer can include residual connection and normalization. The residual connection is used for feature transfer to prevent network degradation; normalization is used to normalize the activation values of each layer. The residual connection can enhance the fitting ability and performance of the model and improve the model processing efficiency.

[0096] Based on the above structure, the normalized result output by the multi-head attention module can be processed through two fully connected layers in the feed-forward neural network module, and the output fully connected result is input into the normalization layer for residual processing and normalization processing, so as to use the result output by the feed-forward neural network module as the sequence vector output by the encoder, and this sequence vector is the final output vector of the encoder.

[0097] Based on this, the application-related data input to the encoder includes the historical application sequence and the target application. Therefore, it can be considered that the input is a sequence of N + 1, and the output is still a sequence of N + 1. Moreover, each position in the output sequence vector corresponds to the representation of the historical application app at each moment in the processed historical application sequence and the representation of the target application target app.

[0098] It should be noted that the Transformer structure can be used for feature extraction of the historical application sequence, and its time complexity is quadratic with the sequence length N. When the sequence length further increases, the model time consumption will also increase significantly. Based on this, in order to improve the processing efficiency and reduce the model time consumption, various types of Transformer improved model structures with linear complexity can also be adopted, as long as they can achieve feature extraction of the sequence, and no specific limitation is made here.

[0099] In step S520, feature extraction is respectively performed on the user profile and the context environment to obtain the user profile vector and the context vector.

[0100] In the embodiments of the present disclosure, the user profile and the context environment can be transformed according to the respective embedding matrices corresponding to the user profile and the context environment to obtain the corresponding user profile vector and context vector. The specific method is the same as that in step S510 and will not be elaborated here. It should be noted that the context environment and the user profile only need to be processed through the embedding layer and do not need to be processed through the encoder.

[0101] In step S530, the multiple sequence vectors, the application vector, the user profile vector, and the context vector are fused to obtain the target vector.

[0102] In the embodiments of the present disclosure, the sequence vectors corresponding to each application in the historical application sequence can be averaged to obtain the average sequence vector; further, the average sequence vector, the application vector, the user profile vector, and the context vector are concatenated to obtain an overall vector as the target vector.

[0103] Figure 5The technical solution in [the above] can obtain the target vector according to the application correlation data including the historical application sequence and the target application. Since the historical application sequence contains the usage characteristics of multiple applications by the user at historical moments, it can truly reflect the user behavior characteristics, provide more useful information for feature processing, and improve the accuracy of the obtained target vector. Moreover, the Transformer encoder structure is used to process the user's historical application sequence, which can not only achieve parallel processing of the sequence but also has a powerful feature extraction ability, improving the processing efficiency and reliability.

[0104] After obtaining the target vector, the target vector can be input into a classification predictor (i.e., a classifier) to determine the predicted usage probability of the target application and the predicted interaction frequency of the target application. The predicted usage probability refers to the predicted usage probability of the target application at the next moment or the click-through rate of the target application at the next moment. The predicted interaction frequency can be the interaction frequency of the user with respect to the target application.

[0105] Two classification predictors can be constructed to respectively predict the predicted usage probability and the predicted interaction frequency of the target application. Exemplarily, the predicted usage probability can be determined by the first classification predictor, and the predicted interaction frequency can be determined by the second classification predictor. The first classification predictor can be a binary classifier, and the second classification predictor can be a multi-classifier. Both the first classification predictor and the second classification predictor adopt the multi-layer perceptron structure MLP. Except for the different number of neurons in the output layer, the hidden layer structures are the same, but the network parameters of the first classification predictor and the second classification predictor are not shared and are two independent network structures.

[0106] Both the first classification predictor and the second classification predictor can be composed of multiple multi-layer perceptron layers and a classification layer. Refer to Figure 8As shown in [Figure 0], both the first classification predictor and the second classification predictor can include two multilayer perceptron (MLP) layers, namely 801 and 802, and can also include a fully connected layer represented by 803 and a classification layer represented by 804 (sigmoid / softmax). Among them, each MLP layer is sequentially composed of a Dropout layer represented by 8011, a fully connected layer represented by 8012, a normalization layer represented by 8013, and an activation function represented by 8014. Among them, the Dropout layer is used to randomly delete and reduce the number of neurons, making the network simpler, that is, to discard some neurons in the network of the MLP layer, so as to process the target vector. Dropout means that during the model processing, the weights of some neurons in the network of the MLP layer are randomly made ineffective. Those non-working neurons can be temporarily considered not to be part of the network structure, but their weights are retained to prevent the model from overfitting. The fully connected layer is used to perform a fully connected process on the vector output by the Dropout layer; the normalization layer is used to normalize the fully connected result output by the fully connected layer. The activation function is used to perform a non-linear mapping on the normalized result output by the normalization layer.

