Mail opening rate optimization method, system and equipment based on artificial intelligence and medium

Abstract

The invention provides a mail opening rate optimization method, system and device based on artificial intelligence and a medium, and the method specifically comprises the steps: obtaining player attribute data, carrying out class group division on a plurality of players according to the player attribute data, and determining a corresponding mail content type for the players of each class group; acquiring historical active time data of players, and determining active probability distribution data of each player in a future preset time period through a time sequence prediction algorithm according to the historical active time data of the players; obtaining social text data of each player, and determining specific mail content of each player according to the social text data; and determining an optimal mail decision of each player through a multi-target optimization algorithm based on the mail content type, the active probability distribution data and the specific mail content by taking a mail opening rate as an optimization target. According to the invention, personalized customization of the mail content and the sending opportunity is realized, and the mail opening rate and the conversion effect are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based email open rate optimization method, system, device and medium. Background Art

[0002] With the booming online gaming industry, gaming companies are facing an increasingly competitive environment. In order to maintain the activity and loyalty of players, gaming companies generally use email as an important channel to communicate with players, and enhance players' gaming experience and participation by sending various notifications, event information, and reward collection. However, with the surge in the number of emails and the lack of personalization and accuracy in email content, the open rate of players' emails is generally low, which has become a major challenge for gaming companies.

[0003] Most existing email sending strategies are based on static rules or simple user classification, lacking in-depth understanding and accurate prediction of players' personalized preferences. This "one-size-fits-all" email sending method often results in a mismatch between email content and players' actual needs, reducing the appeal and open rate of emails. In addition, email titles and content creativity often rely on manual writing, which is not only inefficient, but also difficult to achieve large-scale personalized customization, further limiting the relevance and effectiveness of emails. Specifically, there are the following deficiencies:

[0004] 1. Lack of personalization: The email content is not accurately customized according to the player’s personalized preferences, resulting in the email content being out of touch with the player’s interests, reducing the email’s appeal and opening rate.

[0005] 2. Insufficient prediction ability: Lack of accurate prediction of players’ future active time leads to unreasonable timing of email sending and missed the best time to interact with players.

[0006] 3. Creativity relies on manual labor: The creative writing of email titles and content is highly dependent on manual labor, which is difficult to achieve automation and large-scale production, limiting the diversity and innovation of emails.

[0007] 4. Single optimization method: The optimization of email sending strategy is often limited to a single dimension, such as content optimization or time optimization, and lacks a multi-objective optimization method that comprehensively considers multiple factors. Summary of the invention

[0008] The purpose of the present invention is to provide an email open rate optimization method, system, device and medium based on artificial intelligence, which realizes the personalized customization of email content and sending timing, effectively improves email open rate and conversion effect, and provides accurate user reach solution for game operation to solve at least one of the above-mentioned prior art problems.

[0009] In a first aspect, the present invention provides an email open rate optimization method based on artificial intelligence, and the method specifically comprises:

[0010] Acquire player attribute data, classify multiple players into groups according to the player attribute data, and determine a corresponding email content type for each group of players;

[0011] Obtain the player's historical active time data, and determine the activity probability distribution data of each player in a preset time period in the future through a time series prediction algorithm based on the player's historical active time data;

[0012] Obtaining social text data of each player, and determining specific email content of each player based on the social text data;

[0013] Taking the email opening rate as the optimization goal, based on the email content type, the active probability distribution data and the specific email content, the best email decision for each player is determined through a multi-objective optimization algorithm.

[0014] In a second aspect, the present invention provides an email open rate optimization system based on artificial intelligence, and the system specifically includes:

[0015] A first optimization module, configured to obtain player attribute data, classify multiple players into groups according to the player attribute data, and determine a corresponding email content type for each group of players;

[0016] The second optimization module is used to obtain the player's historical active time data, and determine the activity probability distribution data of each player in a future preset time period through a time series prediction algorithm based on the player's historical active time data;

[0017] A third optimization module is used to obtain social text data of each player and determine the specific email content of each player according to the social text data;

[0018] The fourth optimization module is used to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the active probability distribution data and the specific email content, with the email opening rate as the optimization target.

[0019] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, the artificial intelligence-based email opening rate optimization method as described in any one of the above methods is implemented.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for optimizing email opening rate based on artificial intelligence as described in any one of the above methods is implemented.

[0021] Compared with the prior art, the present invention has at least one of the following technical effects:

[0022] 1. The present invention can deeply explore the personalized preferences of players, accurately predict the active time of players, automatically generate personalized email content, and determine the best email sending strategy through a multi-objective optimization algorithm, thereby improving the email opening rate and player participation, and bringing greater commercial value to game companies.

[0023] 2. The present invention realizes personalized customization of email content and sending timing, effectively improves email opening rate and conversion effect, and provides accurate user reach solutions for game operations.

[0024] 3. The present invention generates personalized email content for different groups of players based on the players' attribute data, historical active time data and social text data, thereby improving the pertinence and attractiveness of the emails.

[0025] 4. The present invention uses a time series prediction algorithm to accurately predict the player's future active time, ensuring that the email is sent during the time period when the player is most likely to be active, thereby increasing the email opening rate.

[0026] 5. The present invention utilizes natural language processing technology and collaborative filtering algorithms to automatically generate personalized email content, thereby reducing the burden of manual writing and improving the diversity and innovation of emails.

[0027] 6. The present invention uses a multi-objective optimization algorithm to comprehensively consider multiple factors such as email content, sending time and user behavior, determine the best email sending strategy, and achieve a significant increase in email opening rate.

