技术领域technical field
本发明涉及游戏运营领域，尤其是涉及一种游戏平台信息推送方法、系统、存储介质及设备。The invention relates to the field of game operation, in particular to a game platform information push method, system, storage medium and device.
背景技术Background technique
随着互联网技术的快速发展，游戏产业也日新月异，各类游戏层出不穷，游戏运营商之间的竞争也越来越激烈，在游戏体验差别不大的情况下，稳定的运营与良好的服务决定了游戏能否脱颖而出。而在当前获取新用户的成本越来越高，用户流失也是游戏运营中常见现象，为了维持用户的数量，游戏平台往往会对用户进行游戏信息的推送以保证用户的粘性。With the rapid development of Internet technology, the game industry is also changing with each passing day. Various types of games are emerging one after another, and the competition among game operators is becoming more and more fierce. In the case that the game experience is not very different, stable operation and good service determine Can the game stand out? At present, the cost of acquiring new users is getting higher and higher, and user churn is also a common phenomenon in game operations. In order to maintain the number of users, game platforms often push game information to users to ensure user stickiness.
然而，在实际业务中，游戏平台在对用户进行信息推送时往往针对全体用户进行信息推送，容易出现推送信息与用户实际需求不符合的情况，过多的无关推送信息占用网络资源的同时由于与用户需求不匹配，缺乏对用户的吸引力，使得推送信息效果差。However, in actual business, the game platform often pushes information to all users when pushing information to users, which is prone to the situation that the push information does not meet the actual needs of users, and too much irrelevant push information occupies network resources. User needs do not match, lack of attractiveness to users, making the push information ineffective.
发明内容SUMMARY OF THE INVENTION
本发明的目的在于克服现有技术的缺点与不足，提供一种推送内容与用户需求匹配、匹配性高的游戏平台信息推送方法、系统、存储介质及设备。The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a game platform information push method, system, storage medium and device that match the push content with user requirements and have high matching.
一种游戏平台信息推送方法，包括以下步骤：A method for pushing game platform information, comprising the following steps:
获取服务器中的用户数据；Get user data from the server;
基于用户数据建立用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签；Create user basic information tags, user consumption prediction tags, game preference tags and lost user tags based on user data;
根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送。Group users according to user data and tags, and push information according to the assigned user groups.
相对于现有技术，本案通过对服务器中海量用户数据进行数据分析并建立相应的标签，可以帮助运营人员快速、精准地定位用户，全方位、多角度地掌握用户特征，通过对用户进行分群并根据已分配的用户群体进行信息推送，提高了匹配的准确性，节省了网络资源。Compared with the prior art, in this case, by analyzing the massive user data in the server and establishing corresponding labels, it can help operators to locate users quickly and accurately, and grasp user characteristics in an all-round and multi-angle manner. Information is pushed according to the assigned user groups, which improves the accuracy of matching and saves network resources.
进一步地，所述基于用户数据建立用户基础信息标签的步骤包括：Further, the step of establishing a user basic information label based on the user data includes:
获取用户注册时间，注册时长，注册设备等信息建立注册行为标签；Obtain user registration time, registration time, registered equipment and other information to establish registration behavior labels;
统计用户登录IP并解析出用户登录地址，建立用户所在城市等级标签；Count user login IP and parse out the user login address, and establish the city level label of the user;
统计用户登录时间点，计算用户活跃时段，建立用户活跃类型标签；Count user login time points, calculate user active period, and create user active type labels;
根据用户登录时间，计算用户活跃指数，建立用户活跃指数标签；According to the user login time, calculate the user activity index and establish the user activity index label;
根据用户参与活动的消费次数、消费金额，分析用户浏览行为，建立用户消费行为标签和用户消费指数标签。通过建立用户基础信息标签，方便全方位、多角度地对用户进行分析，实现推送内容与用户喜好之间的快速匹配。According to the consumption times and consumption amount of users participating in activities, the user's browsing behavior is analyzed, and the user's consumption behavior label and the user's consumption index label are established. By establishing user basic information labels, it is convenient to analyze users in an all-round and multi-angle manner, and to achieve rapid matching between push content and user preferences.
进一步地，所述基于用户数据建立用户消费预测标签的步骤包括：Further, the step of establishing a user consumption prediction label based on user data includes:
从服务器日志采集用户数据；Collect user data from server logs;
提取一个周期内付费用户和无付费用户的用户基础属性特征和游戏行为属性特征，并对所述用户基础属性特征和游戏行为属性特征进行特征工程处理；Extract basic user attribute features and game behavior attribute features of paying users and non-paying users within a cycle, and perform feature engineering on the user basic attribute features and game behavior attribute features;
利用梯度提升决策树构建分类模型并通过K折交叉验证法调整分类模型参数；Use gradient boosting decision tree to build classification model and adjust the parameters of classification model through K-fold cross-validation method;
提取一个周期内付费用户的用户基础属性特征和游戏行为属性特征，并对所述用户基础属性特征和游戏行为属性特征进行特征工程处理；Extracting basic user attribute features and game behavior attribute features of paying users within a cycle, and performing feature engineering on the user basic attribute features and game behavior attribute features;
利用梯度提升回归构建回归模型并通过K折交叉验证法调整回归模型参数；Use gradient boosting regression to build a regression model and adjust the parameters of the regression model through K-fold cross-validation;
制定用户消费等级规则；Formulate user consumption level rules;
将所述分类模型、回归模型和用户消费等级规则整合为用户消费预测模型。使用梯度提升回归构建回归模型，可以灵活地处理各类型数据，包括离散型特征和连续型特征，对异常值的鲁棒性更强，准确率更高。The classification model, regression model and user consumption level rules are integrated into a user consumption prediction model. Using gradient boosting regression to build a regression model can flexibly process various types of data, including discrete features and continuous features, with stronger robustness to outliers and higher accuracy.
进一步地，所述基于用户数据建立游戏偏好标签的步骤包括：Further, the step of establishing a game preference tag based on user data includes:
统计用户在注册每种类型游戏前产生相关行为事件数，并以此计算获取对应类型游戏的初始喜爱度；Count the number of related behavior events that users generate before registering for each type of game, and use this to calculate the initial favorite degree of the corresponding type of game;
统计用户在该周期对每种类型游戏投入的时间以及在上一周期投入的时间，并以此计算出每种类型游戏时间增长量的增长率；Count the time users invested in each type of game in this cycle and the time invested in the previous cycle, and use this to calculate the growth rate of each type of game time increase;
根据以下方式计算每种游戏类型的用户喜爱度：User likeability is calculated for each game type according to:
其中，αi为用户游戏时间增长量的增长率，Δx为用户投入的游戏时间，Tj表示用户喜爱度的初始值，ti-tj表示时间间隔，M表示用户在某类游戏中的消费总额，λ为常量，表示消费额度对热度增长的比例，Ti表示用户在ti周期对某类游戏的喜爱度。通过对构建用户游戏偏好标签，方便后续运营人员针对指定人群实施具体的营销策略。Among them, αi is the growth rate of the user's game time increase, Δx is the game time invested by the user, Tj represents the initial value of the user's favorite degree, ti -tj represents the time interval, and M represents the user's game time in a certain type of game. The total consumption, λ is a constant, represents the ratio of consumption to the increase in popularity, and Ti represents the user's preference for a certain type of game in the ti period. By constructing user game preference tags, it is convenient for subsequent operators to implement specific marketing strategies for designated groups of people.