[0107] Based on the above structure, the target vector can be feature-transformed through the Dropout layer, fully connected layer, normalization layer, and activation function in the multiple MLP layers of the first classification predictor, fully connected according to the transformed target vector, and classified to achieve feature fitting of the target vector, and the prediction usage probability of the target application at the next moment can be determined according to the fitted target vector. That is, the first classification predictor directly outputs the probability that the target application will be clicked at the next moment.

[0108] Similarly, in the process of determining the predicted interaction frequency, a binning and discretization operation is performed on the user interaction frequency. When numerical features span different orders of magnitude, the model may only be sensitive to large feature values, and at this time, binning operations need to be considered. The binning operation is used to offline continuous features into a series of discrete features of 0 / 1 to achieve discretization. Through the binning and discretization operation, the calculation speed can be improved and the robustness can be enhanced. Based on this, the target vector can be feature-transformed through the Dropout layer, fully connected layer, normalization layer, and activation function in the multiple MLP layers of the second classification predictor, fully connected according to the transformed target vector, and classified to output the probability that the user's interaction frequency falls on each bin. That is, what the second classification predictor outputs is the probability that the interaction frequency falls on each bin. Further, the interaction frequency corresponding to the bin with the highest probability can be determined as the predicted interaction frequency, that is, the interaction frequency of the user for the target application at the next moment.

[0109] Figure 9 schematically shows a flowchart for behavior prediction, refer to Figure 9 as shown in, mainly including the following steps:

[0110] In step S910, obtain application-related data, which includes user profile 901, target application 902, context environment 903 of the target application, and historical application sequence 904;

[0111] In step S920, process the application-related data according to the feature vector layer 905 and the encoder 906 to obtain the target vector 907;

[0112] In step S930, input the target vector 907 into the first classification predictor 908 to obtain the predicted usage probability 909;

[0113] In step S940, input the target vector 907 into the second classification predictor 910 to obtain the predicted interaction frequency 911.

[0114] In the embodiments of the present disclosure, by obtaining the target vector based on the historical application sequence, user profile, target application, and context environment of the target application determined from the historical applications, and fitting the target vector to obtain the predicted usage probability and predicted interaction frequency at the next moment on the target application. Since the comprehensiveness and accuracy of the data input into the model are improved through the historical application sequence, the accuracy of the obtained predicted usage probability and predicted interaction frequency can be improved, and the authenticity is also increased. At the same time, predictions can be made for each user, improving the pertinence.

[0115] Continue to refer to Figure 2 as shown in, in step S230, determine the application scenario based on the target application corresponding to the predicted usage probability, and obtain the baseline value of the interaction frequency corresponding to the application scenario.

[0116] In the embodiments of the present disclosure, after obtaining the predicted usage probability, if the predicted usage probability is greater than the preset threshold, it can be considered that the application to be clicked at the next moment is the target application. The preset threshold can be used to describe whether the target application is the application to be clicked at the next moment, and the preset threshold can be set according to actual needs, and no specific limitation is made here. If the predicted usage probability is less than the preset threshold, the target application can be updated, and steps S210 to S230 are re-executed to recalculate the predicted usage probability and the predicted interaction frequency. It should be noted that the target application can be any one of all the applications included in the user-installed application list.

[0117] When determining the target application to be clicked at the next moment according to the predicted usage probability, the application scenario can be determined based on the secondary category of the target application. The application scenario can be, for example, a video category or a game category, etc. After determining the application scenario, the screen refresh rate reference value and the interaction frequency baseline value corresponding to the application scenario can be determined. Among them, the interaction frequency baseline value is used to represent the average user interaction frequency baseline value of the application in the application scenario, and can be determined according to the average interaction frequency of all applications in the application scenario. The screen refresh rate reference value can be determined according to the application scenario, and the screen refresh rate reference values of different application scenarios are different.

