A deep learning-based mobile phone theme recommendation optimization method

Abstract

This application discloses a deep learning-based method for optimizing mobile phone theme recommendations, relating to the technical field of mobile phone theme recommendation. The method includes: collecting real-time usage characteristics of a user's mobile phone theme usage to determine if there are any usage problems; if usage problems exist, collecting real-time usage information to determine if the user accepts the usage problems; if the user does not accept the usage problems, collecting the user's personal information to evaluate a theme-changing intention value; collecting the user's mobile phone usage scenario information to evaluate a scenario-changing intention value; constructing a deep learning model by combining the theme-changing intention value and the scenario-changing intention value; evaluating the recommendation success rate; and when the recommendation success rate reaches a preset recommendation success rate standard, collecting the user's theme preference data and recommending a mobile phone theme. This application improves the accuracy of deep learning-based mobile phone theme recommendation optimization.

Description

Technical Field

[0001] This application relates to the technical field of mobile phone theme recommendation, and in particular to a deep learning-based optimization method for mobile phone theme recommendation. Background Technology

[0002] Phone themes allow users to easily personalize their phones by downloading a custom theme app and setting their own wallpaper, screen saver, ringtone, interface, and icons. Different themes create a more immersive experience, moving users away from a static interface, images, and colors. Due to the sheer number of phone themes, theme recommendations are helpful for users to quickly find suitable ones. However, current theme recommendations often focus solely on user preferences, prioritizing themes to increase the recommendation rate. They fail to consider users' need to change themes, leading to frequent and inconsistent recommendations and reducing the success rate of theme recommendations. Summary of the Invention

[0003] The purpose of this invention is to provide a deep learning-based mobile phone theme recommendation optimization method to solve the problems mentioned in the background art.

[0004] This application provides a deep learning-based method for optimizing mobile theme recommendations, employing the following technical solution:

[0005] Collect real-time usage characteristics of users' mobile phone themes, and determine whether there are any usage problems based on these characteristics;

[0006] If there are usage issues, real-time usage information of the user's mobile theme will be collected, and the user's acceptance of the usage issues will be determined based on the real-time usage information.

[0007] If the user does not accept the usage problem, the user's personal information will be collected, and the user's willingness to change the theme will be evaluated based on the user's personal information.

[0008] Collect information on the scenarios in which users use their mobile phones, and evaluate the willingness to switch scenarios based on this information.

[0009] A deep learning model for mobile theme recommendation is constructed by combining theme change intention value and scene change intention value;

[0010] The success rate of mobile theme recommendation is evaluated using a deep learning model. When the success rate reaches the preset recommendation success rate standard, user theme preference data is collected, and mobile themes are recommended based on the theme preference data.

[0011] Preferably, the step of collecting real-time usage characteristics of users' mobile phone themes and determining whether there are usage problems based on these real-time usage characteristics specifically includes:

[0012] Collect real-time usage characteristics of users using mobile phone themes, including real-time interaction characteristics and real-time mobile phone running status characteristics;

[0013] Obtain the user's historical usage characteristics before using the mobile phone theme, including historical interaction characteristics and historical mobile phone running status characteristics;

[0014] By comparing real-time and historical mobile phone operating status characteristics, the performance degradation value of the mobile phone operating performance can be confirmed.

[0015] By comparing real-time interaction features with historical interaction features, the interaction performance degradation value was confirmed.

[0016] By combining performance degradation and interaction degradation values, the problem value of the user's use of the mobile theme is determined. When the problem value reaches the preset problem value threshold, it is determined that the user has a problem using the mobile theme; otherwise, it is determined that the user does not have a problem using the mobile theme.

[0017] Preferably, the step of comparing real-time interaction features and historical interaction features to confirm the interaction performance degradation value specifically includes:

[0018] By comparing real-time interaction features with historical interaction features, the feature comparison results are obtained, and the positioning time difference of the user's location software position is extracted based on the feature comparison results.

[0019] The loading time difference is extracted based on the feature comparison results, which represents the average software loading time.

[0020] The response time difference of the user's page scrolling is extracted based on the feature comparison results;

[0021] The interaction degradation value is obtained by comprehensively evaluating the positioning time difference, loading time difference, and response time difference.

[0022] Preferably, the step of collecting real-time usage information of the user's mobile phone theme if a usage problem exists, and determining whether the user accepts the usage problem based on the real-time usage information, specifically includes:

[0023] Collect real-time usage information of the user's topic, including topic adjustment information and topic usage duration;

[0024] Collect the historical usage time of themes used by users, extract the average replacement time of the phone theme due to usage issues from the historical usage time, and record it as the problem replacement time.

[0025] Calculate the difference between the duration of topic usage and the duration of issue replacement, and determine whether the duration difference reaches the preset duration difference threshold;

[0026] If the time difference threshold is reached, it is determined that the user accepts the usage problem;

[0027] If the duration difference threshold is not reached, determine whether the user has adjusted the phone theme parameters based on the theme adjustment information.

[0028] If the phone theme parameters have not been adjusted, it is determined that the user accepts the usage problem; otherwise, it is determined that the user does not accept the usage problem.