进一步地，所述基于用户数据建立流失用户标签的步骤包括：Further, the step of establishing a lost user label based on user data includes:
利用用户唯一标识将用户数据及标签关联，获得特征数据集；Use the user's unique identifier to associate user data and tags to obtain a feature data set;
利用梯度提升树和特征权值对特征数据集进行特征选择；Feature selection on feature datasets using gradient boosting trees and feature weights;
利用K折交叉验证法训练若干基学习模型，并根据若干基学习模型的输出结果构建融合模型；Use the K-fold cross-validation method to train several basic learning models, and build a fusion model according to the output results of several basic learning models;
调用融合模型识别流失用户并生成流失用户标签信息。通过利用多种单一模型学习不同用户各种各样的行为特点，调用融合模型识别流失用户，提高对流失用户的预测准确性。Call the fusion model to identify churn users and generate churn user label information. By using a variety of single models to learn various behavioral characteristics of different users, call the fusion model to identify the lost users, and improve the prediction accuracy of the lost users.
进一步地，还包括以下步骤：将用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签的标签名称作为键，标签下的用户转换为该键对应的值，将该键和值以相互对应的方式存储至数据库中；其中，该步骤具体包括：Further, it also includes the following steps: using the user basic information label, the user consumption prediction label, the game preference label and the label name of the lost user label as the key, the user under the label is converted into the value corresponding to the key, and the key and value are as follows. The mutually corresponding ways are stored in the database; wherein, this step specifically includes:
获取标签名称、最大用户标识及具有该标签的所有用户的唯一标识；Get the tag name, the maximum user ID, and the unique IDs of all users with the tag;
将标签中所有用户的唯一标识通过位图算法转换为k位的位图数组；其中，k＝1+N/32，N为标签的用户个数；Convert the unique identifiers of all users in the label into a k-bit bitmap array through a bitmap algorithm; wherein, k=1+N/32, and N is the number of users in the label;
将位图数组转换为十六进制字符串，并以标签名称作为键，所述十六进制字符串作为该键对应的值，存储在数据库中。利用位图算法对标签信息进行存储，不仅减少了使用的内存空间，也大大地提高了查询性能。Convert the bitmap array to a hexadecimal string, and use the label name as a key, and the hexadecimal string is stored in the database as the value corresponding to the key. Using the bitmap algorithm to store the tag information not only reduces the memory space used, but also greatly improves the query performance.
进一步地，所述根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送的步骤具体包括：Further, the steps of grouping users according to user data and labels and pushing information according to the assigned user groups specifically include:
针对消费高于设定阈值和消费低于设定阈值的用户进行优惠券信息的推送，并监控其消费情况；Push coupon information for users whose consumption is higher than the set threshold and whose consumption is lower than the set threshold, and monitor their consumption;
针对流失用户及即将流失的用户进行召回信息推送，并监控其召回用户数量；Push recall information for lost users and users who are about to be lost, and monitor the number of recalled users;
监控信息推送前后用户数量的变化并进行展示。通过对不同标签的用户实现不同信息的推送并进行用户数量变化的监控，直观地展示了游戏平台的信息推送效果，为后续活动提供数据支持。Monitor and display changes in the number of users before and after information is pushed. By pushing different information to users with different labels and monitoring the changes in the number of users, the information push effect of the game platform is intuitively displayed, and data support is provided for follow-up activities.
本发明还提供了一种游戏平台信息推送系统，包括：The present invention also provides a game platform information push system, including:
数据获取模块，用于获取服务器中的用户数据；The data acquisition module is used to acquire user data in the server;
标签建立模块，用于基于用户数据建立用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签；The label establishment module is used to establish user basic information label, user consumption prediction label, game preference label and lost user label based on user data;
信息推送模块，用于根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送。The information push module is used to group users according to user data and tags and push information according to the assigned user groups.
相对于现有技术，本案提供了一整套系统化的模块共同协作，通过对服务器中海量用户数据进行数据分析并建立相应的标签，可以帮助运营人员快速、精准地定位用户，全方位、多角度地掌握用户特征，实现推送内容与用户喜好之间的快速匹配，提高了匹配的准确性，节省了网络资源。Compared with the existing technology, this case provides a set of systematic modules to work together. By analyzing the massive user data in the server and establishing corresponding tags, it can help operators to locate users quickly and accurately, in an all-round and multi-angle manner. It can grasp user characteristics and realize fast matching between push content and user preferences, which improves the accuracy of matching and saves network resources.
本发明还提供了一种计算机可读存储介质，其上储存有计算机程序，其特征在于：该计算机程序被处理器执行时实现上述游戏平台信息推送方法的步骤。The present invention also provides a computer-readable storage medium on which a computer program is stored, characterized in that: when the computer program is executed by a processor, the steps of the above method for pushing game platform information are implemented.
本发明还提供了一种计算机设备，包括储存器、处理器以及储存在所述储存器中并可被所述处理器执行的计算机程序，所述处理器执行所述计算机程序时实现上述的游戏平台信息推送方法的步骤。The present invention also provides a computer device, comprising a storage, a processor, and a computer program stored in the storage and executable by the processor, when the processor executes the computer program, the above-mentioned game is implemented The steps of the platform information push method.
为了更好地理解和实施，下面结合附图详细说明本发明。For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings.
附图说明Description of drawings
图1是本发明实施例1所述游戏平台信息推送方法的流程图；1 is a flowchart of a method for pushing game platform information according to Embodiment 1 of the present invention;
图2是本发明中实施例1中所述游戏平台信息推送方法架构图；Fig. 2 is the framework diagram of the game platform information push method described in Embodiment 1 of the present invention;
图3是本发明实施例1中利用K折交叉验证构建融合模型流程图；3 is a flow chart of constructing a fusion model by utilizing K-fold cross-validation in Embodiment 1 of the present invention;
图4是本发明实施例1中融合模型的结构图；4 is a structural diagram of a fusion model in Embodiment 1 of the present invention;
图5是本发明实施例2中所述游戏平台信息推送系统的结构图。FIG. 5 is a structural diagram of the game platform information push system described in Embodiment 2 of the present invention.
具体实施方式Detailed ways
实施例1Example 1
请参阅图1-2，其是本发明所述游戏平台信息推送方法的流程图。Please refer to FIGS. 1-2 , which are flowcharts of the method for pushing game platform information according to the present invention.