[0118] Table 2

[0119] Application scenario Refresh rate reference value (HZ) Reading category 20 Game category 120 Video category 60 Social category 40

[0120] Since the application scenarios of the terminal are relatively complex, the requirements for the smoothness of the screen in different application scenarios are completely different. For example, in the reading scenario, the user's reading takes a long time, and the screen can be displayed in a low-frequency form of 20HZ when reading statically. For the game scenario, the user is in a dynamically changing environment with frequent interactions, and a screen refresh rate above 100HZ is required to achieve smooth screen display. Therefore, different screen refresh rate reference values can be set for different application scenarios, as shown in Table 2 for example. The value set in Table 2 is a reference value, and subsequent adjustments will be made up and down based on the reference value according to the high or low interaction frequency of the user to meet the personalized requirements.

[0121] For example, if the application scenario of the target application is a video category, referring to Table 2, the screen refresh rate reference value can be 60HZ.

[0122] In step S240, the screen refresh rate of the terminal is determined by combining the screen refresh rate reference value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency.

[0123] In the embodiments of the present disclosure, the screen refresh rate refers to the display frame rate of the terminal, that is, the number of times the terminal screen image is refreshed per second. The higher the terminal refresh rate, the better the display performance, and a more smooth and coherent screen display effect can be brought. The screen refresh rate can be, for example, 120HZ, or 144HZ, etc. In order to solve the situation where the terminal refresh rate remains unchanged, the screen refresh rate can be dynamically adjusted.

[0124] To solve the technical problems existing in the related art, it is possible to predict the interaction state between the user and the target application in the application scenario at the next moment in advance based on the user's historical application sequence, that is, to predict the usage probability and the interaction frequency. Further, based on the predicted usage probability and the interaction frequency in the above steps, through the application scenario and the user's screen interaction frequency, the screen refresh rate is adjusted on the preset screen refresh rate baseline value, and the screen refresh rate of the terminal at the next moment is obtained through the pre-adjustment strategy.

[0125] On this basis, Figure 10 schematically shows a flowchart for determining the screen refresh rate. Refer to Figure 10 shown in, which mainly includes the following steps:

[0126] In step S1010, obtain the difference between the predicted interaction frequency and the interaction frequency baseline value in the application scenario;

[0127] In step S1020, determine the ratio of the difference to the interaction frequency baseline value;

[0128] In step S1030, perform an addition operation on the screen refresh rate baseline value in the application scenario and the ratio to determine the screen refresh rate of the terminal.

[0129] Among them, the screen refresh rate of the terminal refers to the screen refresh rate of the terminal in the application scenario at the next moment. The specific calculation method can refer to formula (3) shown:

[0130]

[0131] Among them, V base represents the screen refresh rate baseline value in the application scenario, freq base is the average interaction frequency baseline value of the user corresponding to the application scenario, freq pred is the predicted interaction frequency of the target application at the next moment output by the model, v f is the unit value of the refresh rate adjustment, and V is the screen refresh rate of the terminal at the next moment that will be finally adjusted.

[0132] In the embodiments of the present disclosure, since the baseline value of the screen refresh rate and the baseline value of the interaction frequency corresponding to the application scenario can be adjusted based on the predicted prediction interaction frequency, the screen refresh rate in this application scenario can be determined. The baseline value of the screen refresh rate and the baseline value of the interaction frequency for each application scenario are different, so the final screen refresh rate of the terminal is also different. Therefore, for the same application scenario, the interaction frequencies of different users are different, and the adjustment methods of the screen refresh rate are also different, and the finally obtained screen refresh rates are also different. Therefore, the pertinence is improved, and the personalized adjustment of the screen refresh rate is realized, and the screen refresh rate can be automatically increased or decreased according to the actual usage behavior. For example, when a user uses a certain short video application, when the user's upswipe interaction rate is high, the screen refresh rate should also be appropriately increased.

[0133] In the embodiments of the present disclosure, the historical application sequence of the user is processed using a deep learning model to predict in advance the target application clicked by the user at the next moment and the predicted interaction frequency of the user, so as to realize the prediction of the user's behavior at the next moment in advance. Furthermore, the baseline value of the screen refresh rate and the baseline value of the interaction frequency in the application scenario corresponding to the target application are compared and adjusted according to the predicted interaction frequency. On this basis, taking the application scenario and the predicted interaction frequency as the basis, the baseline value of the screen refresh rate and the baseline value of the interaction frequency corresponding to the application scenario are adjusted according to the predicted interaction frequency, and the final screen refresh rate of the terminal at the next moment is obtained. By predicting the click-through rate and interaction frequency of the target application at the next moment and adjusting the screen refresh rate based on the prediction result, the pre-adjustment of the screen refresh can be realized. By realizing the pre-adjustment method through advance prediction, the user experience is improved, and sufficient time is provided for the adjustment of the mobile phone hardware.