[0029] Preferably, the step of collecting user personal information and assessing the user's willingness to change themes based on that personal information if the user does not accept the usage problem specifically includes:

[0030] Collect users' personal information and analyze it to determine the user's tolerance level for mobile phone themes;

[0031] Real-time difficulty information is extracted based on real-time usage information. It is then determined whether the real-time difficulty information meets the tolerance standard. If it does, the difference between the real-time difficulty information and the tolerance standard is calculated and used as the user's theme change intention value.

[0032] If the tolerance for difficulty is not met, the average time it takes for users to change themes will be collected and recorded as the daily change time.

[0033] The difference between the duration of topic usage and the duration of daily topic replacement is calculated, and the topic replacement willingness value is obtained by combining real-time difficulty information.

[0034] Preferably, the step of collecting user personal information and analyzing the user's personal information to obtain the user's tolerance level for mobile phone themes specifically includes:

[0035] The phone themes that were replaced due to usage issues were recorded as problem themes. The common characteristics of the problem themes were analyzed to obtain multiple problem characteristics.

[0036] Calculate the worst value for each problem feature, and derive the basic tolerance standard based on the worst values ​​of all problem features;

[0037] Collect user personal information, including the user's proficiency in switching themes and proficiency in using mobile applications;

[0038] The basic tolerance standard is adjusted based on the user's personal information to obtain the user's tolerance level for mobile phone themes.

[0039] Preferably, the step of calculating the difference between the topic usage time and the daily replacement time, and combining this with real-time difficulty information to obtain the topic replacement intention value, specifically includes:

[0040] Collect users' historical usage topics, filter and remove problematic topics from the historical usage topics to get the daily topics;

[0041] Extract the average values ​​of different daily characteristics of daily topics, and form daily standards based on the average values ​​of all daily characteristics;

[0042] Based on the real-time difficulty information, the real-time value of the feature corresponding to the daily feature is extracted, and the average difference between the real-time value of the feature and the daily standard is calculated and recorded as the real-time difficulty value.

[0043] The difference between the duration of topic usage and the duration of daily topic replacement is calculated, and the topic replacement willingness value is obtained by combining the real-time difficulty value.

[0044] Preferably, the step of collecting scenario information of users using their mobile phones and evaluating the scenario change intention value based on the scenario information specifically includes:

[0045] Collect real-time scene information of users using their mobile phones, and determine whether users have spare time based on the real-time scene information;

[0046] If the user has spare time, obtain the user's theme-changing scenario information, compare the similarity between the real-time scenario information and the changed scenario information, and use it as the scenario-changing willingness value;

[0047] If the user has no free time, obtain the real-time functions of the user's mobile phone and determine the impact value of the mobile phone theme on the real-time functions;

[0048] Collect data on users' demand for real-time functions to obtain demand values ​​for real-time functions, and combine these with impact values ​​to obtain a willingness to change scenarios value.

[0049] Preferably, the step of obtaining the user's real-time mobile phone usage data and determining the impact of the mobile phone theme on real-time functionality if the user has no free time specifically includes:

[0050] Obtain real-time data on user phone usage and extract evaluation features of these real-time functions;

[0051] Collect real-time feature information of the evaluation features corresponding to the real-time function, and evaluate the real-time performance value of the real-time function based on the real-time feature information;

[0052] Collect the best performance value of the user's mobile phone when using real-time functions, calculate the difference between the real-time performance value and the best performance value, and use it as the impact value of the mobile phone theme on real-time functions.

[0053] Preferably, the step of collecting user demand for real-time functions and obtaining the demand value for real-time functions specifically includes:

[0054] Collect user operation information for real-time functions, and determine whether the user is completely waiting for the real-time functions to load based on the operation information;

[0055] If users do not wait entirely for the real-time function to load, the frequency of repeated user actions is counted as the demand value for the real-time function.

[0056] In summary, this application includes at least one of the following beneficial technical effects:

[0057] 1. Based on the user's interaction with the phone theme, including software location, loading, response, and phone performance, determine if the user is experiencing any usage issues. If issues are found, further assess the user's acceptance of the theme based on theme adjustment information and usage duration. If the user is not experiencing issues, evaluate their willingness to change the theme based on personal information and then evaluate their willingness to change the theme based on contextual information. A deep learning model for phone theme recommendation is constructed by combining the theme and contextual willingness to change themes. This model evaluates the success rate of theme recommendations and recommends themes with higher success rates, improving the accuracy of deep learning-based phone theme recommendation optimization.

[0058] 2. A basic tolerance standard is established using the worst-case values ​​of the problem features of phone themes that have been changed due to usage issues. This standard is then adjusted based on the user's proficiency in switching themes and their familiarity with the phone application, resulting in a user's tolerance difficulty standard for phone themes. Real-time difficulty information is extracted from the user's usage. If the real-time difficulty information meets the tolerance difficulty standard, the difference between the real-time difficulty information and the tolerance difficulty standard is calculated as the user's theme-changing willingness value. If the real-time difficulty information does not meet the tolerance difficulty standard, the phone theme is compared with the daily characteristics of historically used themes, and the difference is used as the real-time difficulty value. This difference, combined with the difference between the theme usage time and the daily change time, yields the theme-changing willingness value. By considering the user's tolerance level for phone theme issues, the impact of the phone theme on the user, and the user's willingness to change themes, it is beneficial to recommend phone themes when the user's willingness to change is stronger. This improves the success rate of phone theme recommendations and reduces the inconvenience caused to users when they do not want to change themes, thus enhancing the convenience of deep learning-based phone theme recommendations.