一种游戏平台信息推送方法，包括以下步骤：A method for pushing game platform information, comprising the following steps:
S1：获取服务器中的用户数据；S1: Get user data in the server;
所述获取服务器中的用户数据步骤中，所述服务器中的用户数据指多种产品上报至服务器的数据(包括但不限于自有平台产品数据、爬虫采集的第三方数据)，通过ETL(Extract-Transform-Load)技术得到的用户全体数据集，具体包括：In the step of obtaining the user data in the server, the user data in the server refers to the data reported by a variety of products to the server (including but not limited to product data on its own platform, third-party data collected by crawlers), through ETL (Extract -Transform-Load) technology to obtain the entire user data set, including:
上报至服务器数据库中的结构化、非结构化数据和上报至服务器本地磁盘日志文件；其中，所述数据库包括但不限于MySQL、MongoDB、HBase数据库，所述上报至服务器数据库中的结构化、非结构化数据包括用户注册表、用户登录行为表、用户订单表、用户充值表、用户浏览明细表等等，具体包括用户唯一标识、用户年龄、注册时间、登录时间、消费金额、优惠券消费明细等结构化数据；所述上报至服务器本地磁盘日志文件，包括跟进用户时产生的日志文件等非结构化数据。其中，对于非结构化数据，将系统中对应用户产生的日志进行组合，拼接成一个对应用户唯一标识符的独立文本，并利用分词工具进行清洗。具体地，使用分词工具进行分词处理并去除停用词，得到语料集，再通过TF-IDF算法获取用户关键词，所述TF-IDF算法常用于获取文件内词语的重要程度，其计算过程在本实施例中不做具体限定。本步骤中所述分词工具包括但不限于CoreNLP、Jieba等分词工具。The structured and unstructured data reported to the server database and the local disk log file reported to the server; wherein, the database includes but not limited to MySQL, MongoDB, and HBase databases, and the structured and unstructured data reported to the server database. Structured data includes user registration form, user login behavior table, user order table, user recharge table, user browsing list, etc., including user unique ID, user age, registration time, login time, consumption amount, coupon consumption details and other structured data; the log files reported to the local disk of the server include unstructured data such as log files generated when following up users. Among them, for unstructured data, the logs generated by the corresponding users in the system are combined, spliced into an independent text corresponding to the user's unique identifier, and cleaned using word segmentation tools. Specifically, a word segmentation tool is used to perform word segmentation processing and remove stop words to obtain a corpus, and then obtain user keywords through the TF-IDF algorithm. The TF-IDF algorithm is often used to obtain the importance of words in a file. The calculation process is as follows: There is no specific limitation in this embodiment. The word segmentation tools in this step include but are not limited to CoreNLP, Jieba and other word segmentation tools.
S2：基于用户数据建立用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签。在其它实施例中，本步骤中用户也可根据实际需求增加或重新建立其它用户标签信息，便于后续探索更多用户标签信息，最大限度地发挥数据驱动的价值。S2: Based on user data, establish user basic information labels, user consumption prediction labels, game preference labels and lost user labels. In other embodiments, in this step, the user can also add or re-create other user tag information according to actual needs, so as to facilitate subsequent exploration of more user tag information and maximize the value of data-driven.
其中，针对用户数据进行数据分析，所述建立用户基础信息标签步骤包括：Wherein, performing data analysis on user data, the step of establishing user basic information label includes:
S211：获取用户注册时间，注册时长，注册设备等信息建立注册行为标签；S211: Obtain user registration time, registration time, registered equipment and other information to establish a registration behavior label;
S212：统计用户登录IP并解析出用户登录地址，建立用户所在城市等级标签；S212: Count the user login IP and parse out the user login address, and establish a city level label where the user is located;
S213：统计用户登录时间点，计算用户活跃时段，建立用户活跃类型标签；其中，所述活跃类型包括日间活跃型还是夜间活跃型。S213: Count user login time points, calculate user active time period, and establish a user active type label; wherein, the active type includes day active type or night active type.
S214：根据用户登录时间，计算用户活跃指数，建立用户活跃指数标签，其中，用户活跃指数可按时间粒度划分，如近T天登录天数，近T天离线天数等来综合评价，具体地，所述用户活跃指数通过下述方式获取：S214: Calculate the user activity index according to the user login time, and establish a user activity index label, where the user activity index can be divided according to time granularity, such as the number of login days in the past T days, the number of offline days in the past T days, etc. for comprehensive evaluation. The above user activity index is obtained in the following ways:
其中，ActiveIndex为用户活跃指数，wi为活跃指标的权重值，T为用户活动周期，Xi为用户在周期T天内活跃指标，N为评价所述活跃指标的事件数；例如，在本实施例中，N＝3，X1为周期内离线天数，X2为周期内登录天数，X3为周期内最后一次登录日期与首次登陆日期之差。Among them, ActiveIndex is the user activity index,wi is the weight value of the activity index, T is the user activity period, Xi is the user activity index within T days of the cycle, and N is the number of events that evaluate the activity index; for example, in this implementation In the example, N=3, X1 is the number of offline days in the cycle, X2 is the number of login days in the cycle, and X3 is the difference between the last login date and the first login date in the cycle.
度量用户活跃度指数标签为：The label to measure the user activity index is:
其中，ActiveIndexLabel为活跃度指数，当活跃度指数为1时表示该用户低活跃，活跃度指数为2时表示该用户一般活跃，活跃度指数为3表示该用户高度活跃。Among them, ActiveIndexLabel is the activity index. When the activity index is 1, it means that the user is lowly active, when the activity index is 2, it means that the user is generally active, and when the activity index is 3, it means that the user is highly active.
S215：根据用户参与活动的消费次数、消费金额，结合用户浏览行为数据，建立用户消费行为标签和用户消费指数标签。S215: Create a user consumption behavior label and a user consumption index label according to the consumption frequency and consumption amount of the user participating in the activity and in combination with the user browsing behavior data.
其中，S215步骤中建立用户消费行为标签包括：Wherein, establishing the user consumption behavior label in step S215 includes:
结合用户参与活动的消费次数与总消费次数的比值，建立用户对营销活敏感指数标签；Combined with the ratio of the consumption times of users participating in activities to the total consumption times, establish the user's sensitivity index label to marketing activities;
结合用户浏览行为数据对用户的理性消费指数做定性分析，例如分析用户点击进入购买装备的详情页的次数、浏览装备数量与其首次浏览至购买装备的时长关系，将用户定性为冲动消费型、理性消费型、犹豫消费型等，例如，当用户首次进入装备详情页且没有去浏览其他装备详情页，就购买此装备，则将其定性为冲动消费型等方式建立用户理性消费指数标签。Combining user browsing behavior data to qualitatively analyze the user's rational consumption index, such as analyzing the number of times users click to enter the detailed page for purchasing equipment, the relationship between the number of browsing equipment and the duration of the first browsing to purchasing equipment, and characterize users as impulsive consumption type, rational Consumption type, hesitant consumption type, etc. For example, when a user enters the equipment details page for the first time and does not browse other equipment details pages, and buys this equipment, it will be characterized as impulsive consumption type and other methods to establish a user rational consumption index label.
其中，S215步骤中，建立用户消费指数标签的步骤包括：Wherein, in step S215, the step of establishing the user consumption index label includes:
统计所有用户的消费频次、最近一次消费时间、消费总金额、最大消费金额、特价消费频次、高价消费频次作为数据集；Count the consumption frequency of all users, the last consumption time, the total consumption amount, the maximum consumption amount, the special consumption frequency, and the high price consumption frequency as a data set;
计算数据集中每一类消费属性的四分位数，根据属性值与各四分位数的大小关系对用户消费属性进行评分；所述评分项包括消费频次评分、最近一次消费时间评分、消费总金额评分、最大消费金额评分、优惠消费频次评分、高价消费频次评分。Calculate the quartile of each type of consumption attribute in the data set, and score the user consumption attribute according to the relationship between the attribute value and each quartile; the scoring items include the consumption frequency score, the last consumption time score, and the total consumption. Amount score, maximum consumption amount score, preferential consumption frequency score, high price consumption frequency score.
对各属性评分项加权计算获得用户消费指数，The user consumption index is obtained by weighted calculation of each attribute rating item,
其中，wi分别为各项评分的权重，i为评分项个数，i∈[1,2,3,4,5,6]。Among them,wi is the weight of each rating, i is the number of rating items, i∈[1,2,3,4,5,6].