[0028] 7. The present invention dynamically adjusts the frequency of email sending based on the players' historical email interaction behavior data, ensuring that the frequency of email sending matches the players' acceptance and feedback willingness, further improving the email opening rate and player satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1It is a flowchart of an email opening rate optimization method based on artificial intelligence provided by an embodiment of the present invention;

[0031] Figure 2 It is a structural diagram of an email open rate optimization system based on artificial intelligence provided by an embodiment of the present invention;

[0032] Figure 3 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0034] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0035] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0036] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0037] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0038] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0039] In the embodiment of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A flow chart of an artificial intelligence-based email open rate optimization method disclosed in an embodiment of the present invention is shown, and the details are as follows:

[0040] S101, obtaining player attribute data, classifying a plurality of players into groups according to the player attribute data, and determining a corresponding email content type for the players in each group.

[0041] In this embodiment, in order to obtain player attribute data, it is first necessary to determine the type of data to be collected. Common player attribute data include: natural attributes, such as age, gender, region, etc.; game behavior attributes, such as login frequency, daily / weekly online time, game level, consumption amount and frequency, role preference, number of social interaction days, etc.; device attributes, such as the type of device used (iOS / Android), device brand, device model, etc.

[0042] Based on the obtained player attribute data, cluster analysis and other methods can be used to divide players into categories, such as: high-spending active players, these players have high consumption amounts, long online time, and high game levels, and are the core user group in the game; medium-spending moderately active players, these players have a certain consumption capacity, but compared with high-spending active players, their online time and game levels may be slightly lower; low-spending lightly active players, these players spend less but occasionally log in to the game, and may be the potential user group of the game; inactive players, these players have not logged into the game for a long time and may have been lost.

[0043] Cluster analysis can comprehensively consider multiple attributes of players and automatically divide players into different groups through algorithms. Commonly used clustering algorithms include K-means, DBSCAN, etc.

[0044] Different email marketing strategies can be developed for different groups of players to increase the opening rate and conversion rate of emails. For example, for high-spending active players, email content types include exclusive promotions, high-end game props recommendations, VIP privileges descriptions, new gameplay previews, etc.; for medium-spending moderately active players, email content types include limited-time discounts, in-game event invitations, suggestions for improving the gaming experience, friend interaction reminders, etc.; for low-spending lightly active players, email content types include return rewards, game highlights introductions, simple task guides, social interaction invitations, etc.; for inactive players, email content types include recall emails, including game update introductions, reward return packages, friend invitation return, etc.

[0045] In this embodiment, through accurate grouping and personalized email content, the opening rate and conversion rate of emails can be significantly improved, thereby improving marketing effectiveness. Providing game content and preferential activities that meet the needs of players in different groups can enhance players' stickiness and loyalty to the game. By understanding players' consumption habits and behavioral characteristics through data analysis, marketing resources and development resources can be allocated more reasonably, improving resource utilization efficiency.

[0046] In some embodiments, in the above step S101, the grouping of multiple players according to the player attribute data and determining the corresponding email content type for each group of players specifically includes:

[0047] Obtaining a user feature vector by performing feature selection and feature extraction on the player attribute data;

[0048] Based on the user feature vector, a K-Means clustering algorithm is used to cluster multiple players, and players with similar attributes are divided into the same group to obtain multiple different player groups;

[0049] For each player group, analyze the common characteristics of each player in each player group in terms of game content preference, activity participation and willingness to pay, and obtain the characteristic analysis results of each player group;

[0050] Based on the feature analysis result, content elements are determined from a preset email material library to generate email content types corresponding to each player group.

[0051] In this embodiment, the original attribute data of the players is obtained, and the data is cleaned and preprocessed to remove noise data and outliers to ensure data quality. For the preprocessed player attribute data, feature selection algorithms such as chi-square test and mutual information are used to screen out key attribute features that have a significant impact on player clustering. For the screened key attributes of the players, high-dimensional attribute data is mapped to a low-dimensional space through feature extraction methods such as principal component analysis PCA to form a player feature vector. According to the similarity of the player feature vectors, the K-Means clustering algorithm is used to cluster the players, and the players with similar distances are divided into the same cluster cluster by calculating the Euclidean distance between samples. The clustering effect is evaluated by the elbow rule or silhouette coefficient, and the optimal cluster number K value is selected to obtain K different player groups. If the clustering effect is not ideal, return to adjust the feature selection method. The feature distribution of each player group is analyzed, the common characteristics of each group are summarized, and the group is named according to business needs, such as "high-paying users" and "casual players". The player clustering results are associated with the player ID and stored. When a new player registers, their attribute data is extracted to determine their category group, and personalized service recommendations and precision marketing are performed.

[0052] Obtain players' game content preferences, activity participation and willingness to pay data, perform cluster analysis on each player group, and obtain the common feature vector of each group; Based on the common feature vector, use the association rule mining algorithm to find the association rules between game content preferences, activity participation and willingness to pay for each player group, and form the feature analysis results; Based on the feature analysis results, determine the game content elements, participation activity types and willingness to pay of each player group; In the preset email material library, obtain the materials that match the above content elements, activity types and willingness to pay, and generate email content suitable for each player group through the combination optimization algorithm; Perform sentiment tendency analysis on the generated email content to determine whether the content meets the preference characteristics of the player group, and return and regenerate if it does not meet the requirements; Determine the best time to send emails based on the active time periods and activity participation patterns of the player group, and automatically send personalized emails through the email push system; Collect players' feedback data on email content, use the reinforcement learning algorithm to optimize the email content generation model, and continuously improve the matching degree between email content and player preferences.