[0059] 3. Collect real-time scenario information of user phone usage to determine if the user has spare time. If the user has spare time, obtain the scenario information for changing themes, and compare the similarity between the real-time scenario information and the changed scenario information as a scenario-based theme-changing intention value. If the user does not have spare time, evaluate the performance of the user's current functions and assess the impact of the phone theme on the user's current performance. Combine this with the real-time function demand value obtained from the frequency of repeated user operations to obtain a comprehensive scenario-based theme-changing intention value. Evaluating users' willingness to change phone themes in different scenarios helps to recommend phone themes in appropriate scenarios, better match user usage, improve user experience, and enhance the intelligence of deep learning-based phone theme recommendations. Attached Figure Description

[0060] Figure 1 This is a schematic diagram illustrating the specific steps of an embodiment of a deep learning-based mobile phone theme recommendation optimization method according to the present invention. Detailed Implementation

[0061] The following examples and... Figure 1 The present invention will be described in further detail, but the embodiments of the present invention are not limited thereto.

[0062] This invention discloses a deep learning-based method for optimizing mobile phone theme recommendations, specifically including the following steps:

[0063] Step S1: Collect real-time usage characteristics of the user's mobile theme, and determine whether there are any usage problems based on the real-time usage characteristics.

[0064] If there are no usage issues, obtain the average usage time of the user's mobile theme, calculate the difference between the average usage time and the actual usage time of the current mobile theme, and use it as the theme change intention value.

[0065] Step S2: If there is a usage problem, collect the user's real-time usage information of the mobile phone theme, and determine whether the user accepts the usage problem based on the real-time usage information.

[0066] Step S3: If the user does not accept the usage problem, collect the user's personal information and evaluate the user's willingness to change the theme based on the user's personal information.

[0067] If a user accepts the usage problem, it is considered equivalent to the phone theme not having a usage problem. Therefore, the evaluation method for their willingness to change the theme is the same as the evaluation method when there is no usage problem.

[0068] Step S4: Collect scenario information of the user's mobile phone usage, and evaluate the scenario change intention value based on the scenario information.

[0069] Step S5: Combine the theme change intention value and scene change intention value to construct a deep learning model for mobile theme recommendation.

[0070] Deep learning models exhibit autonomous behavior, dynamically allocating weights to different features and automatically adjusting weight distribution to meet constraints. Both themes and environments change intentions, influencing a user's willingness to change their phone's theme, thus affecting the theme recommendation success rate. Deep learning models can autonomously adjust the weights of these intentions to achieve the desired recommendation success rate. Users set learning boundaries through initialization, optimizers, and regularization; within these boundaries, the model autonomously explores the optimal solution through backpropagation. This collaboration between the two creates an intelligent system that is both directionally guided and flexibly adaptable. This mechanism avoids the rigidity of traditional rule engines and prevents the uncontrollability of purely data-driven approaches, further optimizing the deep learning model for phone theme recommendations.

[0071] Step S6: Use a deep learning model to evaluate the success rate of mobile phone theme recommendations. When the success rate reaches the preset success rate standard, collect the user's theme preference data and recommend mobile phone themes based on the theme preference data.

[0072] When the recommendation success rate reaches a preset standard, recommending phone themes can significantly improve the success rate. Of course, the success rate is also influenced by user preferences. Recommending themes that users prefer can further improve the success rate, which can be achieved using existing models and methods for recommending phone themes based on user preferences. For example, features of the themes used in a user's history can be extracted, and the theme with the highest similarity can be selected for recommendation.

[0073] In practical applications, current mobile theme recommendations often only consider situations where users want to change their theme. This leads to either frequent recommendations when users don't intend to change their theme, causing annoyance and inconvenience, or recommendations only appear when a user clearly wants to change their theme, missing opportunities when the user actually wants to, significantly reducing the success rate of theme recommendations. However, by evaluating users' individual willingness to change their theme, and their willingness to change in different scenarios, it's beneficial to increase the success rate of theme recommendations while minimizing inconvenience and improving the utilization rate of deep learning models for theme recommendations. For example, if a user feels uncomfortable using theme A, recommending a new theme at this time is more likely to encourage the user to switch, not only avoiding annoyance but also addressing the user's discomfort promptly, greatly improving the user experience.

[0074] The steps for collecting real-time usage characteristics of users' mobile phone themes and determining whether there are any usage problems based on these characteristics are as follows:

[0075] Step S11: Collect real-time usage characteristics of the user's mobile phone theme. Real-time usage characteristics include real-time interaction characteristics and real-time mobile phone running status characteristics.

[0076] Step S12: Obtain the user's historical usage characteristics before using the mobile phone theme. The historical usage characteristics include historical interaction characteristics and historical mobile phone running status characteristics.

[0077] Step S13: Compare the real-time mobile phone operating status characteristics with the historical mobile phone operating status characteristics to confirm the performance degradation value of the mobile phone operating performance.

