APP optimization method based on AI vertical model accurate putting and user behavior analysis

Abstract

The invention relates to the field of artificial intelligence technology and application, and aims to solve the problem of poor APP design optimization effect based on an AI technology in the prior art. In order to achieve the purpose, the invention provides an APP optimization method based on AI vertical model accurate putting and user behavior analysis, and the method comprises the steps: collecting user data and external data, and carrying out the processing of the user data and the external data; constructing an AI vertical model based on the processed data; integrating the constructed AI vertical model into a target delivery service; collecting behavior data of a user and inputting the behavior data into the AI vertical model; user behaviors are analyzed based on an AI vertical model; and carrying out design optimization on the APP based on a user behavior analysis result. According to the method, the design optimization of the APP can be more scientifically and efficiently carried out, the blind optimization cost is reduced, the product iteration speed is increased, the design optimization effect of the APP is improved, the APP can more flexibly respond to market changes, and the user stickiness is enhanced while the user experience is improved.

Description

Technical Field

[0001] The present invention relates to artificial intelligence technology and application fields, and specifically to an APP optimization method based on AI vertical model precise delivery and user behavior analysis. Background Art

[0002] With the rapid development of mobile Internet, smartphone applications (APPs) have become an indispensable part of people's daily lives. However, in the fierce market competition, how to make APPs stand out from many similar products and attract and retain users has become a major challenge for developers and operators. Traditional APP optimization methods are mostly based on experience judgment or simple data analysis, which makes it difficult to accurately capture user needs and behavior patterns, resulting in limited optimization effects.

[0003] In recent years, the rise of artificial intelligence (AI) technology has provided new ideas and methods for APP optimization. AI technology can process and analyze massive amounts of data, dig out the deep laws and potential needs behind user behaviors, and provide strong support for the precise delivery and design optimization of APPs. However, there are still some problems with the existing AI technology when applied to APP optimization: on the one hand, general AI models are often difficult to adapt to the specific needs of different industries and scenarios, resulting in limited accuracy and practicality of analysis results; on the other hand, the integration of AI models with business systems is not high, making it difficult to achieve real-time data updates and dynamic adjustments of models, affecting the sustainability and stability of optimization effects.

[0004] In view of this, this field needs an APP optimization method based on AI vertical model precise delivery and user behavior analysis to solve the above problems. Summary of the invention

[0005] In order to solve the above technical problems, that is, to solve the problem that the design optimization effect of APP based on AI technology in the prior art is not good.

[0006] The present invention provides an APP optimization method based on AI vertical model precise delivery and user behavior analysis, the method comprising:

[0007] S1: Collecting user data and external data and processing the user data and the external data;

[0008] S2: Build AI vertical models based on processed data;

[0009] S3: Integrate the constructed AI vertical model into the target delivery business;

[0010] S4: Collecting user behavior data and inputting it into the AI ​​vertical model;

[0011] S5: Analyze user behavior based on the AI ​​vertical model;

[0012] S6: Optimize the design of the APP based on the results of user behavior analysis.

[0013] In certain preferred embodiments, the model structure of the AI ​​vertical model includes an input layer, a hidden layer, and an output layer, wherein the input layer is used to receive processed user data and external data features, the number of the hidden layers is multiple, the hidden layer is used to perform nonlinear transformation and feature extraction on the data, and the output layer is used to output the user behavior prediction result;

[0014] The formula of the AI ​​vertical model is:

[0015]

[0016] Among them, Y is the output result, and the input data of the AI ​​vertical model is x i is the i-th eigenvector, n1 is the number of eigenvectors, W1 is the weight matrix connecting the hidden layer and the output layer, W2 is the weight matrix connecting the intermediate hidden layer, is the connection weight between the i-th feature of the input layer and all neurons in the first hidden layer, b1 is the bias term of the output layer, b2 is the bias term of the middle hidden layer, b3 is the bias term of the first hidden layer, g, h and f are activation functions respectively.

[0017] In certain preferred embodiments, the user data includes at least one of user personal information data, user behavior data, and user preference data; and the external data includes at least one of market data, social media data, and technical data.