[0134] Figure 11 Schematically shows a flowchart for determining the screen refresh rate of the terminal. Refer to Figure 11 As described above, it mainly includes the following steps:

[0135] In step S1101, application association data is obtained;

[0136] In step S1102, feature extraction is performed on the application association data to obtain vectors, feature vectors, context vectors, and user portrait vectors corresponding to multiple features;

[0137] In step S1103, the vectors corresponding to multiple features are fused with the position encoding to obtain an intermediate sequence vector, and the feature vector is fused with the position encoding to obtain an intermediate application vector;

[0138] In step S1104, the intermediate sequence vector is encoded to obtain a sequence vector, and the intermediate application vector is encoded to obtain an application vector;

[0139] In step S1105, average pooling is performed on multiple sequence vectors to obtain an average sequence vector;

[0140] In step S1106, the average sequence vector, the application vector, the context vector, and the user profile vector are concatenated to obtain a target vector;

[0141] In step S1107, the target vector is input into the first classification predictor to obtain a predicted usage probability;

[0142] In step S1108, the target vector is input into the second classification predictor to obtain a predicted interaction frequency;

[0143] In step S1109, the application scenario is determined according to the target application corresponding to the predicted usage probability, and the screen refresh rate reference value and the interaction frequency baseline value are determined;

[0144] In step S1110, the screen refresh rate is determined by combining the screen refresh rate reference value, the interaction frequency baseline value, and the predicted interaction frequency.

[0145] In the technical solution of the embodiments of the present disclosure, the dynamic adjustment method of the screen refresh rate of the terminal depends on the model's advance prediction of the user's behavior at the next moment. In addition to user information, app information, and context environment factors, this prediction model introduces a historical behavior sequence represented by the user's historical application sequence. The historical application sequence contains rich user habits, providing more useful information to the subsequent prediction and classification modules of the model. It can perform behavior prediction based on information that more conforms to the user's own habits, improving the prediction accuracy of the model, and thus accurately determining the predicted usage probability of the target application at the next moment and the predicted interaction frequency of the user. Using the Transformer encoder structure to process the user's historical application sequence can not only achieve parallel processing of the sequence but also has a powerful feature extraction ability, improving the accuracy and comprehensiveness of the obtained feature vectors. In addition, the app that the user will click next and the corresponding user interaction frequency are predicted simultaneously. Based on the predicted interaction frequency, the screen refresh rate reference value, and the interaction frequency baseline value, the pre-dynamic adjustment of the refresh rate is performed, avoiding the lag in the related technology and being able to adjust the screen refresh rate in a timely and accurate manner, improving the accuracy and reliability. Moreover, by pre-adjusting the refresh rate, it is avoided to perform after the user perceives it, improving the user experience. And, since the adjustment can be based on the user's application association data, the pertinence can be improved, making the adjusted screen refresh rate more in line with the actual situation of the user, with a higher matching degree, realizing personalized adjustment of the application refresh rate, and also reducing the system power consumption.

[0146] In the embodiments of the present disclosure, a screen refresh rate adjustment device is provided. Referring to Figure 12 as shown, the screen refresh rate adjustment device 1200 may include:

[0147] A data acquisition module 1201, configured to acquire application-related data;

[0148] An application prediction module 1202, configured to perform feature extraction on the application-related data, determine the predicted usage probability of the target application included in the application-related data, and determine the predicted interaction frequency of the target application;

[0149] A baseline value determination module 1203, configured to determine an application scenario based on the target application corresponding to the predicted usage probability, and acquire an interaction frequency baseline value corresponding to the application scenario;

[0150] A refresh rate adjustment module 1204, configured to determine the screen refresh rate of the terminal by combining the screen refresh rate reference value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency.

[0151] In an exemplary embodiment of the present disclosure, the application prediction module includes: a target vector determination module, configured to perform feature extraction on the application-related data to obtain a target vector; a feature fitting module, configured to perform feature fitting on the target vector to determine the predicted usage probability of the target application, and determine the predicted interaction frequency of the target application.