[0078] Historical operating status characteristics refer to the average performance values ​​of various phone operating characteristics when the current phone theme is not being used. These characteristics include parameters such as CPU utilization, memory consumption, power consumption, and screen resolution. The degradation value of each characteristic is calculated, and after normalization, a performance deep learning model is used to assign weight ratios. The performance degradation value is then calculated based on these weight ratios. For example, memory consumption performance degradation value × weight ratio 1 + power consumption performance degradation value × weight ratio 2 = performance degradation value.

[0079] Step S14: Compare real-time interaction features with historical interaction features to confirm the interaction performance degradation value.

[0080] Step S15: Combine the performance degradation value and the interaction degradation value to confirm the problem value of the user's use of the mobile theme. When the problem value reaches the preset problem value threshold, it is determined that the user has a problem with the use of the mobile theme; otherwise, it is determined that the user does not have a problem with the use of the mobile theme.

[0081] In practical applications, when constructing a deep learning-based problem evaluation model that combines performance degradation and interaction degradation in mobile application user experience optimization scenarios, dynamic weight allocation can be achieved through deep neural networks. Specifically, the first step is to build a dataset containing multi-dimensional features: collecting performance monitoring data and interaction behavior logs for different themes in real-world usage scenarios, and labeling them with comprehensive problem value tags. A dual-channel network architecture with an attention mechanism is designed. The performance feature branch uses LSTM to process temporal resource consumption data, while the interaction feature branch uses CNN to extract spatial patterns of touch events. The final problem value prediction is generated through an adaptive weight fusion layer. In a real-world test case, a certain anime theme showed an 18% rendering frame rate decay in the performance dimension (performance degradation value 0.18) and an average click latency of 112ms in the interaction dimension (interaction degradation value 0.22). The trained model automatically assigned a weight ratio of 0.62:0.38, and the problem value was evaluated based on this weight ratio. The problem value threshold can be set according to the actual user experience. For example, most users typically only experience noticeable discomfort when the problem value is 30, so the problem value threshold can be set to 30. If the issue value does not reach the preset issue value threshold, it can be considered a minor issue that does not affect user experience and is assumed to be non-existent. For example, each theme consumes a certain amount of power; if a theme consumes little power, it can be considered to have no issue.

[0082] The steps for identifying the interaction performance degradation value by comparing real-time interaction features with historical interaction features are as follows:

[0083] Step S141: Compare real-time interaction features and historical interaction features to obtain feature comparison results, and extract the positioning time difference of the user's location software position based on the feature comparison results.

[0084] Interaction features include the time it takes for the user to locate the software, the software loading time, and the response time for the user to swipe pages. Therefore, by comparing real-time interaction features with historical interaction features, data such as location time difference, loading time difference, and response time difference can be calculated. Historical interaction features refer to the average value of various interaction features when the current phone theme is not used. For example, if the average time for a user to locate all software when the current phone theme is not used is 1 second, then 1 second is the location time for the user to locate software in the historical interaction features.

[0085] Step S142: Extract the loading time difference of the average software loading time based on the feature comparison results.

[0086] The average software loading time refers to the average loading time of all software, that is, the average time from clicking to entering the software.

[0087] Step S143: Extract the response time difference of the user's page swiping based on the feature comparison results.

[0088] The response time for a user to swipe a page refers to the time it takes for the user to swipe until the page is fully scrolled.

[0089] Step S144: Combine the positioning time difference, loading time difference, and response time difference to comprehensively evaluate and obtain the interaction decrease value.

[0090] In practical applications, phone themes impact phone performance, both in terms of key performance parameters and by causing inconvenience to user interaction. Some themes alter wallpapers and app icons, making it difficult for users to locate apps. The time a user spends clicking an app icon from the home screen can be used as the app location time. Furthermore, decreased phone performance can lead to sluggishness, including stuttering when scrolling and excessively long app loading times. By establishing a deep learning model, weights are assigned to location time difference, loading time difference, and response time to determine the interaction degradation value.

[0091] If usage issues are identified, real-time usage information of the user's phone theme will be collected. Based on this information, the user's acceptance of the usage issues will be determined. The specific steps are as follows:

[0092] Step S21: Collect real-time usage information of the user's topic. The real-time usage information includes topic adjustment information and topic usage duration.

[0093] Theme adjustment information refers to the user's adjustments to various parameters of the theme, including adjustments to brightness, font, and dynamic effects. Theme usage duration refers to the duration of use of the current theme.

[0094] Step S22: Collect the historical usage time of the user's theme, extract the average replacement time due to usage problems from the historical usage time and record it as the problem replacement time.

[0095] The system collects the user's historical usage time of different themes, calculates the issue value for each theme, sets a range for these issue values, and extracts the average replacement time for the themes within that range as the issue replacement time. For example, if a user has used 10 themes, and the issue values ​​for themes A, B, and C are within the issue value range, then the average usage time of themes A, B, and C is calculated as the issue replacement time. The usage time is the replacement time, and the issue value range can be set according to the user's situation. For example, the issue value range for themes with an average replacement time of 3 days can be used as the set issue value range, because the user's usage time is relatively short, indicating that the replacement is more likely due to theme incompatibility.

[0096] Step S23: Calculate the time difference between topic usage time and question replacement time, and determine whether the time difference reaches the preset time difference threshold.