[0018] In certain preferred embodiments, the user personal information data includes at least one of basic information data, geographic location information data, identity information data and account information data; the user behavior data includes at least one of APP usage behavior data and device usage behavior data; the user preference data includes at least one of content preference data, function preference data and interaction preference data; the market data includes at least one of industry dynamics data, market size and share data and price trend data; the social media data includes at least one of user social relationship data, social behavior data and social influence data; the technical data includes at least one of device performance data, network technology data and APP technical indicator data.

[0019] In some preferred implementations, step S4 also includes processing the collected user behavior data:

[0020] Performing data cleaning on the user's behavior data, wherein the data cleaning includes the steps of removing duplicate data, processing missing values, and correcting erroneous data;

[0021] Performing data conversion on the cleaned data, wherein the data conversion includes the steps of standardizing the data and classifying and coding the data;

[0022] Perform data integration on the data after data conversion.

[0023] In some preferred implementations, in the step of performing data cleaning on the user's behavior data, the formula for processing missing values ​​is:

[0024]

[0025] in, is the numerical feature z of the original user behavior data i The filling value of the missing value, n2 is the number of non-missing values, z ij is the i-th eigenvalue of the j-th non-missing value sample.

[0026] In some preferred embodiments, in the step of performing data conversion on the cleaned data, the formula for normalizing the data is:

[0027]

[0028] Among them, μ i is the mean of the ith feature, σ i is the standard deviation of the ith feature, n3 is the sample size.

[0029] In certain preferred embodiments, the user's behavior data includes at least one of login frequency and time, browsing history, operation history, purchase behavior, device model, network environment, and APP installation and uninstallation.

[0030] In certain preferred embodiments, in step S5, analyzing the user behavior includes at least one of user behavior prediction analysis, user behavior pattern recognition analysis, and behavior association analysis.

[0031] In certain preferred embodiments, in step S6, the design optimization of the APP includes at least one of APP interface optimization, APP operation optimization and APP performance optimization.

[0032] From the above, it can be seen that the APP optimization method based on AI vertical model precise delivery and user behavior analysis provided by the present invention has the following beneficial technical effects:

[0033] The APP optimization method based on AI vertical model precise delivery and user behavior analysis provided by the present invention constructs a highly adaptable AI vertical model by systematically collecting and processing user data and external data. The model can deeply understand user behavior characteristics and demand preferences. After integrating this model into the target delivery business, it can realize real-time capture and intelligent analysis of user behavior data, and then accurately guide the content push, function optimization and interface design of the APP. This process not only significantly improves the personalization of user experience and enhances user stickiness, but also enables the APP to respond to market changes more flexibly, effectively improving delivery efficiency and conversion rate. At the same time, based on the in-depth analysis results of the AI ​​vertical model, the design optimization of the APP can be carried out more scientifically and efficiently, which not only reduces the cost of blind optimization, but also speeds up the product iteration speed and improves the design optimization effect of the APP. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0035] Figure 1 This is a flow chart of the APP optimization method based on AI vertical model precise delivery and user behavior analysis of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] Based on the problem pointed out in the background technology that the design optimization effect of APP based on AI technology in the prior art is not good. The present invention provides an APP optimization method based on AI vertical model precision delivery and user behavior analysis, which aims to make the design optimization of APP more scientific and efficient, reduce the cost of blind optimization, speed up product iteration, improve the design optimization effect of APP, make it more flexible to respond to market changes, and enhance user stickiness while improving user experience.