[0053] Exemplarily, firstly, feature selection and feature extraction are performed on the attribute data of the players, and the principal component analysis (PCA) method is used to reduce the dimension of the original data, and 10 principal components with greater influence on the player classification are extracted as user feature vectors. Then, based on the user feature vectors, the K-Means clustering algorithm is used to cluster 10,000 players. By calculating the Euclidean distance between each player, players with close distance and similar attributes are divided into the same group, and finally 5 different player groups are obtained. Then, for each player group, the common characteristics of each player in terms of game content preference, activity participation and willingness to pay are analyzed. For example, for player group A, statistical analysis shows that 80% of the players in this group prefer strategy game content, participate in game activities 5 times a week on average, and the average monthly payment amount is more than 50 yuan. Based on the above feature analysis results, we selected strategy game-related materials from the pre-set 100 email material libraries as content elements, and combined the activity participation and willingness to pay of this group to generate personalized email content, such as "Warriors, the kingdom needs your strength. Participate in the weekend siege event to get double gold coin rewards and rare heroes waiting for you. You can participate for only 6 yuan. Don't miss the opportunity!" Similarly, we also conduct corresponding analysis on other player groups to generate differentiated email content that meets the characteristics of each group, and push it accurately to improve player activity and payment conversion.

[0054] S102, obtaining players' historical active time data, and determining the activity probability distribution data of each player in a future preset time period through a time series prediction algorithm based on the players' historical active time data.

[0055] In this embodiment, first, it is necessary to collect the historical active time data of each player. Such data usually comes from the log records of the game server, including but not limited to the player's login time, online time, daily / weekly / monthly active days, etc. In order to more comprehensively analyze the player's active mode, the player's specific behavior data in the game can also be collected, such as the time to complete the task, the time to participate in the activity, etc.

[0056] After obtaining the raw data, data preprocessing is required, including data cleaning (removing outliers, missing value processing, etc.), data conversion (such as converting timestamps to date and time formats), feature extraction (such as calculating the average online time and active frequency of players, etc.). These preprocessing steps are aimed at improving data quality and providing accurate and effective input for subsequent time series prediction algorithms.

[0057] Next, using the preprocessed player historical active time data, the time series prediction algorithm is applied to predict the player's active probability distribution in the future preset time period. Through the time series prediction algorithm, the daily active probability of each player in the future preset time period can be obtained. In order to more intuitively display the prediction results, these probability values ​​can be plotted as a probability distribution graph or curve graph. In addition, the prediction results can be further processed and analyzed according to actual needs, such as calculating the average active probability, maximum active probability and other statistical indicators of the player in a certain period of time in the future.

[0058] In this embodiment, the time series prediction algorithm can be used to accurately predict the player's future activity probability, providing strong data support for game operators.

[0059] In some embodiments, in the above step S102, determining the activity probability distribution data of each player in a future preset time period by a time series prediction algorithm based on the player's historical activity time data specifically includes:

[0060] Using the STL algorithm to decompose the player's historical active time data to obtain trend item data, cycle item data and random item data;

[0061] The ARIMA model is used to model the trend item data to obtain the trend item prediction result;

[0062] The periodic item data is analyzed by Fourier transform method, and the exponential smoothing model is used for modeling to obtain the periodic item prediction result;

[0063] Using a moving average model to model the random item data to obtain a random item prediction result;

[0064] The trend item prediction result, the period item prediction result and the random item prediction result are combined to obtain a predicted value of player active time within a future preset time period.

[0065] According to the predicted value of the player's active time, probability distribution fitting is performed through a Gaussian mixture model to obtain the active probability distribution data of each player in a preset time period in the future.

[0066] In this embodiment, the STL (seasonal decomposition model) algorithm is used to decompose the time series data to obtain trend items, period items and random item data. According to the trend item data, the ARIMA model is used for modeling processing to obtain ARIMA model parameters; for the ARIMA model, the time period to be predicted is input to obtain the trend item prediction data within the time period; if the confidence of the prediction data is lower than the preset threshold, the ARIMA model is updated online according to the residual of the prediction data to obtain new ARIMA model parameters; according to the new ARIMA model parameters, the time period is re-predicted until the confidence of the prediction data is higher than the threshold; the actual trend item data within the time period is obtained, the actual data is compared with the prediction data, and the prediction error is calculated; if the prediction error is greater than the preset threshold, the actual data is added to the training set, and the ARIMA model is retrained to obtain the updated model parameters; according to the updated ARIMA model parameters, the trend item of the next time period is predicted and the prediction result is output.

[0067] The Fourier transform method is used to perform periodic analysis on the periodic item data. The main periodic characteristics are determined by analyzing the frequency components and amplitudes in the Fourier transform results. According to the main periodic characteristics of the periodic item data, a suitable exponential smoothing model is selected, and multiple candidate exponential smoothing models are established by setting different smoothing coefficients. The historical periodic item data are used to train each candidate exponential smoothing model, and the optimal exponential smoothing model is screened out by comparing the fitting effects of each model on the training data. The periodic item data are input into the optimal exponential smoothing model, and the prediction results of the periodic item data in the future are obtained through the inference calculation of the model. The predicted periodic item results are post-processed to remove possible outliers, smooth and correct the predicted periodic item data, and obtain the final periodic item prediction results. The periodic item prediction results are combined with the original time series data, and the complete time series prediction data is generated as the final prediction output by addition or product.