[0097] The timeframe for resolving issues reflects a user's tolerance for problems. Therefore, the difference in timeframe can be used to determine whether a user accepts a problem. For example, if a user typically re-resolves a problematic theme after 3 days, and the theme has already been used for 30 days, then the user is considered to have accepted the problem; otherwise, they would have already replaced the theme.

[0098] Step S24: If the time difference threshold is reached, then it is determined that the user accepts the usage problem.

[0099] The duration difference threshold can be set based on the user's actual situation, such as the average time difference between users changing problematic topics. For example, if topic A is changed every 3 days, topic B every 6 days, and topic C every 10 days, then the duration difference threshold can be set to 4 days. This is because the maximum time difference for changing problematic topics is 4 days.

[0100] Step S25: If the duration difference threshold is not reached, determine whether the user has adjusted the phone theme parameters based on the theme adjustment information.

[0101] Step S26: If the phone theme parameters have not been adjusted, determine that the user accepts the usage problem; otherwise, determine that the user does not accept the usage problem.

[0102] In practical applications, if the preset time difference threshold has not been reached, it is difficult to determine whether the user accepts the issue based on the duration. If the user has adjusted the phone theme parameters, it indicates that the user feels the current phone theme has a problem, and therefore the user has improved the usability of the current phone theme by adjusting the phone theme parameters. If the user has not adjusted the phone theme parameters, it is assumed that the user accepts the current phone theme. Phone themes will have a certain impact on phone performance and user usage, but different users have different levels of acceptance of the impact. For example, if a phone theme is very power-consuming, user A may accept the power consumption, while user B may not. In this case, there is no usage problem for user A, but there is a usage problem for user B.

[0103] If the user does not accept the usage issues, the following steps are taken: Collecting the user's personal information and assessing their willingness to change the theme based on that information.

[0104] Step S31: Collect user personal information and analyze the user's personal information to obtain the user's tolerance level for mobile phone themes.

[0105] Step S32: Extract real-time difficulty information based on real-time usage information, determine whether the real-time difficulty information meets the tolerance standard, and if it does, calculate the difference between the real-time difficulty information and the tolerance standard as the user's theme change intention value.

[0106] Real-time usage information includes not only theme adjustment information and theme usage duration, but also various features corresponding to the tolerance difficulty standard. The real-time values ​​of these features combine to form real-time difficulty information. For example, this includes features selected from various phone performance and interaction features, such as power consumption and loading time, specifically derived from the problem features in the tolerance difficulty standard. Assuming the problem features are power consumption, display performance, and location software location, then the real-time difficulty information is the real-time data corresponding to power consumption, display performance, and location software location; in other words, the real-time difficulty information includes the real-time feature values ​​corresponding to the problem features.

[0107] Step S33: If the tolerance for difficulty standard is not met, the average time for users to change themes is collected and recorded as the daily change time.

[0108] The average time users spend changing themes here refers to the average usage time during normal use, excluding themes changed due to issues. This can be differentiated by time period; for example, themes changed within a week can be filtered out, and the average usage time of the remaining themes can be calculated as the daily change time.

[0109] Step S34: Calculate the time difference between topic usage time and daily replacement time, and combine it with real-time difficulty information to obtain the topic replacement willingness value.

[0110] In practical applications, it's clear that everyone's tolerance level and approach to the inconvenience caused by phone themes differ; in other words, different users have different tolerance standards for phone themes. Therefore, even with the same theme, different users will have different willingness to change it. Analyzing users' tolerance difficulty standards allows us to determine their willingness to change themes. When the real-time difficulty information reaches the user's tolerance difficulty standard, it's considered that the user's willingness to change the phone theme is relatively strong. Therefore, the larger the difference between the real-time difficulty information and the tolerance difficulty standard, the greater the willingness to change the theme. The difference between the real-time difficulty information and the tolerance difficulty standard is calculated by taking the difference between the real-time difficulty information and the tolerance difficulty standard, normalizing these differences, and then summing them up. For example, if the tolerance difficulty standard is 5% battery consumption / hour and an average loading time of 3 seconds, while the real-time difficulty information is 7% battery consumption / hour and an average loading time of 5 seconds, then we calculate 2% / hour and 2 seconds. These two data points are then normalized and summed to obtain the difference between the real-time difficulty information and the tolerance difficulty standard. If a user's real-time difficulty information does not meet the tolerance standard, it is assumed that the user's willingness to switch is not strong enough, and the willingness to switch is determined based on daily usage.

[0111] The steps for collecting user personal information and analyzing it to determine the user's tolerance level for mobile phone themes are as follows:

[0112] Step S311: Record the phone themes that were replaced due to usage issues as problem themes, and count the common features of problem themes to obtain multiple problem features.

[0113] The issue theme can be identified by the duration of the issue, and common characteristics of these themes can be statistically analyzed and used as issue features. If a user changes their phone theme because of an issue, it indicates that the issue is a fundamental problem that the user finds unacceptable.

[0114] Step S312: Calculate the worst value of each problem feature and obtain the basic tolerance standard based on the worst values ​​of all problem features.

[0115] The basic tolerance standard is formed based on the worst value of all problem characteristics, which is the user's bottom line of tolerance.