根据用户消费指数，建立用户消费指数标签。According to the user consumption index, a user consumption index label is established.
具体地，步骤S2中针对用户数据进行数据分析，建立用户消费预测标签的步骤包括：Specifically, in step S2, data analysis is performed on the user data, and the steps of establishing a user consumption prediction label include:
S221：从服务器日志采集并预处理用户数据；其中，所述预处理用户数据步骤包括：S221: Collect and preprocess user data from server logs; wherein, the step of preprocessing user data includes:
关联用户设备号和用户账号，过滤关联设备异常用户和过滤关联多个用户账号的设备号，此类用户可能为非正常用户如代练等。Associate user device numbers and user accounts, filter abnormal users associated with devices, and filter device numbers associated with multiple user accounts. Such users may be abnormal users such as power leveling.
过滤异常付费用户；Filter out abnormal paying users;
忽略属性值大量缺失的属性；Ignore properties with a large number of missing property values;
对渠道类型、手机平台等离散特征进行独热编码；One-hot encoding of discrete features such as channel type and mobile platform;
量化用户游戏事件；Quantify user game events;
将手机号、身份证等用户基础属性二元化。Binary user basic attributes such as mobile phone number and ID card.
S222：提取一个周期内付费用户和无付费用户的用户基础属性特征和游戏行为属性特征，并对所述用户基础属性特征和游戏行为属性特征进行特征工程处理；S222: Extract basic user attribute features and game behavior attribute features of paying users and non-paying users within a cycle, and perform feature engineering on the user basic attribute features and game behavior attribute features;
其中，所述用户基础属性包括：性别、年龄、平台类型、注册渠道、已注册天数、登录天数、关联账号数、VIP等级、登录次数、常用IP地区；所述用户游戏行为属性包括：用户游戏数量、用户游戏类型、用户游戏评级、用户游戏角色等级、用户游戏评级、游戏事件数、参与重要活动次数、游戏时长；所述对所述用户基础属性特征和游戏行为属性特征进行特征工程处理步骤具体包括：Wherein, the basic user attributes include: gender, age, platform type, registration channel, days registered, days logged in, number of associated accounts, VIP level, number of logins, and commonly used IP regions; the user game behavior attributes include: user games Quantity, user game type, user game rating, user game character level, user game rating, number of game events, number of important activities participated, and game duration; the feature engineering processing steps for the user basic attribute features and game behavior attribute features Specifically include:
对游戏事件分类并统计每种类别事件总数；过滤应用PCA或其它降维法处理时线性关系很大的特征，同时计算同类特征的均值、中值等作为新的特征加入数据集；过滤方差为0或者接近0的属性；Classify game events and count the total number of events of each category; filter features with a large linear relationship when PCA or other dimensionality reduction methods are applied, and calculate the mean, median, etc. of similar features as new features to add to the data set; the filtering variance is 0 or close to 0 attributes;
应用RF值、F值及互信息值对特征进行评分，选择三种评分前十的用户特征的交集作为用户分类模型关键特征。The features are scored using the RF value, the F value and the mutual information value, and the intersection of the three top ten user features is selected as the key feature of the user classification model.
S223：利用梯度提升决策树构建分类模型并通过K折交叉验证法调整参数；具体地，使用梯度提升决策树(Gradient Boosting Decision Tree，GBDT)建立分类模型，通过K折交叉验证法得到最优分类模型，用于预测用户是否付费。S223: Use the gradient boosting decision tree to build a classification model and adjust the parameters through the K-fold cross-validation method; specifically, use the gradient boosting decision tree (Gradient Boosting Decision Tree, GBDT) to build the classification model, and obtain the optimal classification through the K-fold cross-validation method A model for predicting whether a user will pay.
S224：提取一个周期内付费用户的用户基础属性和游戏行为属性特征，并对所述用户基础属性和游戏行为属性特征进行特征工程处理；具体地，提取用户数据集中周期T内付费用户的用户基础属性、用户消费行为属性及用户游戏行为属性，并使用S222所述方法对所述用户基础属性特征和游戏行为属性特征进行特征工程处理。其中用户消费行为属性包括：消费总额、消费次数、周期内最后一次消费时间、周期内最后一次消费金额、平均消费金额、高单价商品消费频次、最大消费金额。S224: Extract basic user attributes and game behavior attribute features of paying users within a period, and perform feature engineering processing on the basic user attributes and game behavior attribute features; attribute, user consumption behavior attribute and user game behavior attribute, and use the method described in S222 to perform feature engineering on the user basic attribute feature and game behavior attribute feature. The user consumption behavior attributes include: total consumption, consumption times, the last consumption time in the cycle, the last consumption amount in the cycle, the average consumption amount, the consumption frequency of high-unit-price commodities, and the maximum consumption amount.
S225：利用梯度提升回归构建回归模型并通过K折交叉验证法调整参数；在一个优选的实施例中，使用梯度提升回归(Gradient boosting regression，GBR)建立回归模型，通过K折交叉验证法得到最优回归模型，预测用户的付费金额。S225: Use gradient boosting regression to build a regression model and adjust parameters through K-fold cross-validation method; in a preferred embodiment, use gradient boosting regression (Gradient boosting regression, GBR) to build a regression model, and obtain the maximum value through K-fold cross-validation method The optimal regression model predicts the user's payment amount.
本发明使用的梯度提升分类模型、梯度提升回归模型采用的K折交叉验证法，每次选出K-1个作为训练集，余下的1个作为测试集，此时获得K组训练集和测试集。选择一组训练集，用初始权重训练出一个基学习器，根据该基学习器的误差率来更新样本中的权重，使得该基学习器中误差率高的训练样本点权重变高，即使得误差率高的样本在下一个基学习器中得到重视；然后利用调整权重后的训练集来训练下一个基学习器，循环直到基学习器数量达到指定数量T；最后将这T个基学习器组合，得到强学习器；遍历K组训练集，得到K组强学习器，从中挑选均方误差最小的强学习器。使用梯度提升策略，可以得到灵活地处理各类型数据，包括离散型特征和连续型特征，对异常值的鲁棒性较强，同时比使用单一的模型得到的准确率也更高。The K-fold cross-validation method used by the gradient boosting classification model and the gradient boosting regression model used in the present invention selects K-1 as the training set each time, and the remaining 1 as the test set. At this time, K groups of training sets and test sets are obtained. set. Select a set of training sets, train a basic learner with initial weights, and update the weights in the samples according to the error rate of the basic learner, so that the weight of the training sample points with high error rate in the basic learner becomes higher, that is, the The samples with high error rate are valued in the next basic learner; then use the weighted training set to train the next basic learner, and loop until the number of basic learners reaches the specified number T; finally, these T basic learners are combined , get a strong learner; traverse K groups of training sets, get K groups of strong learners, and select the strong learner with the smallest mean square error. Using the gradient boosting strategy, it is possible to flexibly process various types of data, including discrete features and continuous features, with strong robustness to outliers, and higher accuracy than using a single model.
S226：制定用户消费等级规则；具体地，计算用户数据集周期T内付费的用户群的分位数，根据分位数制定5个用户消费区间，定义每个区间的用户消费等级，并按照从小到大的顺序依次定义为1到5级。S226: Formulate user consumption level rules; specifically, calculate the quantiles of the user groups paying in the user data set period T, formulate 5 user consumption intervals according to the quantiles, define the user consumption grades of each interval, and calculate the quantiles according to the quantiles. The order of the largest is defined as 1 to 5.