[0068] For random item data, weighted average calculation is performed to obtain the smoothed value at the current moment as the predicted value of the random item. According to the preset threshold, it is judged whether the deviation between the current predicted value and the actual value exceeds the range. If it exceeds the range, the model parameters are adjusted. By continuously moving the window and updating the data, real-time prediction of random items and dynamic tracking of trend changes are achieved. The noise filtering algorithm is used to remove abnormal noise in the random item data and improve the accuracy of the prediction results. The predicted results are compared with the actual values, the prediction error is calculated, and the model is optimized and adjusted through the error feedback mechanism. The above process is continuously iterated until the prediction error converges to an acceptable range and a stable random item prediction result is obtained.

[0069] The prediction results of trend items, cycle items and random items are weighted and combined to obtain the predicted value of the player's active time in the future preset time period; according to the predicted value of the player's active time, a Gaussian mixture model is used to fit the probability distribution, and the parameters of the Gaussian mixture model are determined by methods such as maximum likelihood estimation; the fitted Gaussian mixture model is used to predict the probability distribution of each player's active time in the future preset time period to obtain the player's active probability distribution data; according to the player's active probability distribution data, it is determined whether each player is active in the future preset time period, and if the probability is greater than the preset threshold, the player is determined to be an active player; for the players determined to be active, the corresponding marketing strategy is determined according to their active probability distribution, and personalized marketing is carried out through push, email and other methods.

[0070] For example, the daily active time data of players in the past 30 days is first decomposed by STL to obtain long-term trend items, periodic items and random items. The trend item shows a slow upward trend, with an average increase of 5 minutes per day. The ARIMA (1,1,0) model is used to model the trend item and predict the trend item data for the next 7 days. After Fourier transform analysis of the periodic item, it is found that the player's active time has an obvious 7-day periodicity, that is, the player's activity is higher every weekend. The exponential smoothing model (α=8) is used to predict the periodic item and obtain the periodic item data for the next week. For the random item, the 5-day moving average model is used for fitting to predict the random fluctuations in the next 7 days. Finally, the trend item prediction value, the periodic item prediction value and the random item prediction value are added and combined to obtain the daily active time prediction data of the player in the next week. Based on the above time series analysis, the predicted active time value of each player in the next 7 days is obtained. In order to further characterize the active probability distribution of players, the Gaussian mixture model (GMM) is selected for modeling. The GMM parameter estimation of each player's active time data is performed through the EM algorithm, and two Gaussian components are fitted, with means of 30 minutes and 120 minutes, standard deviations of 10 minutes and 30 minutes, and weights of 4 and 6. This means that in the next week, there is a 40% probability that the player's active time will be concentrated around 30 minutes, and a 60% probability that it will be around 120 minutes. Using the obtained GMM probability density function, the player's active probability distribution in a specific time period in the future can be predicted, providing accurate data support for personalized operations.

[0071] S103, obtaining social text data of each player, and determining specific email content of each player according to the social text data.

[0072] In this embodiment, in order to obtain the social text data of each player, it is first necessary to determine the source of the data. Such data usually comes from the chat system in the game, the player forum, etc.

[0073] After obtaining the players' social text data, we can use natural language processing (NLP) and sentiment analysis to conduct in-depth analysis and mining of the text content to understand the players' interests, needs, emotional states, and other information. Based on this information, we can customize specific email content for each player. The specific implementation steps include:

[0074] Text preprocessing: Perform preprocessing work such as cleaning, word segmentation, and part-of-speech tagging on social text data to improve the accuracy and efficiency of subsequent analysis.

[0075] Sentiment analysis: Use sentiment analysis technology to identify the emotional tendencies in text (such as positive, negative, neutral, etc.) and understand the emotional state of players.

[0076] Topic modeling: Topic modeling technology is used to discover potential themes and topics in the text and understand players' concerns and discussion content.

[0077] Personalized email content generation: Based on the results of sentiment analysis, topic modeling, and other factors, combined with the player's historical behavior data and in-game status, personalized email content is generated for each player. The email content may include game information, event invitations, discount information, social interaction reminders, etc. that the player is interested in.

[0078] In this embodiment, by providing players with personalized email content, the needs and interests of players can be better met, and the user experience and satisfaction can be improved. Email content based on the interests and needs of players can more effectively attract the attention of players, increase the opening rate and conversion rate of emails, and thus improve marketing effectiveness.

[0079] In some embodiments, in the above step S103, determining the specific email content of each player according to the social text data specifically includes:

[0080] The social text data is processed using natural language processing technology to obtain personalized features of each player, wherein the personalized features include the player's emotional inclination towards different game contents and hot topics that the player is concerned about;

[0081] Using an item-based collaborative filtering algorithm, the correlation between each content element determined in the player group corresponding to each player and the personalized feature is calculated to obtain the top N content elements whose correlation scores are higher than a preset correlation threshold;

[0082] Using sentiment analysis algorithms to perform sentiment recognition on the personalized features, and determining the players' text tone and emotional labels for various game contents and game topics;

[0083] Based on the text tone emotion tags and the first N content elements with relevance scores higher than a preset relevance threshold, specific email content corresponding to each player is generated.