[0116] Step S313: Collect user personal information, including the user's proficiency in switching themes and proficiency in mobile applications.

[0117] Theme switching proficiency can be assessed by the average time users take to switch themes, or by the extent to which users have adjusted phone theme settings. For example, if a phone theme has 10 settings and a user has only adjusted 2, the coverage rate is 20%, which can be used as a measure of the user's theme switching proficiency. User proficiency with mobile applications refers to their familiarity with using the phone. This can be assessed by the average time users take to complete software operations. For example, if a user takes an average of 2 minutes to send a text message, placing them in just under 10% of the general user base, their mobile application proficiency can be assessed as 10%.

[0118] Step S314: Adjust the basic tolerance standard based on the user's personal information to obtain the user's tolerance difficulty standard for mobile phone themes.

[0119] In practical applications, a basic tolerance standard is formed based on the worst-case scenario of a user's perception of a particular theme, generally reflecting the user's acceptable tolerance threshold. However, in daily life, phone themes can become unbearable for users, and their desire to change them is strong. However, due to factors such as lack of experience or difficulty in changing themes, users may not make the change. Therefore, the user's tolerance standard needs further adjustment to obtain a more accurate tolerance difficulty standard. The adjustment percentage is determined based on the user's personal information. For example, a deep learning model can be used to assign weights to the user's proficiency in switching themes and their proficiency with the phone application, generating an adjustment percentage formula: Adjustment Percentage = 1 / User's Proficiency in Switching Themes × Corresponding Weight Ratio a + 1 / User's Proficiency with the Phone Application × Corresponding Weight Ratio b. The higher the user's proficiency in switching themes and their proficiency with the phone application, the smaller the adjustment percentage, because the user can switch themes skillfully, and the actual tolerance level is close to the basic tolerance standard. The tolerance difficulty standard is then lowered according to the adjustment percentage. For example, if the adjustment percentage is 10%, and the basic tolerance standard is 10% of battery consumption per hour, then lowering it by 10% would result in a tolerance difficulty standard of 9% per hour.

[0120] The steps for calculating the difference between topic usage time and daily topic replacement time, and combining this with real-time difficulty information to obtain the topic replacement intention value, are as follows:

[0121] Step S341: Collect the user's historical usage topics, filter and remove problem topics from the historical usage topics to obtain daily topics.

[0122] Step S342: Extract the average value of different daily features of daily topics, and form a daily standard based on the average value of all daily features.

[0123] Daily features can be extracted autonomously by deep learning models or set by users, including various features such as interaction features, mobile phone performance features, and visual features.

[0124] Step S343: Extract the real-time value of the feature corresponding to the daily feature based on the real-time difficulty information, calculate the average difference between the real-time feature value and the daily standard and record it as the real-time difficulty value.

[0125] Calculate the difference between each feature in the real-time hard information and the daily feature. A difference indicating a performance disadvantage compared to the daily feature is positive; otherwise, it is negative. For example, if the daily feature is power consumption of 1% / hour, and the real-time hard information shows power consumption of 2% / hour, then the difference is 1% / hour. If the real-time hard information shows power consumption of 0.5% / hour, then the difference is -0.5% / hour. Calculate the average of the differences between all real-time feature values ​​and the daily standard, normalize them, and use this average as the real-time hard value.

[0126] Step S344: Calculate the difference between the duration of topic usage and the duration of daily topic replacement, and combine it with the real-time difficulty value to obtain the topic replacement willingness value.

[0127] In practical applications, if the real-time difficulty information does not reach the tolerance standard, and the user can tolerate the current theme's problems, it indicates that the user's willingness to change the phone theme is not very strong and they are not affected by the problems. Therefore, in this case, a comparison with the characteristics of commonly used phone themes is used to confirm the user's willingness to change the theme. In other words, if the user is not concerned about the problems caused by the phone theme, then the evaluation is based on the theme's performance characteristics, such as visual appeal and brightness, and the user experience. The real-time difficulty value obtained by comparing with commonly used phone themes reflects the impact of the current phone theme. Secondly, different users change phone themes at different frequencies. Therefore, the closer the current theme's usage time is to the usual change time, the stronger the user's willingness to change the theme will be. For example, if User A changes their phone theme on average every month, it reflects that the user's weariness time for the theme is about one month. The closer the current phone theme's usage time is to one month, the more likely the user is to change it due to weariness, thus the greater the willingness to change the theme. The theme change willingness value obtained by combining market difference and real-time difficulty value can be obtained by allocating weights using a deep learning model.

[0128] The steps for collecting user mobile phone usage scenario information and evaluating the scenario switching intention value based on the scenario information are as follows:

[0129] Step S41: Collect real-time scene information of the user's mobile phone usage, and determine whether the user has spare time based on the real-time scene information.

[0130] To determine if a user has free time, one can observe their usage of different mobile applications. These applications can be categorized into work-related and non-work-related applications. Further categorization of work and non-work-related scenarios can then be used to determine if there is free time during non-work-related scenarios.

[0131] Step S42: If the user has spare time, obtain the user's theme change scene information, compare the similarity between the real-time scene information and the changed scene information, and use it as the scene change intention value.