[0038] like Figure 1 As shown, the APP optimization method based on AI vertical model precise delivery and user behavior analysis of the present invention includes:

[0039] S1: collect user data and external data and process them;

[0040] In a preferred scenario, user data includes at least one of user personal information data, user behavior data, and user preference data; external data includes at least one of market data, social media data, and technical data. More preferably, user personal information data includes at least one of basic information data (such as name, age, gender, date of birth, contact information, etc.), geographic location information data (such as permanent address, work address, real-time positioning information, etc.), identity information data (such as ID card number, occupation, education, marital status, family status, etc.) and account information data (such as registration time, account ID, login password, bound third-party account, etc.); user behavior data includes at least one of APP usage behavior data (such as login frequency and time, browsing history, operation history, purchase behavior, etc.) and device usage behavior data (such as device model, network environment, application installation and uninstallation, etc.); user preference data includes content preference data (such as topics of interest, preferred media types, preferred brands or product types, etc.), function preference data (such as commonly used APP functions, interface design preferences, etc.) and interaction preference data The market data includes at least one of industry dynamics data (such as competitor information, industry development trends, etc.), market size and share data (such as the size of the target market, the market share of its own APP, etc.) and price trend data (such as the price range of similar products or services, price fluctuation factors, etc.); social media data includes at least one of user social relationship data (such as friend lists and social groups, etc.), social behavior data (such as published content, interactive behavior, etc.) and social influence data (such as the number and quality of fans, content dissemination scope, etc.); technical data includes at least one of device performance data (such as processor performance, memory capacity and performance, storage capacity and type, etc.), network technology data (such as network bandwidth, network delay, network protocol, etc.) and APP technical indicator data (such as version information, performance indicators and compatibility data, etc.).

[0041] S2: Build AI vertical models based on processed data;

[0042] In some preferred cases, the model structure of the AI ​​vertical model includes an input layer, a hidden layer and an output layer, the input layer is used to receive processed user data and external data features, the number of hidden layers is multiple, the hidden layer is used to perform nonlinear transformation and feature extraction on the data, and the output layer is used to output the user behavior prediction results;

[0043] The formula for the AI ​​vertical model is:

[0044]

[0045] Among them, Y is the output result, and the input data of the AI ​​vertical model is x i is the i-th eigenvector, n1 is the number of eigenvectors, W1 is the weight matrix connecting the hidden layer and the output layer, W2 is the weight matrix connecting the intermediate hidden layer, is the connection weight between the i-th feature of the input layer and all neurons in the first hidden layer, b1 is the bias term of the output layer, b2 is the bias term of the middle hidden layer, b3 is the bias term of the first hidden layer, g, h and f are activation functions respectively;

[0046] Specifically, W3 is a p×l-dimensional matrix, where p is the dimension of the feature vector of the input layer, l is the number of neurons in the first hidden layer, can be viewed as a row vector of length l, which represents the connection weights between the i-th feature in the input layer and all l neurons in the first hidden layer. In other words, Each element of (where j = 1, 2, ..., l) represents the connection weight between the i-th feature of the input layer and the j-th neuron of the first hidden layer;

[0047] In the above, the formula of the activation function h is:

[0048] in,

[0049] The formula for the activation function g is:

[0050] g(y2)=max(0,y2), where

[0051] The formula for the activation function f is:

[0052] in,

[0053] S3: Integrate the constructed AI vertical model into the target delivery business;

[0054] Preferably, a RESTful API interface is used to ensure that the AI ​​vertical model can be seamlessly connected to the business system. The interface has stability and efficiency in data transmission, can receive data sent by the business system, and return the output results of the model to the business system, and define clear request and response formats, including data input formats (such as user data and external data in JSON format) and output formats (such as user behavior results predicted by the model or recommended content). In addition, a data transmission channel is established to transmit relevant data in the target delivery business to the AI ​​vertical model for processing. During the data transmission process, the security and integrity of the data are ensured, and encrypted transmission technology and data verification mechanisms are used to pre-process the data transmitted to the model to make it meet the input requirements of the model, such as data format conversion, feature extraction, etc. At the same time, the results of the model output are post-processed and converted into a form that the business system can understand and apply, such as converting user behavior prediction results into specific delivery strategies or recommended content.