[0152] In an exemplary embodiment of the present disclosure, the application-related data includes a user profile, the context environment of the target application, and a historical application sequence; the target vector determination module includes: a first feature extraction module, configured to perform feature extraction on the historical application sequence to obtain a plurality of sequence vectors, and perform feature extraction on the target application to obtain an application vector; a second feature extraction module, configured to perform feature extraction on the user profile and the context environment respectively to obtain a user profile vector and a context vector; a feature fusion module, configured to fuse the plurality of sequence vectors, the application vector, the user profile vector, and the context vector to obtain the target vector.

[0153] In an exemplary embodiment of the present disclosure, the first feature extraction module includes: a feature processing module, configured to obtain vectors corresponding to a plurality of features included in each application in the historical application sequence based on an embedding matrix; a vector fusion module, configured to fuse the vectors to obtain an intermediate sequence vector; a vector encoding module, configured to encode the intermediate sequence vector through an encoder to obtain the sequence vector.

[0154] In an exemplary embodiment of the present disclosure, the vector fusion module includes: a splicing module, configured to splice the vectors to obtain a spliced vector, and fuse the spliced vector with a position information vector to obtain an intermediate sequence vector; wherein, the position information vector includes at least one of a position encoding and a time interval encoding.

[0155] In an exemplary embodiment of the present disclosure, the first feature extraction module includes: a target application processing module, configured to obtain a feature vector of the target application based on an embedding matrix, and fuse the feature vector with a positional encoding to obtain an intermediate application vector; a target application encoding module, configured to encode the intermediate application vector to obtain the application vector.

[0156] In an exemplary embodiment of the present disclosure, the vector encoding module includes: a multi-head attention processing module, configured to perform convolution processing on the intermediate sequence vector through a multi-head self-attention layer in an encoder to obtain a multi-head self-attention vector; a fully-connected processing module, configured to perform a fully-connected processing on the multi-head self-attention vector to obtain a sequence vector corresponding to the intermediate sequence vector; the sequence vector includes representation vectors of each application in a historical application sequence.

[0157] In an exemplary embodiment of the present disclosure, the multi-head attention processing module includes: a vector mapping module, configured to map the intermediate sequence vectors at each position into corresponding reference matrices through a plurality of linear transformation matrices; a weighting processing module, configured to perform logical processing on the reference matrices based on scaled dot-product attention to determine a weighted sum at each position to determine the output of each head; a splicing processing module, configured to splice the outputs of each head to obtain a splicing result, and perform a fully-connected processing on the splicing result to obtain a multi-head self-attention vector.

[0158] In an exemplary embodiment of the present disclosure, the fully-connected processing module is configured to: perform a fully-connected processing on the normalization result obtained by performing residual processing and normalization processing on the multi-head self-attention vector to obtain a fully-connected result; perform residual processing and normalization processing on the fully-connected result to obtain the sequence vector.

[0159] In an exemplary embodiment of the present disclosure, the feature fusion module includes: a pooling module, configured to perform average pooling on the plurality of sequence vectors to obtain an average sequence vector; a splicing control module, configured to splice the average sequence vector, the application vector, the user profile vector, and the context vector to obtain the target vector.

[0160] In an exemplary embodiment of the present disclosure, the feature fitting module includes: a usage probability prediction module, configured to perform feature fitting on the target vector through a plurality of multi-layer perceptron layers of a first classification predictor, and determine a predicted usage probability of the target application at the next moment according to the fitted target vector.

[0161] In an exemplary embodiment of the present disclosure, the feature fitting module includes: a probability calculation module, configured to perform feature fitting on the target vector through multiple multi-layer perceptron layers of the second classification predictor, and determine the probabilities of the user interaction frequency in each bucket according to the fitted target vector; an interaction frequency prediction module, configured to determine the interaction frequency corresponding to the bucket with the maximum probability as the predicted interaction frequency.

[0162] In an exemplary embodiment of the present disclosure, the multi-layer perceptron layer includes: a dropout layer, configured to discard some neurons of the network of the multi-layer perceptron layer to process the target vector; a fully connected layer, configured to perform a fully connected process on the vector output by the dropout layer; a normalization layer, configured to perform a normalization process on the fully connected result; and an activation function, configured to perform a non-linear mapping on the normalized result output by the normalization layer.

[0163] In an exemplary embodiment of the present disclosure, the refresh rate adjustment module includes: a comparison module, configured to obtain the difference between the predicted interaction frequency and the interaction frequency baseline value in the application scenario, and determine the ratio of the difference to the interaction frequency baseline value; an addition operation module, configured to perform an addition operation on the screen refresh rate reference value in the application scenario and the ratio to determine the screen refresh rate of the terminal.