[0084] In this embodiment, natural language processing is performed on the social text data to extract personalized features such as player emotional tendencies and topics of interest; an item-based collaborative filtering algorithm is used to calculate the correlation score between each player's personalized feature and the content elements that the player group to which the player belongs has been exposed to; it is determined whether the correlation score is higher than a preset correlation threshold, and if so, the content element is added to a candidate set; the content elements in the candidate set are sorted according to the correlation score, and the top N are selected as recommended content for the target player; the recommended content is displayed to the target player, and the player's feedback data on the recommended content is recorded for optimizing the personalized recommendation model.

[0085] Use emotion recognition technology to perform semantic analysis on players' text expressions, identify the emotional tendencies of the expressions, and generate corresponding emotion tags. Use text generation algorithms to automatically combine and generate personalized email content based on the selected content elements and corresponding emotion tags. In the process of email content generation, incorporate game topics and keywords that players are interested in to increase the appeal and click-through rate of the email. Perform emotion analysis on the generated email content to ensure that the overall emotional tone of the email is consistent with the player's preferences and avoid emotional deviations. Push personalized email content to the corresponding players to achieve precision marketing based on the player's personality characteristics and emotional tendencies.

[0086] Exemplarily, firstly, natural language processing technology is used to process the social text data of players on game forums, chat rooms and other platforms. Through emotional dictionary matching and emotional polarity calculation, the emotional tendency scores of players for different game contents (such as PVP gameplay, PVE copies, social systems, etc.) are obtained, ranging from -1 to 1, where -1 means very negative and 1 means very positive. At the same time, the LDA topic model is used to mine the text data for topics, setting the number of topics to 10 and the number of iterations to 1000, and extracting the hot topics and their weights that players are concerned about, such as the topic weight of "game balance" is 3, and the topic weight of "new character online" is 2. Then, the item-based collaborative filtering algorithm is used to calculate the Pearson correlation coefficient between each player and each content element that has been determined in the player group to which he belongs, and select the top 5 content elements with a correlation score greater than 6. Then, the dictionary-based sentiment analysis algorithm is used to perform sentiment recognition on the text data in the player's personalized features to obtain the player's emotional labels for each game content and topic, such as "like", "hate", "neutral", etc. Finally, we customize the content of the email for each player based on the content elements and emotional tags with the highest relevance scores. For example, for players whose emotional inclination toward PVP gameplay is 8 and whose weight toward the topic of "game balance" is 3, the email will focus on the optimization and adjustment of the new version of the PVP system, the impact of the balance changes on players, etc., and use a positive tone to attract players' attention.

[0087] S104, taking the email opening rate as the optimization target, based on the email content type, the activity probability distribution data and the specific email content, determining the best email decision for each player through a multi-objective optimization algorithm.

[0088] In this embodiment, in email marketing, improving email opening rate is one of the most important goals. To achieve this goal, a multi-objective optimization algorithm can be used to determine the best email decision for each player by combining the email content type, the player's active probability distribution data and the specific email content.

[0089] The email opening rate is taken as the main optimization target, and other related indicators such as click-through rate and conversion rate are considered as auxiliary targets. Multi-objective optimization algorithms such as non-dominated sorting genetic algorithm (NSGA-II) are used to establish a multi-objective optimization model based on the email content type, active probability distribution data and specific email content, taking into account factors such as email sending time, email content type and player interest matching, and personalized customization degree.

[0090] Randomly generate a set of initial solutions containing different email sending strategies (such as sending time, email content, etc.). Evaluate each solution in the initial solution set and calculate its value on each objective function (such as email open rate, click rate, etc.). According to the evaluation results, use methods such as non-dominated sorting and crowding distance to select excellent solutions as parents. Generate new solution sets as children through operations such as crossover and mutation, and merge the parents and children to form a new solution set. Repeat the steps of evaluation, selection, and crossover mutation until the termination condition is met (such as reaching the maximum number of iterations, objective function value convergence, etc.). Select the email sending strategy with the best overall performance from the final solution set as the best email decision for each player. : Send emails to players according to the optimized email sending strategy. Collect email open rate, click rate, conversion rate and other data through the email tracking system.

[0091] In this embodiment, the email sending strategy is optimized through a multi-objective optimization algorithm, which can significantly improve the email opening rate and increase the interaction opportunities between players and the game. The email content is formulated based on the player's active probability distribution data and interest preferences, making the email more in line with the actual needs of the players and improving the personalized experience.

[0092] In some embodiments, in the above step S104, the optimization target of the email opening rate is to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the activity probability distribution data and the specific email content, specifically including:

[0093] The player's email opening rate is used as the reward function, the email content type, the active probability distribution data and the specific email content are used as the state space, and the email content strategy and the email sending strategy are used as the action space;

[0094] Based on the reward function, the state space and the action space, a DQN reinforcement learning algorithm is used to iteratively make decisions and calculate cumulative rewards to determine the best email decision for each player.

[0095] In this embodiment, the player's historical email data, including email content type, active probability distribution data and specific email content, is obtained as the state space. For different player groups, a variety of email content strategies and email sending strategies are designed to form an action space. According to the player's opening of historical emails, the email opening rate is calculated as an immediate reward value. The DQN reinforcement learning algorithm is adopted, based on the state space, action space and reward function, and the email decision is optimized through neural network iteration. In each iteration, an action is selected according to the current state, and the immediate reward and the next state are obtained, and this quadruple is stored in the experience replay pool. A batch of quadruple groups are randomly selected from the experience replay pool, the target Q value and the predicted Q value are calculated, and the mean square error is used as the loss function for back propagation to update the neural network parameters. After multiple rounds of iterations, a trained DQN model is obtained, which can select the best email content and sending strategy according to the player's state, thereby maximizing the cumulative reward and improving the player's email opening rate.