[0132] The context information for changing themes includes details such as the time and location of the change. By statistically analyzing the context information of users changing themes, common features are extracted to form context information for changing themes. For example, user A's context information might be changing themes at their residence between 9 and 10 PM. Higher similarity indicates a greater willingness to change themes. Similarity can be evaluated using existing models such as cosine similarity.

[0133] Step S43: If the user has no free time, obtain the real-time functions of the user's mobile phone and determine the impact value of the mobile phone theme on the real-time functions.

[0134] Step S44: Collect the user's demand for real-time functions to obtain the demand value for real-time functions, and combine it with the impact value to obtain the willingness to change the scenario value.

[0135] In practical applications, if users lack spare time—that is, in work or event settings where they don't have free time to change their phone theme—their willingness to do so will be lower. However, it will also be influenced by the real-time functions the user needs to use their phone. For example, if a user needs to open a document using a certain software at work, but the document loads too slowly due to the phone theme, and the user is in a hurry to open the document—meaning the need is high—then the user will be more willing to spend a little time changing their phone theme to solve the immediate problem.

[0136] If the user does not have spare time, the steps to obtain the user's real-time mobile phone usage data and determine the impact of the mobile phone theme on real-time functionality are as follows:

[0137] Step S431: Obtain the real-time functions of the user's mobile phone and extract the evaluation features of the real-time functions.

[0138] Evaluation features refer to the features extracted from a function, that is, the features that users value most for that function. Because different functions have different requirements—including phone functions and software functions; for example, financial apps prioritize security, while communication apps prioritize message sending speed—extracting evaluation features helps to confirm the needs of real-time functions.

[0139] Step S432: Collect real-time feature information of the evaluation features corresponding to the real-time function, and evaluate the real-time performance value of the real-time function based on the real-time feature information.

[0140] The percentage of comments for each evaluation feature is used as the weight ratio. All weighted features and their corresponding weight ratios are multiplied and summed to establish a calculation formula, such as adjusting the percentage calculation formula, to calculate the real-time performance value.

[0141] Step S433: Collect the best performance value of the user's mobile phone using the real-time function, calculate the difference between the real-time performance value and the best performance value, and use it as the impact value of the mobile phone theme on the real-time function.

[0142] In practical applications, the greater the difference between the real-time performance value and the optimal performance value, the greater the impact of the phone theme on the current real-time function, and therefore the greater the impact value.

[0143] The steps to collect user feedback on the level of demand for real-time functionality and obtain the demand value for real-time functionality are as follows:

[0144] Step S441: Collect user operation information for real-time functions, and determine whether the user is completely waiting for the real-time functions to load based on the operation information.

[0145] Step S442: If the user does not wait for the real-time function to load completely, then the frequency of the user's repeated operations is counted as the demand value of the real-time function.

[0146] In practical applications, if users can simply wait for real-time functions to load (i.e., they don't repeat the operation), it indicates that the user's demand for that real-time function is low, at its minimum. Conversely, the more frequently users repeat the operation, the greater the demand for the real-time function, and thus the higher the demand value. For example, when using a banking app, if the page doesn't load, the user will refresh multiple times; the higher the refresh frequency, the more impatient the user is. If users don't repeat the operation, it means they are not in a hurry and can tolerate the impact of the theme on their current use. Therefore, in scenarios without spare time, their willingness to change their phone theme is low. However, if users are in a hurry, their willingness to change the theme in that scenario will increase.

[0147] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A deep learning-based method for optimizing mobile phone theme recommendations, characterized in that, Includes the following steps: Collect real-time usage characteristics of users' mobile phone themes, and determine whether there are any usage problems based on these characteristics; If there are usage issues, real-time usage information of the user's mobile theme will be collected, and the user's acceptance of the usage issues will be determined based on the real-time usage information. If the user does not accept the usage problem, the user's personal information will be collected, and the user's willingness to change the theme will be evaluated based on the user's personal information. Collect information on the scenarios in which users use their mobile phones, and evaluate the willingness to switch scenarios based on this information. A deep learning model for mobile theme recommendation is constructed by combining theme change intention value and scene change intention value; The success rate of mobile theme recommendation is evaluated using a deep learning model. When the success rate reaches the preset success rate standard, user theme preference data is collected, and mobile themes are recommended based on the theme preference data. The step of collecting real-time usage characteristics of users' mobile phone themes and determining whether there are usage problems based on these characteristics is as follows: Collect real-time usage characteristics of users using mobile phone themes, including real-time interaction characteristics and real-time mobile phone running status characteristics; Obtain the user's historical usage characteristics before using the mobile phone theme, including historical interaction characteristics and historical mobile phone running status characteristics; By comparing real-time and historical mobile phone operating status characteristics, the performance degradation value of the mobile phone operating performance can be confirmed. By comparing real-time interaction features with historical interaction features, the interaction performance degradation value was confirmed. By combining performance degradation and interaction degradation values, the problem value of the user's use of the mobile theme is determined. When the problem value reaches the preset problem value threshold, it is determined that the user has a problem using the mobile theme; otherwise, it is determined that the user does not have a problem using the mobile theme. If usage issues exist, the step of collecting real-time usage information of the user's mobile phone theme and determining whether the user accepts the usage issues based on the real-time usage information is as follows: Collect real-time usage information of the user's topic, including topic adjustment information and topic usage duration; Collect the historical usage time of themes used by users, extract the average replacement time of the phone theme due to usage issues from the historical usage time, and record it as the problem replacement time. Calculate the difference between the duration of topic usage and the duration of issue replacement, and determine whether the duration difference reaches the preset duration difference threshold; If the time difference threshold is reached, it is determined that the user accepts the usage problem; If the duration difference threshold is not reached, determine whether the user has adjusted the phone theme parameters based on the theme adjustment information. If the phone theme parameters have not been adjusted, it is determined that the user accepts the usage problem; otherwise, it is determined that the user does not accept the usage problem.