[0164] In an exemplary embodiment of the present disclosure, the multiple features of each application in the historical application sequence include an application identifier, an application type, and an interaction frequency; the feature processing module includes: a first processing module, configured to perform feature transformation on the application identifier through the embedding matrix corresponding to the application identifier to obtain a first vector; a second processing module, configured to perform feature transformation on the application type through the embedding matrix corresponding to the application type to obtain a second vector; and a third processing module, configured to perform feature transformation on the interaction frequency through the embedding matrix corresponding to the interaction frequency to obtain a third vector.

[0165] In an exemplary embodiment of the present disclosure, the apparatus further includes: an embedding matrix acquisition module, configured to determine a feature encoding according to the categories of the features in the historical application sequence and the target application, and determine the embedding matrix according to the feature encoding and a preset dimension.

[0166] In an exemplary embodiment of the present disclosure, the embedding matrix acquisition module includes: a first feature encoding determination module, configured to, if the feature is a discrete feature, encode the discrete feature in the order of arrangement to determine the feature encoding; and a second feature encoding determination module, configured to, if the feature is a continuous feature, convert the continuous feature into a discrete feature and encode the discrete feature to determine the feature encoding.

[0167] In an exemplary embodiment of the present disclosure, the data acquisition module includes: a positive sample data acquisition module, configured to acquire buried point data and obtain positive sample data according to the buried point data; a negative sample data acquisition module, configured to determine negative sample data according to the user application installation list; and a data acquisition control module, configured to determine application association data according to the positive sample data, the negative sample data, and the historical application sequence.

[0168] In an exemplary embodiment of the present disclosure, the positive sample data acquisition module includes: a sampling probability calculation module, configured to adjust the sampling probability of popular applications in the buried point data and obtain positive sample data according to the adjusted sampling probability.

[0169] It should be noted that the specific details of each part in the above screen refresh rate adjustment device have been described in detail in the embodiments of the screen refresh rate adjustment method. For the undisclosed details, reference can be made to the embodiments of the method, and thus will not be elaborated herein.

[0170] An exemplary embodiment of the present disclosure further provides an electronic device. The electronic device may be the above terminal 101. Generally, the electronic device may include a processor and a memory. The memory is used to store executable instructions of the processor, and the processor is configured to execute the above screen refresh rate adjustment method by executing the executable instructions.

[0171] Next, taking Figure 13 the mobile terminal 1300 as an example, an exemplary description of the structure of the electronic device will be given. Those skilled in the art should understand that, except for the components specifically for mobile purposes, Figure 13 the structure in

[0172] can also be applied to fixed-type devices. Figure 13 As

[0173] The processor 1301 may include one or more processing units. For example, the processor 1301 may include an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor, and / or an NPU (Neural-Network Processing Unit), etc. The screen refresh rate adjustment method in this exemplary embodiment may be executed by the AP, the GPU, or the DSP. When the method involves neural network-related processing, it may be executed by the NPU. For example, the NPU may load neural network parameters and execute neural network-related algorithm instructions.

[0174] The encoder may encode (i.e., compress) an image or video to reduce the data size for easy storage or transmission. The decoder may decode (i.e., decompress) the encoded data of the image or video to restore the image or video data. The mobile terminal 1300 may support one or more encoders and decoders. For example, image formats such as JPEG (Joint Photographic Experts Group), PNG (Portable Network Graphics), BMP (Bitmap), etc., and video formats such as MPEG (Moving Picture Experts Group) 1, MPEG10, H.1063, H.1064, HEVC (High Efficiency Video Coding), etc.

[0175] The processor 1301 may form a connection with the memory 1302 or other components through the bus 1303.

[0176] The memory 1302 may be used to store computer-executable program code, and the executable program code includes instructions. The processor 1301 executes various functional applications and data processing of the mobile terminal 1300 by running the instructions stored in the memory 1302. The memory 1302 may also store application data, such as storing files such as images and videos.

[0177] The communication function of the mobile terminal 1300 can be implemented through the mobile communication module 1304, antenna 1, wireless communication module 1305, antenna 2, modulation and demodulation processor, baseband processor, etc. Antenna 1 and antenna 2 are used to transmit and receive electromagnetic wave signals. The mobile communication module 1304 can provide mobile communication solutions such as 3G, 4G, 5G, etc. applied to the mobile terminal 1300. The wireless communication module 1305 can provide wireless communication solutions such as wireless local area network, Bluetooth, near field communication, etc. applied to the mobile terminal 1300.