[0096] Exemplarily, first, the player's email opening rate is used as the reward function of the reinforcement learning algorithm, where the reward for emails with an opening rate greater than 80% is 1, the reward for an opening rate between 50% and 80% is 5, and the reward for an opening rate less than 50% is -1. Then, the email content type (such as event notifications, game updates, holiday greetings, etc.), the player's active probability distribution data (such as the probability of a player logging into the game every day) and the specific email content (i.e., personalized text) are used as the state space. Next, the email content strategy (such as email subject, content length, emotional tone, etc.) and the email sending strategy (such as sending time, sending frequency, etc.) are used as the action space. In the DQN algorithm, a neural network is used as a Q function approximation, the state s is input, and the Q value of each action a is output. The ε-greedy strategy is used for exploration, that is, an action is randomly selected with a probability of ε, and the action with the largest Q value is selected with a probability of 1-ε. At each decision step, an action is selected according to the current state, the next state and reward are observed, and the quadruple (s, a, r, s') is stored in the experience replay pool. A batch of quads are randomly drawn from the replay pool, the target Q value is calculated, and the mean square error is used as the loss function to update the neural network parameters. The training is iterated continuously until the Q function converges. Finally, for each player, according to his state characteristics, the trained Q function is used to select the optimal email content strategy and sending strategy, that is, the action with the largest Q value, as the best email decision for the player. For example, for a player who has a 90% probability of logging into the game every day and an emotional inclination of 8 for PVP gameplay, the DQN algorithm may choose to send a personalized email with the subject "New version of PVP gameplay first experience" at 8 o'clock on the second night after the game version is updated, with a word count of about 100 words and a positive emotional tone, in order to maximize the email open rate.

[0097] In some embodiments, in the above step S104, the optimization target of the email opening rate is to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the activity probability distribution data and the specific email content, specifically including:

[0098] Taking the email content type, the activity probability distribution data and the specific mailing content as input, a decision tree algorithm is used to make a decision on the best time and content of sending emails for each player;

[0099] Determine whether the decision result output by the decision tree algorithm meets a preset email opening rate threshold, and if so, determine the decision result as a candidate optimal email decision for the corresponding player;

[0100] The candidate best email decision is used as the initial population, and a genetic algorithm is used to optimize the candidate best email decision. The population is continuously iterated and updated through selection, crossover, and mutation operations until a preset convergence standard is reached or a preset upper limit of iteration times is reached, so as to obtain the best email decision for each player.

[0101] In this embodiment, the player's email content preference data, active distribution data and specific email content are used as input, and the best email sending time and content are determined for each player through a decision tree algorithm; it is determined whether the decision result output by the decision tree algorithm meets a preset email open rate threshold; if the decision result meets the preset threshold, the decision result is determined as the candidate best email decision plan for the corresponding player.

[0102] The candidate best email decision plan is used as the initial population of the genetic algorithm. For each candidate decision in the initial population, its fitness is evaluated. Decisions with high fitness have a greater probability of being selected for continued inheritance. The selected decision individuals are crossover-operated, that is, part of the content of two decisions is randomly selected to be exchanged to generate new decision individuals. The newly generated decision individuals are mutated, that is, the decision content is randomly changed with a certain probability to introduce new features. The decision individuals after selection, crossover, and mutation are added to the population to replace individuals with lower fitness, and the iterative update of the population is completed. Determine whether the current population has reached the preset convergence standard or the upper limit of the number of iterations. If not, return to continue iterative optimization. Select the decision with the highest fitness from the converged population as the best email decision for the player for subsequent email sending.

[0103] Exemplarily, first, the email content type, active probability distribution data and specific email content are used as input features, and the decision tree algorithm is used to make decisions on the best email sending time and content for each player. For example, for a player, based on its historical active data, its active probability distribution from Monday to Friday is 2, 3, 1, 3, 1. At the same time, combined with the email content type of activity notification, the decision tree algorithm may output the decision result of sending activity notification emails on Tuesday and Thursday. Next, it is determined whether the expected email opening rate corresponding to the decision result exceeds the preset threshold, such as 6. If it meets the threshold, the decision result is added to the candidate best email decision set. Then, the candidate best email decisions of all players are used as the initial population, each decision corresponds to an individual, and the genetic algorithm is used for optimization. Through operations such as selection, crossover, and mutation, the population is continuously iterated and updated, in which the selection operation will give priority to retaining individuals with high email opening rates, the crossover operation will exchange information between candidate decisions of different players, and the mutation operation will fine-tune certain decisions. For example, the average email opening rate of the current generation population is 65. After 50 iterations, the average email opening rate increases to 78, and the increase in 10 consecutive iterations is less than 0.1. In this case, the population is considered to have converged and the iteration is terminated. Finally, the optimal individual corresponding to each player is selected from the converged population, which is the optimal email decision for that player, including the optimal sending time and email content. Through the above method, key factors such as player activity patterns and email content attributes are comprehensively utilized, and the advantages of decision trees and genetic algorithms are combined to formulate the optimal email sending strategy for each player in a targeted manner, thereby significantly improving the overall email opening rate and marketing effect.