[0055] S4: Collect user behavior data and input it into the AI ​​vertical model;

[0056] In some preferred cases, the above step S4 includes:

[0057] S41: Collecting user behavior data, wherein the user behavior data includes at least one of login frequency and time, browsing history, operation history, purchase behavior, device model, network environment, and APP installation and uninstallation; the device model, network environment and other information can be obtained through the system interface or third-party tools;

[0058] S42: Processing the collected user behavior data:

[0059] S421: performing data cleaning on the user's behavior data, wherein the data cleaning includes the steps of removing duplicate data, processing missing values, and correcting erroneous data;

[0060] In the step of removing duplicate data, duplicate data records are identified and deleted according to the unique identifier of the data (such as the timestamp and operation type combination of the user operation record) to ensure the uniqueness of the data;

[0061] The formula for handling missing values ​​is:

[0062]

[0063] in, is the numerical feature z of the original user behavior data i The filling value of the missing value, n2 is the number of non-missing values, z ijis the i-th eigenvalue of the j-th non-missing value sample; that is, for the missing value of the numerical feature, according to the formula for processing missing values Filling

[0064] In the step of correcting erroneous data, the rationality and legality of the data are checked through data validation rules. For example, for the amount field in the purchase behavior data, check whether it is a positive number and within a reasonable range, and correct or mark the abnormal data.

[0065] S422: performing data conversion on the cleaned data, wherein the data conversion includes the steps of standardizing the data and classifying and coding the data;

[0066] The formula for normalizing the data is:

[0067]

[0068] Among them, μ i is the mean of the ith feature, σ i is the standard deviation of the ith feature, n3 is the number of samples;

[0069] In the step of classifying and encoding data, text or discrete data is classified and encoded, such as encoding the user's operation type (such as click, slide, long press, etc.) and one-hot encoding the network environment (such as 4G, WiFi, etc.) for model processing;

[0070] S423: Perform data integration on the converted data; for example, the cleaned and converted data can be integrated with historical data to form a complete user behavior data set, ensuring the consistency and integrity of the data and providing a reliable data foundation for subsequent data analysis and model input.

[0071] S43: Input the processed data into the AI ​​vertical model. For example, the pre-processed user behavior data is organized and formatted according to the input requirements of the AI ​​vertical model, and then input into the model for analysis and prediction. Specifically, if the input requirements of the model are feature data in a two-dimensional array format and target data in a one-dimensional array format, the user behavior data is sorted into the corresponding format and then input into the model. During the data input process, the order and dimension of the data are matched to ensure that the model can correctly receive and process the data.

[0072] S5: Analyze user behavior based on AI vertical model;

[0073] Preferably, analyzing user behavior includes at least one of user behavior prediction analysis, user behavior pattern recognition analysis, and behavior association analysis.

[0074] S6: Optimize the design of the APP based on the results of user behavior analysis;

[0075] Preferably, the design optimization of the APP includes at least one of APP interface optimization (such as layout adjustment and / or visual design improvement), APP operation optimization (such as process simplification and / or function improvement) and APP performance optimization (such as loading speed improvement and / or stability improvement).

[0076] The following is an example of how to optimize the interface layout of a game app, assuming that the app is a game app:

[0077] Collect user data and external data. For user personal information data: collect basic information such as age, gender, registration region, etc. of game users, and obtain game users' preferences for game types (such as action, strategy, role-playing, etc.) through the preference options set in the game. At the same time, the account creation time and the commonly used login device information (device model, operating system version) are recorded; for user behavior data, it includes APP usage behavior data and device usage behavior data respectively. For APP usage behavior data, the game user's login time, duration, and frequency are recorded, and the stay time and switching frequency in different game scenes (main interface, level, mall, etc.) are tracked, as well as operation records (such as clicks, task execution, etc.) and purchase behavior (prop purchase details) are tracked. For device usage behavior data, device hardware information (CPU, memory, graphics card, etc.), screen resolution, brightness settings and network connection status (network type, delay, packet loss rate) are collected; for user preference data, it specifically includes content preference data analysis (game users' preference for character professions, appearance styles, game maps, level difficulty, plot, etc.), function preference data (observing the game users' frequency and dependence on social functions, rankings, achievement systems, in-game activities, etc.) and interaction preference data (recording game users' operating habits on the game interface and setting preferences for sound effects, vibration feedback, etc.); for external data, it includes market data, social media data and technical data; for market data :It includes industry dynamic data (game industry media, forums, etc., collecting information on changes in policies and regulations, technological innovation trends, competitors' new games and marketing strategies, etc.), market size and share data (obtaining the overall size of the game market, market segmentation, and the market share and ranking of this game), price trend data (monitoring the pricing of props, recharge discounts and virtual currency exchange ratios of similar types of games, as well as the price trends of game peripheral products); for social media data, it includes user social relationship data (obtaining the friend relationship network and related information of game users on social media platforms under authorization), social behavior data (recording the interactive behavior of game users under the official social media account of the game and the chat, team formation and other behaviors in the social functions within the game), social influence data (evaluating the game users' game-related influence on social media); for technical data, it includes device performance data (continuously collecting hardware performance parameters and technical specifications of mainstream game devices), network technology data (network technology development and network coverage, speed, stability data of different network operators) and APP technical indicator data (regularly monitoring the version number, update frequency, crash rate, loading time, frame rate, etc. of the game APP).

[0078] Process the data, including data cleaning, data conversion and data integration, as follows:

[0079] For data cleaning: Remove duplicate data: Check the game users' login, operation and other records, remove duplicate entries, and ensure data uniqueness; Handle missing values: For numerical data (such as purchase amount, game time), use the mean, median or fill according to trend forecasts; Categorical data is filled with common categories or marked as unknown, pending verification; Correct erroneous data: Establish rules to check the legitimacy of data, such as age range, device information format, etc., and correct or delete abnormal data;

[0080] For data conversion: Standardization: standardize numerical data (such as game time, purchase amount, etc.), and use the Zscore method to make it have a distribution feature with a mean of 0 and a standard deviation of 1; Classification and encoding: classify and encode text or discrete data (such as gender, character occupation, etc.), such as converting gender into 0 and 1 encoding, and performing one-hot encoding on character occupation;

[0081] For data integration: Integrate the cleaned and converted data, establish a unified data storage system, solve problems such as inconsistent data formats and inconsistent field names among different data sources, and achieve seamless integration and efficient access to data.

[0082] The constructed AI vertical model structure includes input layer, hidden layer and output layer:

[0083] Input layer: Determine the input layer nodes according to data features and analysis objectives; for example, use the game user's age, gender, game time, purchase frequency, device performance, network status, etc. as input features, each feature corresponds to a node, and the classification features are encoded as multiple nodes; perform preprocessing such as normalization on the input data;

[0084] Hidden layer: Set multiple hidden layers. The number and nodes can be adjusted according to data complexity and performance requirements. For example, try 2-3 hidden layers with 32-64 nodes in each hidden layer. Use activation functions such as ReLU to increase nonlinear expression capabilities.

[0085] Output layer: Determine the output layer nodes according to the analysis task; for example, to predict the game user's satisfaction with the interface layout, set a node output probability; to classify game users, set the number of nodes to be equal to the number of categories, and output the probability distribution of the corresponding category;

[0086] To train the model, select the stochastic gradient descent method or its variant (such as Adam, etc.) optimization algorithm, determine the learning rate (such as starting from 0.01 and adjusting according to the convergence situation), batch size (such as 32, 64 or 128, etc.), number of training rounds (determined by observing the performance of the validation set) and other parameters, divide the data into training set, validation set and test set, use the training set data for iterative training, update the model parameters according to the error back propagation, regularly use the validation set to evaluate and adjust the hyperparameters, and stop training when the model performance on the validation set is stable or no longer improves to prevent overfitting; finally, evaluate and optimize the model, select appropriate indicators according to the task, such as accuracy, recall rate, F1 value, etc. for classification tasks, mean square error, mean absolute error, etc. for regression tasks, and other indicators such as AUC value and logarithmic loss can also be selected according to the situation; Model evaluation: Use the test set to evaluate the model, Compare the predicted results with the true labels to calculate the evaluation index values, such as the accuracy of binary classification problems, to fully understand the model performance; Model tuning: Model structure adjustment: Try to increase or decrease the number of hidden layers and nodes to balance model complexity and performance; Hyperparameter optimization: Use grid search, random search or Bayesian optimization to adjust hyperparameters such as learning rate, batch size, regularization parameters; Feature engineering optimization: Consider adding new features (such as composite features derived from behavioral data) or transforming existing features, and delete redundant or irrelevant features; Model integration: Bagging, Boosting and other integration technologies can be used, such as using the random forest algorithm to improve the generalization ability and stability of the model; Data enhancement: Data enhancement can be performed when the amount of data is small, such as rotating and scaling image data, and performing word replacement on text data to increase data diversity and model robustness.