S227：将上述分类模型、回归模型和用户消费等级规则整合为用户消费预测模型。该用户消费预测模型可预测用户未来N天内是否付费，对于预测结果为付费的用户，通过回归模型得到预测的付费金额，将付费金额映射为用户消费等级并输出。S227: Integrate the above classification model, regression model and user consumption level rules into a user consumption prediction model. The user consumption prediction model can predict whether the user will pay in the next N days. For users whose prediction result is payment, the predicted payment amount is obtained through the regression model, and the payment amount is mapped to the user's consumption level and output.
所述基于用户数据建立游戏偏好标签的步骤中，由于用户对某类型游戏的偏好一般都具有时效性，即用户的喜爱程度会随着时间的增长而下降，因此，本发明采用改进的基于牛顿冷却定律(Newton's law of cooling)的时间衰减函数模型，并以此计算用户的游戏偏好标签，具体地，包括以下步骤：In the step of establishing a game preference label based on user data, because the user's preference for a certain type of game is generally time-sensitive, that is, the user's favorite degree will decrease with time. Therefore, the present invention adopts an improved Newton-based game. The time decay function model of Newton's law of cooling is used to calculate the user's game preference label. Specifically, the following steps are included:
S231：统计用户在注册每种类型游戏前产生相关行为事件数，其中所述相关行为包括：搜索游戏、点击平台内游戏介绍、论坛、评价等；每产生一条相关的事件记录，则对应的游戏类型初始喜爱度Tj增加一个单位值；S231: Count the number of related behavior events that the user generates before registering each type of game, wherein the related behaviors include: searching for games, clicking on game introductions on the platform, forums, evaluations, etc.; each time a related event record is generated, the corresponding game The initial favorite degree Tj of the type increases by one unit value;
S232：统计用户在该周期对每种类型游戏投入的时间以及在上一周期投入的时间，并以此计算出每种类型游戏时间增长量的增长率；若用户近期喜欢玩某个类型的游戏，则会投入大量时间，操作频繁且行为丰富；若用户对该游戏失去兴趣，往往会逐渐减少游戏时间及操作行为。因此，用户游戏时间增长量的增长率αi可表示为：S232: Count the time invested by the user in each type of game in this cycle and the time invested in the previous cycle, and calculate the growth rate of the time increase of each type of game based on this; if the user likes to play a certain type of game recently , it will spend a lot of time, operate frequently and have rich behaviors; if the user loses interest in the game, the game time and operation behaviors will be gradually reduced. Therefore, the growth rate αi of the user's game time increase can be expressed as:
其中，tj为某一周期，ti为另一游戏周期，αi为用户游戏时间增长量的增长率，ΔXj为用户在周期tj投入的游戏时间，ΔXi为用户在周期ti投入的游戏时间。Among them, tj is a certain period, ti is another game period, αi is the growth rate of the user's game time increase, ΔXj is the game time invested by the user in period tj , ΔXi is the user's game time in period ti Invested game time.
若αi为正，表示用户的喜爱度上升；反之下降。If αi is positive, it means that the user's favorability increases; otherwise, it decreases.
S233：根据以下方式计算每种游戏类型的用户喜爱度Ti值：S233: Calculate the user preference Ti value of each game type according to the following methods:
其中，αi为用户游戏时间增长量的增长率，Δx为用户投入的游戏时间，Tj为用户喜爱度的初始值，ti-tj表示时间间隔，M表示用户在某类游戏中的消费总额，λ为常量，表示消费额度对热度增长的比例，Ti表示用户在ti周期对某类游戏的喜爱度。Among them, αi is the growth rate of the user's game time increase, Δx is the game time invested by the user, Tj is the initial value of the user's favorite degree, ti -tj represents the time interval, and M represents the user's game time in a certain type of game. The total consumption, λ is a constant, represents the ratio of consumption to the increase in popularity, and Ti represents the user's preference for a certain type of game in the ti period.
S234：根据用户对周期内各游戏类型的喜爱度进行排序，根据喜爱度数值确定用户偏好的游戏类型。其中，对于用户初始喜爱度的计算规则，可根据具体的应用场景制定；对于通过初始喜爱度规则计算后结果值相差不大的同批次游戏类型，后续周期可通过直接比较T值得出用户对同批次游戏类型喜爱程度的排序。S234: Rank the game types in the cycle according to the user's favorite degree, and determine the game type preferred by the user according to the value of the favorite degree. Among them, the calculation rules for the user's initial favorite degree can be formulated according to specific application scenarios; for the same batch of game types whose results are not much different after calculation through the initial favorite degree rules, the user's preference can be determined by directly comparing the T value in subsequent cycles. The order of the favorite degree of the same batch of game types.
基于用户数据建立流失用户标签的步骤包括：The steps to create a churn user tag based on user data include:
S241：利用用户唯一标识将用户数据及标签关联，获得特征数据集；具体地，综合上述用户数据集以及标签数据，通过用户唯一标识关联得到特征数据集。其中，用户数据集包括但不限于用户注册年龄、注册IP、近N天登录次数、近N天评论次数、近N天的点赞次数等字段；基于标签建立模块，标签数据包括但不限于用户登录的城市等级、用户消费类型、消费指数、游戏喜爱程度值等字段。S241 : Associate the user data and the label by using the user's unique identifier to obtain a feature data set; specifically, combine the above-mentioned user data set and the label data, and obtain the feature data set through the association of the user's unique identifier. Among them, the user data set includes but is not limited to the user registration age, registration IP, login times in the past N days, comment times in the past N days, likes in the past N days and other fields; based on the tag building module, the tag data includes but not limited to the user The logged-in city level, user consumption type, consumption index, game love value and other fields.
S242：利用梯度提升树和特征权值对特征数据集进行特征选择；S242: Use the gradient boosting tree and the feature weight to perform feature selection on the feature dataset;
在一个实施例中，所述利用梯度提升树和特征权值对特征数据集进行特征选择的步骤之前，对特征数据集进行特征预处理，步骤包括：对于缺省值，丢弃或填充处理，对于丢弃缺省值占比大的样本，定性特征按0填充，定量特征值则按其均值填充；对于不同量纲，按无量纲化处理保证其处于同一规格，方便比较。所述无量纲化可以采用如标准化、区间缩放等现有技术中常见的数据无量纲化处理方式；对于信息冗余，将其包含的有效信息按区间划分，如近N天在线时长，只关心在线时长是否达到某个阈值，将其处理成0和1用于表示未达阈值和已达阈值；In one embodiment, before the step of performing feature selection on the feature dataset by using the gradient boosting tree and feature weights, feature preprocessing is performed on the feature dataset, and the step includes: for default values, discarding or filling processing, for Samples with a large proportion of default values are discarded, qualitative features are filled with 0, and quantitative feature values are filled with their mean values; for different dimensions, dimensionless processing is performed to ensure that they are in the same specification, which is convenient for comparison. The dimensionlessization can adopt the common data dimensionlessization processing methods in the prior art, such as standardization and interval scaling; for information redundancy, the effective information contained in it is divided into intervals, such as the online duration of nearly N days, only concerned about Whether the online time reaches a certain threshold, it is processed into 0 and 1 to indicate that the threshold has not been reached and the threshold has been reached;
所述利用梯度提升树和特征权值进行特征选择的步骤中，利用基于树的特征选择算法在保证准确度的最优的情况下，再根据特征权值来选择特征，其中根据特征权值来选择特征是考察特征与目标值相关性，采用机器学习的方法训练，得到各个特征的权值系数，移出权值系数较低的特征。在其它实施例中，本步骤也可利用过滤法(Filter)，通过考察特征是否发散并根据其发散性对各个特征进行评分，设定阈值或者待选择阈值的数量进行特征选择。In the step of using the gradient boosting tree and the feature weight for feature selection, the tree-based feature selection algorithm is used to select the feature according to the feature weight under the condition of ensuring the optimum accuracy, wherein the feature weight is selected according to the feature weight. The selection of features is to examine the correlation between the feature and the target value, use the machine learning method to train, obtain the weight coefficient of each feature, and remove the feature with lower weight coefficient. In other embodiments, this step can also use a filtering method (Filter), by examining whether the feature is divergent and scoring each feature according to its divergence, setting a threshold or the number of thresholds to be selected for feature selection.