[0104] In some embodiments, the method further comprises:

[0105] Acquire player's historical email interaction behavior data, and construct time series features based on the player's historical email interaction behavior data, wherein the player's historical email interaction behavior data includes player's email opening behavior, player's email clicking behavior, and player's email conversion behavior;

[0106] Taking the time series features as input, the LSTM recurrent neural network model is used for training and prediction to determine the predicted value of each player's acceptance of the email and the predicted value of their willingness to give feedback;

[0107] Calculate each player's feedback speed characteristics for emails based on the player's historical email interaction behavior data, where the feedback speed characteristics include average email opening interval, average email clicking interval, and average email conversion interval;

[0108] The corresponding email sending frequency for each player in multiple time windows within a future preset time period is determined according to the acceptance prediction value, the feedback willingness prediction value and the feedback speed characteristic.

[0109] In this embodiment, the player's historical email interaction behavior data is obtained, and time series features are constructed for opening behavior, clicking behavior, and conversion behavior. Based on the time series features, a sliding time window is used to extract the behavior sequence within a certain time range. The extracted behavior sequence is input into a pre-constructed LSTM recurrent neural network model, and trained and learned through forward propagation and back propagation algorithms. Based on the LSTM model parameters obtained through training, the player's acceptance of emails and willingness to feedback in the future are predicted.

[0110] According to the players' historical email interaction behavior data, the average email opening interval, average email clicking interval, and average email conversion interval of each player are calculated and used as the feedback speed characteristics of the players to the email. The acceptance prediction value, feedback willingness prediction value, and feedback speed characteristics are weighted and fused to obtain the comprehensive feedback score of each player, and the players are grouped according to the score. Different email sending frequency thresholds are set for different player groups. If the player's comprehensive feedback score exceeds the threshold of the group, the player's email sending frequency in the future preset time period is increased, otherwise the sending frequency is maintained or reduced. In the future preset time period, the group to which each player belongs is dynamically adjusted according to the comprehensive feedback score of each player, and the email sending frequency is updated accordingly to achieve personalized email sending strategy optimization. The players' email interaction behavior is continuously tracked and analyzed, and the LSTM prediction model is regularly retrained with updated behavior data to continuously improve the prediction accuracy of players' feedback willingness and acceptance, and then dynamically optimize the email sending frequency strategy.

[0111] First, we obtain the email interaction records of each player in the past month from the game log database, including email opening time, click time, conversion time, etc., and construct time series features. For example, in the past 30 days, the time series of player A opening emails is [1,0,0,1,1,0,…], the time series of clicking emails is [0,0,1,0,1,0,…], and the time series of conversions is [0,0,0,1,0,0,…]. Then, we use these time series features as input and use the LSTM recurrent neural network model for training and prediction. The input layer dimension of the model is 30, corresponding to the 30-day time series; the hidden layer contains 64 neurons and uses the ReLU activation function; the output layer contains 2 neurons, representing acceptance and feedback willingness respectively, and the Sigmoid activation function is used to map the output value to between 0 and 1. The model uses the binary cross entropy loss function and the Adam optimizer, and is trained for 100 epochs, with a batch size of 128 for each epoch. Finally, based on the prediction results and feedback speed characteristics, the optimal email sending frequency for each player in the next 7 days is determined. Specifically, if the average acceptance prediction value of the player in the next 1 day, 3 days, and 7 days exceeds 6, 7, and 8 respectively, and the average feedback willingness prediction value exceeds 5, 6, and 7 respectively, then 1, 2, and 3 emails will be sent to the player every day within the corresponding time window; otherwise, the sending frequency is reduced to 1, 1, and 2 emails per day. Through the above method, the email sending frequency can be dynamically adjusted according to the personalized preferences of each player, while improving the player experience and maximizing the effect of email marketing.

[0112] Reference Figure 2 An embodiment of the present invention provides an email opening rate optimization system 2 based on artificial intelligence, and the system 2 specifically includes:

[0113] The first optimization module 201 is used to obtain player attribute data, classify multiple players into groups according to the player attribute data, and determine the corresponding email content type for each group of players;

[0114] The second optimization module 202 is used to obtain the player's historical active time data, and determine the activity probability distribution data of each player in a future preset time period through a time series prediction algorithm based on the player's historical active time data;

[0115] The third optimization module 203 is used to obtain social text data of each player and determine the specific email content of each player according to the social text data;

[0116] The fourth optimization module 204 is used to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the activity probability distribution data and the specific email content, with the email opening rate as the optimization target.

[0117] It is understandable that if Figure 1 The contents of the embodiment of the email opening rate optimization method based on artificial intelligence shown in the figure are all applicable to the embodiment of the email opening rate optimization system based on artificial intelligence. The functions specifically implemented by the embodiment of the email opening rate optimization system based on artificial intelligence are the same as those in the embodiment of the figure. Figure 1 The embodiment of the method for optimizing email opening rate based on artificial intelligence is the same as that shown in FIG. Figure 1 The beneficial effects achieved by the embodiment of the email opening rate optimization method based on artificial intelligence shown are also the same.

[0118] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0119] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0120] Reference Figure 3 The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, the artificial intelligence-based email opening rate optimization method as described in any one of the above methods is implemented.

[0121] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0122] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0123] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 3. Further, the memory 302 may also include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, an application program, a boot loader (Boot Loader), data, and other programs, such as the program code of the computer program, etc. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0124] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for optimizing email opening rate based on artificial intelligence as described in any one of the above methods is implemented.