[0178] The display screen 1306 is used to implement the display function, such as displaying user interfaces, images, videos, etc. The camera module 1307 is used to implement the shooting function, such as shooting images, videos, etc., and the camera module may include a color temperature sensor array. The audio module 1308 is used to implement the audio function, such as playing audio, collecting voices, etc. The power module 1309 is used to implement the power management function, such as charging the battery, powering the device, monitoring the battery status, etc. The sensor module 1310 may include one or more sensors for implementing corresponding sensing and detection functions. For example, the sensor module 1310 may include an inertial sensor for detecting the motion pose of the mobile terminal 1300 and outputting inertial sensing data.

[0179] It should be noted that in the embodiments of the present disclosure, a computer-readable storage medium is also provided. The computer-readable storage medium may be included in the electronic device described in the above embodiments; or it may exist alone without being assembled into the electronic device.

[0180] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0181] The computer-readable storage medium can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium can be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination of the above.

[0182] A computer-readable storage medium carries one or more programs which, when executed by an electronic device, cause the electronic device to implement the methods described in the following embodiments.

[0183] From the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to cause a computing device (such as a personal computer, a server, a terminal device, or a network device, etc.) to execute the methods according to the embodiments of the present disclosure.

[0184] In addition, the above drawings are only schematic illustrations of the processes included in the methods according to the exemplary embodiments of the present disclosure, rather than for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes can be executed synchronously or asynchronously in, for example, multiple modules.

[0185] It should be noted that although several modules or units of devices for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-described modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0186] Those skilled in the art will readily think of other embodiments of the present disclosure after considering the specification and practicing the content disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include well-known common general knowledge or conventional technical means in the technical field not disclosed herein. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims. It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A method for adjusting screen refresh rate, characterized in that, comprising: obtaining application-related data; performing feature extraction on the application-related data to determine the predicted usage probability of the target application included in the application-related data, and determining the predicted interaction frequency of the target application; determining an application scenario based on the target application corresponding to the predicted usage probability, and obtaining an interaction frequency baseline value corresponding to the application scenario; combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency to determine the screen refresh rate of the terminal; wherein, combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency to determine the screen refresh rate of the terminal includes: obtaining the difference between the predicted interaction frequency and the interaction frequency baseline value in the application scenario, and determining the ratio of the difference to the interaction frequency baseline value; performing an addition operation on the screen refresh rate baseline value in the application scenario and the product of the ratio and the unit value of the refresh rate adjustment to determine the screen refresh rate of the terminal.

2. The screen refresh rate adjustment method according to claim 1, characterized in that, the performing feature extraction on the application-related data to determine the predicted usage probability of the target application included in the application-related data, and determining the predicted interaction frequency of the target application includes: performing feature extraction on the application-related data to obtain a target vector; performing feature fitting on the target vector to determine the predicted usage probability of the target application, and determining the predicted interaction frequency of the target application.

3. The screen refresh rate adjustment method according to claim 2, characterized in that, the application-related data includes a user profile, the context environment of the target application, and a historical application sequence; the performing feature extraction on the application-related data to obtain a target vector includes: performing feature extraction on the historical application sequence to obtain a plurality of sequence vectors, and performing feature extraction on the target application to obtain an application vector; performing feature extraction on the user profile and the context environment respectively to obtain a user profile vector and a context vector; fusing the plurality of sequence vectors, the application vector, the user profile vector, and the context vector to obtain the target vector.

4. The screen refresh rate adjustment method according to claim 3, characterized in that, each application in the historical application sequence includes a plurality of features, and the performing feature extraction on the historical application sequence to obtain a sequence vector includes: based on an embedding matrix, obtaining vectors corresponding to the plurality of features included in each application in the historical application sequence; fusing the vectors to obtain an intermediate sequence vector; encoding the intermediate sequence vector by an encoder to obtain the sequence vector.

5. The screen refresh rate adjustment method according to claim 4, characterized in that, the fusing the vectors to obtain an intermediate sequence vector includes: concatenating the vectors to obtain a concatenated vector, and fusing the concatenated vector with a position information vector to obtain an intermediate sequence vector; Wherein, the position information vector includes at least one of position encoding and time interval encoding.