S243：利用K折交叉验证法训练若干基学习模型，并根据若干基学习模型的输出结果构建融合模型；具体地，采用基于Stacking的融合方法来提高模型的预测能力，相对比单一预测模型，融合模型的优势在于它能够结合多种单一模型学习到的不同用户各种各样的行为特点，在多种环境下也能体现很好的健壮性，本步骤中构建融合模型包括以下步骤：S243: Use the K-fold cross-validation method to train several basic learning models, and construct a fusion model according to the output results of the several basic learning models; specifically, adopt the fusion method based on Stacking to improve the prediction ability of the model, compared with a single prediction model, fusion The advantage of the model is that it can combine various behavioral characteristics of different users learned by a variety of single models, and it can also reflect good robustness in a variety of environments. In this step, building a fusion model includes the following steps:
将特征选择及特征预处理后的数据集分为N等份，其中每个基学习模型通过训练N-1份数据集，剩余的1份作为测试集，将所有基学习模型的预测结果作为训练集并作为下一步的输入。其中基学习模型包括但不限于XGBoost、随机森林(Random Forest，RF)、逻辑回归(Logistic Regression，LR)和神经网络(Neural Network，NN)。在一个优选的实施例中，在训练基学习模型时采用K折交叉验证的方法区分不同模型的测试集和训练集，以增加模型的之间的差异，提高模型融合的效果，具体包括：遍历上述N-1份训练集，对其中遍历到的每一份训练集，均将其拆分为K等份，每次选出K-1个作为训练集，剩余的1份作为测试集，即获得K组训练集和测试集；对上述K组训练集和测试集分别采用K种不同的基学习模型训练，如XGBoost、RF、LR等，如图3所示；将K个基学习模型训练出模型的预测结果作为第二层模型融合的输入。其中，所述K种不同基学习模型可结合实际预测结果，舍弃部分效果不佳的模型。The data set after feature selection and feature preprocessing is divided into N equal parts, in which each basic learning model is trained by N-1 data sets, the remaining 1 is used as the test set, and the prediction results of all basic learning models are used as training. set and used as the input for the next step. The basic learning models include but are not limited to XGBoost, Random Forest (RF), Logistic Regression (LR), and Neural Network (NN). In a preferred embodiment, the K-fold cross-validation method is used to distinguish the test set and the training set of different models when training the basic learning model, so as to increase the difference between the models and improve the effect of model fusion, which specifically includes: traversing the The above N-1 training sets are divided into K equal parts for each traversed training set, K-1 are selected as the training set each time, and the remaining 1 is used as the test set, that is Obtain K groups of training sets and test sets; use K different basic learning models for training on the above K groups of training sets and test sets, such as XGBoost, RF, LR, etc., as shown in Figure 3; train K basic learning models The prediction result of the output model is used as the input of the second layer model fusion. Among them, the K different basic learning models can be combined with the actual prediction results, and some models with poor performance can be discarded.
S244：调用融合模型识别流失用户并生成流失用户标签信息。S244: Invoke the fusion model to identify the lost users and generate the label information of the lost users.
将基学习模型输出的用户流失概率值作为融合模型的输入。如图4所示，在融合层分别采用XGBoost和LR分别训练和预测用户流失概率，在一个实施例中，对该两个模型输出的概率值取均值作为融合层的最终输出概率。在其它实施例中，对该两个模型输出的概率值采用加权平均的方式作为融合层的最终输出概率。当概率均值大于设定的阈值时，则判定该用户为流失用户。The user churn probability value output by the basic learning model is used as the input of the fusion model. As shown in FIG. 4 , XGBoost and LR are used in the fusion layer to train and predict the user churn probability, respectively. In one embodiment, the average value of the probability values output by the two models is taken as the final output probability of the fusion layer. In other embodiments, a weighted average of the probability values output by the two models is used as the final output probability of the fusion layer. When the probability mean is greater than the set threshold, the user is determined to be a lost user.
输入待预测数据集并调用该融合模型进行计算，输出用户流失列表作为流失用户标签信息。Input the data set to be predicted and call the fusion model for calculation, and output the user churn list as the churn user label information.
在一个实施例中，所述游戏平台信息推送方法还包括以下步骤：将标签名称作为键，将该标签下的用户转换为该键对应的值的方式将所述用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签存储至数据库中；本步骤中通过利用位图算法将标签信息存储在数据库中，即利用一个bit位来标记某个元素对应的值(Value)，将键(Key)作为该元素本身，包括以下步骤：In one embodiment, the game platform information push method further includes the following steps: using the tag name as a key, converting the user under the tag into a value corresponding to the key, converting the user basic information tag, user consumption prediction The tags, game preference tags and lost user tags are stored in the database; in this step, the tag information is stored in the database by using a bitmap algorithm, that is, a bit bit is used to mark the value (Value) corresponding to an element, and the key ( Key) as the element itself, including the following steps:
根据标签模型得到的标签名称Lable_k、最大用户标识M及具有该标签的所有用户：The label name Lable_k, the maximum user ID M and all users with the label are obtained according to the label model:
<uid_1,uid_2,…,uid_n>；<uid_1,uid_2,…,uid_n>;
其中，所有用户唯一标识均映射为自增序列的整型数值，uid_1,uid_2,…,uid_n为该标签下所有用户的用户唯一标识；Among them, all user unique identifiers are mapped to integer values of self-incrementing sequence, uid_1,uid_2,...,uid_n are the user unique identifiers of all users under the label;
将<uid_1,uid_2,…,uid_n>通过位图算法，转化成位图数组L，其中L为：Convert <uid_1,uid_2,…,uid_n> into bitmap array L through bitmap algorithm, where L is:
<b_uid_1,b_uid_2,…,b_uid_k>；<b_uid_1,b_uid_2,…,b_uid_k>;
其中，k＝1+N/32，k为位图数组个数。Among them, k=1+N/32, k is the number of bitmap arrays.
将位图数组L转化成十六进制字符串，并以该标签名称Lable_k作为键(Key)，上述十六进制字符串作为键对应的值(Value)，存储在数据库中。其中，所述数据库包括但不限于Redis、MySQL、Mongo数据库。The bitmap array L is converted into a hexadecimal string, and the label name Lable_k is used as the key (Key), and the above-mentioned hexadecimal string is used as the value corresponding to the key (Value), and is stored in the database. Wherein, the database includes but is not limited to Redis, MySQL, and Mongo database.