[0125] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0126] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0127] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. The email opening rate optimization method based on artificial intelligence is characterized by: The method specifically comprises: Acquire player attribute data, classify multiple players into groups according to the player attribute data, and determine a corresponding email content type for each group of players; Obtain the player's historical active time data, and determine the activity probability distribution data of each player in a preset time period in the future through a time series prediction algorithm based on the player's historical active time data; Obtaining social text data of each player, and determining specific email content of each player based on the social text data; Taking the email opening rate as the optimization goal, based on the email content type, the active probability distribution data and the specific email content, the best email decision for each player is determined through a multi-objective optimization algorithm.

2. The method according to claim 1, characterized in that The dividing the multiple players into groups according to the player attribute data and determining the corresponding email content type for each group of players specifically includes: Obtaining a user feature vector by performing feature selection and feature extraction on the player attribute data; Based on the user feature vector, a K-Means clustering algorithm is used to cluster multiple players, and players with similar attributes are divided into the same group to obtain multiple different player groups; For each player group, analyze the common characteristics of each player in each player group in terms of game content preference, activity participation and willingness to pay, and obtain the characteristic analysis results of each player group; Based on the feature analysis result, content elements are determined from a preset email material library to generate email content types corresponding to each player group.

3. The method according to claim 1, characterized in that: The method of determining the activity probability distribution data of each player in a future preset time period according to the player's historical active time data by a time series prediction algorithm specifically includes: Using the STL algorithm to decompose the player's historical active time data to obtain trend item data, cycle item data and random item data; The ARIMA model is used to model the trend item data to obtain the trend item prediction result; The periodic item data is analyzed by Fourier transform method, and the exponential smoothing model is used for modeling to obtain the periodic item prediction result; Using a moving average model to model the random item data to obtain a random item prediction result; The trend item prediction result, the period item prediction result and the random item prediction result are combined to obtain a predicted value of player active time within a future preset time period. According to the predicted value of the player's active time, probability distribution fitting is performed through a Gaussian mixture model to obtain the active probability distribution data of each player in a preset time period in the future.

4. The method according to claim 2, characterized in that: Determining the specific email content of each player according to the social text data specifically includes: The social text data is processed using natural language processing technology to obtain personalized features of each player, wherein the personalized features include the player's emotional inclination towards different game contents and hot topics that the player is concerned about; Using an item-based collaborative filtering algorithm, the correlation between each content element determined in the player group corresponding to each player and the personalized feature is calculated to obtain the top N content elements whose correlation scores are higher than a preset correlation threshold; Using sentiment analysis algorithms to perform sentiment recognition on the personalized features, and determining the players' text tone and emotional labels for various game contents and game topics; Based on the text tone emotion tags and the first N content elements with relevance scores higher than a preset relevance threshold, specific email content corresponding to each player is generated.

5. The method according to claim 1, characterized in that: The optimization objective of the email opening rate is to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the active probability distribution data and the specific email content, specifically including: The player's email opening rate is used as the reward function, the email content type, the active probability distribution data and the specific email content are used as the state space, and the email content strategy and the email sending strategy are used as the action space; Based on the reward function, the state space and the action space, a DQN reinforcement learning algorithm is used to iteratively make decisions and calculate cumulative rewards to determine the best mail decision for each player.

6. The method according to claim 1, characterized in that The optimization objective of the email opening rate is to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the active probability distribution data and the specific email content, specifically including: Taking the email content type, the activity probability distribution data and the specific mailing content as input, a decision tree algorithm is used to make a decision on the best time and content of sending emails for each player; Determine whether the decision result output by the decision tree algorithm meets a preset email opening rate threshold, and if so, determine the decision result as a candidate optimal email decision for the corresponding player; The candidate best email decision is used as the initial population, and a genetic algorithm is used to optimize the candidate best email decision. The population is continuously iterated and updated through selection, crossover, and mutation operations until a preset convergence standard is reached or a preset upper limit of iteration times is reached, so as to obtain the best email decision for each player.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Acquire player's historical email interaction behavior data, and construct time series features based on the player's historical email interaction behavior data, wherein the player's historical email interaction behavior data includes player's email opening behavior, player's email clicking behavior, and player's email conversion behavior; Taking the time series features as input, the LSTM recurrent neural network model is used for training and prediction to determine the predicted value of each player's acceptance of the email and the predicted value of their willingness to give feedback; Calculate each player's feedback speed characteristics for emails based on the player's historical email interaction behavior data, where the feedback speed characteristics include average email opening interval, average email clicking interval, and average email conversion interval; The corresponding email sending frequency for each player in multiple time windows within a preset time period in the future is determined according to the acceptance prediction value, the feedback willingness prediction value and the feedback speed characteristic.

8. The email opening rate optimization system based on artificial intelligence is characterized by: The system specifically comprises: A first optimization module, configured to obtain player attribute data, classify multiple players into groups according to the player attribute data, and determine a corresponding email content type for each group of players; The second optimization module is used to obtain the player's historical active time data, and determine the activity probability distribution data of each player in a future preset time period through a time series prediction algorithm based on the player's historical active time data; A third optimization module is used to obtain social text data of each player and determine the specific email content of each player according to the social text data; The fourth optimization module is used to determine the best email decision for each player through a multi-objective optimization algorithm based on the email content type, the active probability distribution data and the specific email content, with the email opening rate as the optimization target.

9. A computer device, characterized in that: include: A memory and a processor and a computer program stored in the memory, when the computer program is executed on the processor, implements the artificial intelligence-based email opening rate optimization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method for optimizing email opening rate based on artificial intelligence as described in any one of claims 1 to 7 is implemented.

Images