[0087] Integrate the constructed AI vertical model into the target delivery business, use RESTful API interface to communicate through HTTP protocol, the interface request contains game user data and business parameters (such as game user ID, game status, analysis task type, etc.), and the response returns the model prediction results (such as satisfaction score, interface layout suggestions, game user classification labels, etc.) in JSON format, establish a secure data transmission channel (such as SSL / TLS encryption), preprocess and encapsulate data on the game APP side, decapsulate and further preprocess on the model side to make it meet the input requirements, post-process the output results and convert them into a form that can be used by the game APP (such as converting interface layout suggestions into coordinates, dimensions and other parameters), conduct tests on interface functions, data transmission accuracy, model performance in business scenarios, simulate game scenarios to send a large amount of test data, verify interface stability and response speed, ensure that data is processed correctly and model prediction results meet expectations, debug and repair problems in a timely manner when problems are found, and optimize system performance.

[0088] Collect user behavior data and input it into the AI ​​vertical model, embed a data collection module in the game APP, record the game user's login time, operation, interface interaction, network connection status and other information in real time, use device sensors to obtain additional information (such as acceleration, gyroscope data), ensure timely and accurate data collection, use cache and upload mechanisms to ensure data integrity, and regularly test and optimize the data collection module; data preprocessing: data cleaning: remove duplicate data: check for duplicates based on unique identifiers such as timestamps and operation type combinations, and remove duplicate operation records in a short period of time; handle missing values: fill in missing values ​​of numerical features based on historical game user data or similar game user situations, supplement missing values ​​of categorical features or mark them as unknown, and fill missing values ​​of time features with reasonable speculation; correct erroneous data: establish validation rules to check data According to the legitimacy, such as checking the game score range, operation position data, etc., and reviewing and processing abnormal data; data conversion: standardization processing: standardize numerical data (such as game time, operation frequency, etc.) to eliminate dimension and numerical range differences; classification and coding: classify and code text or discrete data (such as character occupation, game scene, etc.), such as character occupation unique hot coding, game scene number coding; data integration: integrate the cleaned and converted data with historical data to ensure consistency and integrity, handle data association and connection, and deduplicate and integrate data; organize and format the pre-processed data according to the model input requirements, such as combining game user features into vectors, converting time series data into appropriate feature representations, and after checking that the data is correct, input the data into the model through the interface function to trigger analysis and prediction.