一般情况下，一个整数的存储需要占用4个字节，即32个bit位，N个整数的存储，则需占用32N个bit的内存空间，而查询某个整数是否存在，需遍历N个整数来判断，时间复杂度为O(n)；本实施例所述存储方式通过使用一个bit位来表示一个整数，N个整数的存储只需占用N个bit的内存空间，内存变成了原来的1/32，并且，对于Redis数据库，当查询某个整数是否存在时，可采用Redis数据库的内置方法getbit进行判断，时间复杂为O(1)，查询速度由原来的线性阶降至常数阶。上述存储方式不仅减少了使用的内存空间，也大大地提高了查询性能，对多个标签的交集、并集的查询也能通过位运算快速得到结果。In general, the storage of an integer needs to occupy 4 bytes, that is, 32 bits, and the storage of N integers needs to occupy 32N bits of memory space, and to query whether an integer exists, it is necessary to traverse N integers To judge, the time complexity is O(n); the storage method described in this embodiment uses one bit to represent an integer, and the storage of N integers only needs to occupy N bits of memory space, and the memory becomes the original 1/32, and, for the Redis database, when querying whether an integer exists, the built-in method getbit of the Redis database can be used to judge, the time complexity is O(1), and the query speed is reduced from the original linear order to the constant order. The above storage method not only reduces the used memory space, but also greatly improves the query performance. The query for the intersection and union of multiple tags can also quickly obtain results through bit operations.
S3：根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送；S3: Group users according to user data and tags, and push information according to the assigned user groups;
所述根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送步骤中，所述用户分群方式可结合实际运营情况进行分群。例如：针对消费高于设定阈值和消费低于设定阈值的用户进行优惠券信息的推送，并监控其消费情况。具体地：定义每日消费M元以上或者每月消费累计N元以上的用户为VIP用户；定义每日消费m元以上或者每月消费累计n元以上并且为特定渠道的用户为渠道VIP用户；定义最后登录距离现在超过7天的用户为流失用户，最后登录距离现在4到6天区间内的用户为预流失用户，最后登录距离现在1到3天区间内的用户为活跃用户等。In the step of grouping users according to user data and tags and pushing information according to the assigned user groups, the user grouping method may be used for grouping according to actual operation conditions. For example, push coupon information for users whose consumption is higher than the set threshold and whose consumption is lower than the set threshold, and monitor their consumption. Specifically: define users with daily consumption of more than M yuan or accumulated monthly consumption of more than N yuan as VIP users; define users with daily consumption of more than m yuan or monthly consumption of more than n yuan and a specific channel as channel VIP users; Define users whose last login is more than 7 days away as churn users, users whose last login is within the interval of 4 to 6 days from now are pre-churn users, and users whose last login is between 1 and 3 days away from now are active users, etc.
根据已分配的用户群体，实施不同的优惠券信息推送，可以刺激用户消费，提高留存，例如：当需要测试满减券、折扣券这两种券在营销中的对用户消费的刺激程度时，可针对相同用户群体的做A/B测试，并通过测试对照以测试不同的优惠券类型对用户消费的刺激程度，从而制定最佳的优惠券方案。针对已推送优惠券信息的用户群体，监控其消费情况和留存情况。According to the assigned user groups, the implementation of different coupon information push can stimulate user consumption and improve retention. A/B test can be done for the same user group, and the best coupon scheme can be formulated by testing the degree of stimulation of different coupon types to user consumption through test comparison. For user groups who have pushed coupon information, monitor their consumption and retention.
针对流失用户及即将流失的用户进行召回信息推送，并监控其召回情况的步骤具体包括：根据流失用户标签获取已流失玩家、即将流失玩家信息，根据相应玩家的用户画像信息，制定召回信息并进行推送，其中，召回信息推送方式包括但不限于短信通知、代金券发放和回访沟通的方式。针对流失用户召回中成功唤醒的玩家进行运营监控，所述运营监控包括统计和展示所述用户的行为事件，包括登录、搜索关键词、充值、消费等，为运营人员提供提供数据可视化、数据交互以及数据监控功能。在一个其它的实施例中，该步骤还包括监控信息推送前后用户数量的变化并进行展示，所述推送前后用户数量包括：玩家咨询数量、跟进玩家数量、召回玩家数量及其召回比等，方便对该信息平台推送的效果进行直观的了解。The steps of pushing recall information for lost users and users who are about to be lost, and monitoring their recalls include: obtaining information about lost players and players about to be lost according to the lost user tags, formulating recall information according to the user portrait information of the corresponding players, and carrying out Push, among which, the methods of pushing recall information include but are not limited to SMS notification, voucher issuance and return visit communication. Operation monitoring is carried out for players who are successfully awakened in the recall of lost users. The operation monitoring includes statistics and display of the user's behavior events, including login, search keywords, recharge, consumption, etc., to provide data visualization and data interaction for operators. and data monitoring functions. In another embodiment, the step further includes monitoring and displaying changes in the number of users before and after the information is pushed, and the number of users before and after the push includes: the number of players inquired, the number of follow-up players, the number of recalled players and their recall ratio, etc., It is convenient to intuitively understand the effect of the information platform push.
相对于现有技术，本发明提供的一种游戏平台信息推送方法，有效地结合服务器中海量用户数据集，可以快速、精准地定位用户，全方位、多角度地掌握用户特征，并通过已分配的用户群体进行信息推送，实现推送内容与用户喜好之间的快速匹配，提高了匹配的准确性，节省了网络资源；本发明采用融合模型的方式对用户的流失概率进行分析和预测，极大地提高了模型的泛化能力，用户流失预警准确性高。本发明还通过利用位图算法对标签信息进行存储，相对于传统的结构化的标签存储方案，在数据量大的情况下极大地提高了查询效率。Compared with the prior art, the method for pushing game platform information provided by the present invention effectively combines the massive user data sets in the server, can quickly and accurately locate users, grasp user characteristics in an all-round and multi-angle manner, and pass the assigned user characteristics. The user group pushes information, realizes the rapid matching between the pushed content and the user's preferences, improves the accuracy of matching, and saves network resources; the present invention adopts the fusion model to analyze and predict the loss probability of users, which greatly improves the accuracy of matching and saves network resources. The generalization ability of the model is improved, and the accuracy of user churn warning is high. The invention also stores the label information by using the bitmap algorithm, compared with the traditional structured label storage scheme, the query efficiency is greatly improved in the case of a large amount of data.
本发明还提供了一种游戏平台信息推送系统，如图5所示，所述游戏平台信息推送系统包括：The present invention also provides a game platform information push system, as shown in FIG. 5 , the game platform information push system includes:
数据获取模块1，用于获取服务器中的用户数据；A data acquisition module 1 is used to acquire user data in the server;
所述数据获取模块1包括：The data acquisition module 1 includes:
服务器数据获取模块，用于获取上报至服务器数据库中的结构化、非结构化数据和上报至服务器本地磁盘日志文件；The server data acquisition module is used to acquire structured and unstructured data reported to the server database and to the local disk log file of the server;
数据处理模块，用于对于非结构化数据，将系统中对应用户产生的日志进行组合，拼接成一个对应用户唯一标识符的独立文本，并利用分词工具进行清洗。The data processing module is used to combine the logs generated by the corresponding users in the system for unstructured data, splicing them into an independent text corresponding to the user's unique identifier, and use the word segmentation tool for cleaning.