2. The mobile phone theme recommendation optimization method based on deep learning according to claim 1, characterized in that, The step of comparing real-time interaction features and historical interaction features to confirm the interaction performance degradation value is as follows: By comparing real-time interaction features with historical interaction features, the feature comparison results are obtained, and the positioning time difference of the user's location software position is extracted based on the feature comparison results. The loading time difference is extracted based on the feature comparison results, which represents the average software loading time. The response time difference of the user's page scrolling is extracted based on the feature comparison results; The interaction degradation value is obtained by comprehensively evaluating the positioning time difference, loading time difference, and response time difference.

3. The mobile phone theme recommendation optimization method based on deep learning according to claim 2, characterized in that, The step of collecting user personal information and assessing the user's willingness to change the theme based on that information, if the user does not accept the usage problem, is as follows: Collect users' personal information and analyze it to determine the user's tolerance level for mobile phone themes; Real-time difficulty information is extracted based on real-time usage information. It is then determined whether the real-time difficulty information meets the tolerance standard. If it does, the difference between the real-time difficulty information and the tolerance standard is calculated and used as the user's theme change intention value. If the tolerance for difficulty is not met, the average time it takes for users to change themes will be collected and recorded as the daily change time. The difference between the duration of topic usage and the duration of daily topic replacement is calculated, and the topic replacement willingness value is obtained by combining real-time difficulty information.

4. The mobile phone theme recommendation optimization method based on deep learning according to claim 3, characterized in that, The steps of collecting user personal information and analyzing it to determine the user's tolerance level for mobile phone themes are as follows: The phone themes that were replaced due to usage issues were recorded as problem themes. The common characteristics of the problem themes were analyzed to obtain multiple problem characteristics. Calculate the worst value for each problem feature, and derive the basic tolerance standard based on the worst values ​​of all problem features; Collect user personal information, including the user's proficiency in switching themes and proficiency in using mobile applications; The basic tolerance standard is adjusted based on the user's personal information to obtain the user's tolerance level for mobile phone themes.

5. The mobile phone theme recommendation optimization method based on deep learning according to claim 4, characterized in that, The step of calculating the difference between the usage time and the daily replacement time of a topic, and combining this with real-time difficulty information to obtain a topic replacement intention value, is as follows: Collect users' historical usage topics, filter and remove problematic topics from the historical usage topics to get the daily topics; Extract the average values ​​of different daily characteristics of daily topics, and form daily standards based on the average values ​​of all daily characteristics; Based on the real-time difficulty information, the real-time value of the feature corresponding to the daily feature is extracted, and the average difference between the real-time value of the feature and the daily standard is calculated and recorded as the real-time difficulty value. The difference between the duration of topic usage and the duration of daily topic replacement is calculated, and the topic replacement willingness value is obtained by combining the real-time difficulty value.

6. The mobile phone theme recommendation optimization method based on deep learning according to claim 5, characterized in that, The step of collecting user mobile phone usage scenario information and evaluating the scenario change intention value based on the scenario information is as follows: Collect real-time scene information of users using their mobile phones, and determine whether users have spare time based on the real-time scene information; If the user has spare time, obtain the user's theme-changing scenario information, compare the similarity between the real-time scenario information and the changed scenario information, and use it as the scenario-changing willingness value; If the user has no free time, obtain the real-time functions of the user's mobile phone and determine the impact value of the mobile phone theme on the real-time functions; Collect data on users' demand for real-time functions to obtain demand values ​​for real-time functions, and combine these with impact values ​​to obtain a willingness to change scenarios value.

7. The mobile phone theme recommendation optimization method based on deep learning according to claim 6, characterized in that, The step of obtaining the user's real-time mobile phone usage data and determining the impact of the mobile phone theme on real-time functionality if the user has no free time is as follows: Obtain real-time data on user phone usage and extract evaluation features of these real-time functions; Collect real-time feature information of the evaluation features corresponding to the real-time function, and evaluate the real-time performance value of the real-time function based on the real-time feature information; Collect the best performance value of the user's mobile phone when using real-time functions, calculate the difference between the real-time performance value and the best performance value, and use it as the impact value of the mobile phone theme on real-time functions.

8. The mobile phone theme recommendation optimization method based on deep learning according to claim 7, characterized in that, The step of collecting user demand for real-time functions and obtaining the demand value for real-time functions is as follows: Collect user operation information for real-time functions, and determine whether the user is completely waiting for the real-time functions to load based on the operation information; If users do not wait entirely for the real-time function to load, the frequency of repeated user actions is counted as the demand value for the real-time function.

Images