[0089] Analyze user behavior based on AI vertical models. For user behavior prediction analysis, determine the predicted behavior, such as whether the game user will purchase props, whether they can pass the level, whether they will churn, etc., and clarify the prediction task definition and label (such as marking the purchase behavior as "purchase" or "no purchase"); input the game user's historical behavior data and features into the model, and calculate the prediction probability of different behaviors, such as predicting the probability of purchasing props based on purchase records, consumption habits, etc.; result evaluation and application: evaluate the prediction results, classify and sort the game users according to probability, push recommendations and discounts to users with a high probability of purchasing games, compare the prediction results and actual behavior to calculate evaluation indicators (such as accuracy, etc.) to optimize the model and improve the prediction accuracy; for user behavior pattern recognition analysis, use clustering algorithms (such as K-Means, hierarchical clustering) to cluster game user behavior data, and group game users with similar behaviors according to characteristics such as game duration and operation frequency; pattern feature extraction: analyze the behavior of each cluster group Data, extract pattern features, such as role division preferences and collaboration strategies of team-based game users, and coping strategies and game progress advancement methods of single-player challenge game users; pattern application and understanding: provide strategies for game design and operation based on the results of behavioral pattern recognition, such as designing team functions for team-based game users, adding single-player levels and personalized systems for single-player challenge game users, and monitoring pattern changes to adjust the game; for behavioral association analysis, use association rule mining algorithms (such as Apriori and FP-Growth) to find out the association between game user behaviors, such as the association between purchasing equipment and participating in activities; rule evaluation and screening: evaluate and screen rules based on indicators such as support and confidence, and determine meaningful association rules, such as setting thresholds to screen out rules that meet the conditions; strategy formulation and application: formulate operation strategies based on association rules, such as pushing relevant activity information when purchasing props, designing guidance mechanisms based on task and exploration associations, and continuously monitoring and adjusting strategies.

[0090] Based on the results of user behavior analysis, the game APP interface is optimized. For example, the layout is adjusted to highlight the core functions and content: according to the behavior analysis, the common functions and content of game users are determined, such as "start the game", "task list", "mall", etc., and they are placed in a prominent position and optimized in design. Dynamic display areas are set up for popular activities and limited-time tasks. In the level interface, the character information, skill shortcut keys, etc. are reasonably laid out to highlight important prompt information; the setting pages and complex interfaces are simplified, such as the use of folding menus and tab classification display functions to reduce information density, optimize the character creation process, provide default templates and visual previews, and reduce operation steps and inputs; another example is the improvement of visual design: color matching optimization: adjust the color matching based on user preferences and behavior analysis. If the game users prefer bright and vivid colors, increase the saturation and brightness of the main color to create a lively atmosphere; adopt differentiated color strategies in different scenes, such as using warning colors for battle scenes and soft colors for leisure scenes; font and icon design: choose the appropriate font style and size to ensure readability and beauty Improve the visual quality of the game by highlighting important information and displaying general text in a moderate manner; optimize icon design to make it concise, clear and recognizable, and add animation effects and interactive feedback to commonly used icons; special effects and animation design: add special effects and animations according to the game type and game user preferences, such as transition animations, task completion and upgrade special effects, etc., to enhance visual effects and appeal; pay attention to controlling the frequency and intensity of use to avoid affecting performance, provide special effect setting options, optimize loading and playback, and reduce freezes and delays; for example, optimize interactivity: simplify the operation process: analyze the operation process and simplify tedious steps, such as one-click task collection and automatic submission, optimize shopping carts and payment processes, provide automatic combat and resource collection auxiliary functions, etc., to improve efficiency and fun and reduce repetitive work; enhance the feedback mechanism: establish a timely and effective feedback mechanism, provide visual and auditory feedback (such as button effects and sound effects) during operation, and display operation results and status changes (such as health points, experience value updates, etc.) in real time; provide detailed feedback and prompts for important operations and behaviors (such as upgrades to display attribute improvements and new skills unlocked).

[0091] In the above, when performing user behavior pattern recognition and cluster analysis, the similarity of behaviors between users can be measured. For example, cosine similarity can be used to calculate the similarity between user behavior feature vectors. Assume that the behavior feature vectors of user i and user j are and The cosine similarity calculation formula between them is:

[0092]

[0093] Among them, θ ij is the angle between the two vectors, n4 is the number of behavioral feature vectors, v ik is the value of user i on the kth feature, vjk is the value of user j on the kth feature. The closer the cosine value is to 1, the more similar the behaviors of the two users are; the closer the cosine value is to 0, the greater the difference in the behaviors of the two users is. By calculating the cosine similarity between users, users with similar behaviors can be grouped together, thereby better understanding the behavior patterns and characteristics of user groups and providing a basis for personalized recommendations and game operation strategy formulation.

[0094] It should be noted that the method of the present invention can be performed by a single device, such as a computer or a server. The method of the present invention can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the present invention, and the multiple devices will interact with each other to complete the described method.