6. The screen refresh rate adjustment method according to claim 3, wherein, the obtaining of the application vector by performing feature extraction on the target application includes: obtaining a feature vector of the target application based on an embedding matrix, and fusing the feature vector with the position encoding to obtain an intermediate application vector; encoding the intermediate application vector to obtain the application vector.

7. The screen refresh rate adjustment method according to claim 4, wherein, the encoding of the intermediate sequence vector by the encoder to obtain the sequence vector includes: performing convolutional processing on the intermediate sequence vector through a multi-head self-attention layer in the encoder to obtain a multi-head self-attention vector; performing a fully connected process on the multi-head self-attention vector to obtain the sequence vector corresponding to the intermediate sequence vector; the sequence vector includes the representation vectors of each application in the historical application sequence.

8. The screen refresh rate adjustment method according to claim 3, wherein, the fusing of the multiple sequence vectors, the application vector, the user profile vector, and the context vector to obtain the target vector includes: performing average pooling on the multiple sequence vectors to obtain an average sequence vector; concatenating the average sequence vector, the application vector, the user profile vector, and the context vector to obtain the target vector.

9. The screen refresh rate adjustment method according to claim 2, wherein, the determining of the predicted usage probability of the target application by performing feature fitting on the target vector includes: performing feature fitting on the target vector through multiple multi-layer perceptron layers of a first classification predictor, and determining the predicted usage probability of the target application at the next moment according to the fitted target vector.

10. The screen refresh rate adjustment method according to claim 2, wherein, the determining of the predicted interaction frequency of the target application includes: performing feature fitting on the target vector through multiple multi-layer perceptron layers of a second classification predictor, and determining the probabilities of the user interaction frequency in each bucket according to the fitted target vector; determining the interaction frequency corresponding to the bucket with the highest probability as the predicted interaction frequency.

11. The screen refresh rate adjustment method according to claim 4, wherein, the multiple features of each application in the historical application sequence include an application identifier, an application type, and an interaction frequency; the obtaining of the vectors corresponding to the multiple features included in each application in the historical application sequence based on the embedding matrix includes: performing feature transformation on the application identifier through the embedding matrix corresponding to the application identifier to obtain a first vector; performing feature transformation on the application type through the embedding matrix corresponding to the application type to obtain a second vector; performing feature transformation on the interaction frequency through the embedding matrix corresponding to the interaction frequency to obtain a third vector.

12. The screen refresh rate adjustment method according to claim 11, wherein, the method further includes: Determine a feature encoding according to the categories of the features in the historical application sequence and the target application, and determine the embedding matrix according to the feature encoding and a preset dimension.

13. The screen refresh rate adjustment method according to claim 12, wherein, the determining a feature encoding according to the categories of the features in the historical application sequence and the target application includes: if the feature is a discrete feature, encoding the discrete feature in the arrangement order to determine the feature encoding; if the feature is a continuous feature, converting the continuous feature into a discrete feature and encoding the discrete feature to determine the feature encoding.

14. A screen refresh rate adjustment device, wherein, it includes: a data acquisition module, configured to acquire application-related data; an application prediction module, configured to perform feature extraction on the application-related data, determine the predicted usage probability of the target application included in the application-related data, and determine the predicted interaction frequency of the target application; a baseline value determination module, configured to determine an application scenario based on the target application corresponding to the predicted usage probability, and obtain an interaction frequency baseline value corresponding to the application scenario; a refresh rate adjustment module, configured to determine the screen refresh rate of the terminal by combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency; wherein, combining the screen refresh rate baseline value of the application scenario, the interaction frequency baseline value, and the predicted interaction frequency to determine the screen refresh rate of the terminal includes: obtaining a difference between the predicted interaction frequency and the interaction frequency baseline value in the application scenario, and determining a ratio of the difference to the interaction frequency baseline value; performing an addition operation on the screen refresh rate baseline value in the application scenario and the product of the ratio and the unit value of the refresh rate adjustment to determine the screen refresh rate of the terminal.

15. An electronic device, wherein, it includes: a processor; and a memory, configured to store executable instructions of the processor; wherein, the processor is configured to execute the screen refresh rate adjustment method according to any one of claims 1-13 by executing the executable instructions.

16. A computer-readable storage medium, on which a computer program is stored, wherein, the computer program, when executed by a processor, implements the screen refresh rate adjustment method according to any one of claims 1-13.

Images