[0095] It should be noted that the above specific embodiments of the present invention are described. Other embodiments are within the scope of the present invention. In some cases, the actions or steps recorded in the present invention can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0096] It should be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0097] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0098] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0099] While the invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications and variations of these embodiments will be apparent to those skilled in the art in light of the foregoing description.

[0100] One or more embodiments of the present invention are intended to cover all such replacements, modifications and variations that fall within the broad scope of the present invention. Therefore, any omissions, modifications, equivalent replacements, improvements, etc. made within the spirit and principles of one or more embodiments of the present invention should be included in the protection scope of the present invention.

Claims

1. An APP optimization method based on AI vertical model precision delivery and user behavior analysis, characterized in that: The method comprises: S1: Collecting user data and external data and processing the user data and the external data; S2: Build AI vertical models based on processed data; S3: Integrate the constructed AI vertical model into the target delivery business; S4: Collecting user behavior data and inputting it into the AI ​​vertical model; S5: Analyze user behavior based on the AI ​​vertical model; S6: Optimize the design of the APP based on the results of user behavior analysis.

2. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 1 is characterized in that: The model structure of the AI ​​vertical model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to receive processed user data and external data features, the number of the hidden layers is multiple, the hidden layer is used to perform nonlinear transformation and feature extraction on the data, and the output layer is used to output the user behavior prediction result; The formula of the AI ​​vertical model is: Among them, Y is the output result, and the input data of the AI ​​vertical model is x i is the i-th eigenvector, n1 is the number of eigenvectors, W1 is the weight matrix connecting the hidden layer and the output layer, W2 is the weight matrix connecting the intermediate hidden layer, is the connection weight between the i-th feature of the input layer and all neurons in the first hidden layer, b1 is the bias term of the output layer, b2 is the bias term of the middle hidden layer, b3 is the bias term of the first hidden layer, g, h and f are activation functions respectively.

3. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 2 is characterized in that: The user data includes at least one of user personal information data, user behavior data and user preference data; the external data includes at least one of market data, social media data and technical data.

4. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 3 is characterized in that: The user personal information data includes at least one of basic information data, geographic location information data, identity information data and account information data; the user behavior data includes at least one of APP usage behavior data and device usage behavior data; the user preference data includes at least one of content preference data, function preference data and interaction preference data; the market data includes at least one of industry dynamics data, market size and share data and price trend data; The social media data includes at least one of user social relationship data, social behavior data and social influence data, and the technical data includes at least one of device performance data, network technology data and APP technical indicator data.

5. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 2 is characterized in that: Step S4 also includes processing the collected user behavior data: Performing data cleaning on the user's behavior data, wherein the data cleaning includes the steps of removing duplicate data, processing missing values, and correcting erroneous data; Performing data conversion on the cleaned data, wherein the data conversion includes the steps of standardizing the data and classifying and coding the data; Perform data integration on the data after data conversion.

6. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 5 is characterized in that: In the step of performing data cleaning on the user's behavior data, the formula for processing missing values ​​is: in, is the numerical feature z of the original user behavior data i The filling value of the missing value, n2 is the number of non-missing values, z ij is the i-th eigenvalue of the j-th non-missing value sample.

7. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 6 is characterized in that: In the step of converting the cleaned data, the formula for standardizing the data is: Among them, μ i is the mean of the ith feature, σ i is the standard deviation of the ith feature, n3 is the sample size.

8. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 5 is characterized in that: The user's behavior data includes at least one of login frequency and time, browsing history, operation history, purchase behavior, device model, network environment, and APP installation and uninstallation.

9. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 1 is characterized in that: In step S5, analyzing the user behavior includes at least one of user behavior prediction analysis, user behavior pattern recognition analysis, and behavior association analysis.

10. The APP optimization method based on AI vertical model precise delivery and user behavior analysis according to claim 1 is characterized in that: In step S6, the design optimization of the APP includes at least one of APP interface optimization, APP operation optimization and APP performance optimization.

Images