标签建立模块2，用于基于用户数据建立用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签；所述标签建立模块2包括：用户基础信息标签建立模块、用户消费预测标签建立模块、游戏偏好标签建立模块和流失用户标签建立模块。The label establishment module 2 is used to establish user basic information label, user consumption prediction label, game preference label and lost user label based on user data; the label establishment module 2 includes: user basic information label establishment module, user consumption prediction label establishment module , the game preference tag creation module and the lost user tag creation module.
其中，所述用户基础信息标签建立模块包括：注册行为标签建立单元，用于获取用户注册时间，注册时长，注册设备等信息建立注册行为标签；Wherein, the user basic information label establishment module includes: a registration behavior label establishment unit, which is used to obtain user registration time, registration duration, registered equipment and other information to establish a registration behavior label;
用户活跃类型标签建立单元，用于统计用户登录时间点，计算用户活跃时段，建立用户活跃类型标签；The user activity type label establishment unit is used to count the user login time points, calculate the user activity period, and establish the user activity type label;
用户所在城市等级标签建立单元，用于统计用户登录IP解析出用户登录地址，建立用户所在城市等级标签；The city level label establishment unit where the user is located is used to parse the user login address from the statistics of the user login IP, and establish the city level label of the user;
用户活跃指数标签建立单元，用于根据用户登录时间，计算用户活跃指数，建立用户活跃指数标签；The user activity index label establishment unit is used to calculate the user activity index according to the user login time, and establish the user activity index label;
用户消费标签建立单元，用于根据用户参与活动的消费次数、消费金额，结合用户浏览行为数据，建立用户消费行为标签和用户消费指数标签。The user consumption label establishment unit is used to establish the user consumption behavior label and the user consumption index label according to the consumption times and consumption amount of the user participating in the activity, combined with the user browsing behavior data.
所述用户消费预测标签建立模块包括：The user consumption prediction label establishment module includes:
数据采集单元，用于从服务器日志采集并预处理用户数据；A data collection unit for collecting and preprocessing user data from server logs;
特征处理单元，用于提取一个周期内付费用户和无付费用户的用户基础属性特征和游戏行为属性特征，并对所述用户基础属性特征和游戏行为属性特征进行特征工程处理；A feature processing unit, used for extracting basic user attribute features and game behavior attribute features of paying users and non-paying users in a cycle, and performing feature engineering on the basic user attribute features and game behavior attribute features;
分类模型构建单元，用于利用梯度提升决策树构建分类模型并通过K折交叉验证法调整参数；The classification model building unit is used to build a classification model by using gradient boosting decision tree and adjust parameters through K-fold cross-validation method;
回归模型特征处理单元，用于提取一个周期内付费用户的用户基础属性和游戏行为属性特征，并对所述用户基础属性和游戏行为属性特征进行特征工程处理；A regression model feature processing unit, used for extracting basic user attributes and game behavior attribute features of paying users within a cycle, and performing feature engineering on the user basic attributes and game behavior attribute features;
回归模型构建单元，用于利用梯度提升回归构建回归模型并通过K折交叉验证法调整参数；The regression model building unit is used to build a regression model by using gradient boosting regression and adjust parameters through K-fold cross-validation method;
用户消费规则制定单元，用于制定用户消费等级规则；A user consumption rule formulation unit, which is used to formulate user consumption level rules;
整合单元，用于将上述分类模型、回归模型和用户消费等级规则整合为用户消费预测模型。The integration unit is used to integrate the above classification model, regression model and user consumption level rules into a user consumption prediction model.
所述游戏偏好标签建立模块包括：The game preference tag establishment module includes:
初始喜爱度计算单元，用于统计用户在注册每种类型游戏前产生相关行为事件数，并以此计算获取对应类型游戏的初始喜爱度；The initial favorite degree calculation unit is used to count the number of related behavior events generated by the user before registering each type of game, and use this calculation to obtain the initial favorite degree of the corresponding type of game;
增长率计算单元，用于统计用户在该周期对每种类型游戏投入的时间以及在上一周期投入的时间，并以此计算出每种类型游戏时间增长量的增长率；The growth rate calculation unit is used to count the time invested by the user in each type of game in this cycle and the time invested in the previous cycle, and calculate the growth rate of the time increase of each type of game based on this;
喜爱度计算单元，用于根据以下方式计算每种类型的Ti值；A favorability calculation unit for calculating theTi value of each type according to the following manner;
其中，αi为用户游戏时间增长量的增长率，Δx为用户投入的游戏时间，Tj表示用户喜爱度的初始值，ti-tj表示时间间隔，M表示用户在某类游戏中的消费总额，λ为常量，表示消费额度对热度增长的比例，Ti表示用户在ti周期对某类游戏的喜爱度。Among them, αi is the growth rate of the user's game time increase, Δx is the game time invested by the user, Tj represents the initial value of the user's favorite degree, ti -tj represents the time interval, and M represents the user's game time in a certain type of game. The total consumption, λ is a constant, represents the ratio of consumption to the increase in popularity, and Ti represents the user's preference for a certain type of game in the ti period.
用户偏好确定单元，用于根据用户周期内各游戏类型的喜爱度进行排序，根据喜爱度数值确定用户偏好的游戏类型。The user preference determining unit is configured to sort the game types according to the preference degree of each game type in the user cycle, and determine the game type preferred by the user according to the value of the preference degree.
所述流失用户标签建立模块包括：The lost user label establishment module includes:
特征数据集获取单元，用于利用用户唯一标识将用户数据及标签关联，获得特征数据集；A feature data set acquisition unit, used to associate user data and tags by using the user's unique identifier to obtain a feature data set;
特征处理单元，用于对特征数据集进行特征预处理并利用梯度提升树和特征权值进行特征选择；The feature processing unit is used to perform feature preprocessing on the feature dataset and use gradient boosting trees and feature weights to perform feature selection;
融合模型构建单元，用于利用K折交叉验证法训练基学习模型，并构建融合模型；The fusion model construction unit is used to train the base learning model by using the K-fold cross-validation method, and construct the fusion model;
流失用户标签生成单元，用于调用融合模型识别流失用户并生成流失用户标签信息。The lost user label generation unit is used to call the fusion model to identify the lost user and generate the lost user label information.
标签存储模块3，用于将用户基础信息标签、用户消费预测标签、游戏偏好标签和流失用户标签的标签名称作为键，标签下的用户转换为该键对应的值，将该键和值以相互对应的方式存储至数据库中；The label storage module 3 is used to use the user basic information label, the user consumption prediction label, the game preference label and the label name of the lost user label as the key, and the user under the label is converted into the value corresponding to the key, and the key and value are mutually The corresponding method is stored in the database;
所述标签存储模块3包括：The label storage module 3 includes:
获取单元，用于获取标签名称、最大用户标识及具有该标签的所有用户的唯一标识；The obtaining unit is used to obtain the label name, the maximum user ID and the unique ID of all users with the label;
转换单元，用于将标签中所有用户的唯一标识通过位图算法转换为k位的位图数组；其中，k＝1+N/32，N为标签的用户个数；The conversion unit is used to convert the unique identifiers of all users in the label into a bitmap array of k bits through a bitmap algorithm; wherein, k=1+N/32, and N is the number of users of the label;
存储单元，用于将位图数组转换为十六进制字符串，并以标签名称作为键，所述十六进制字符串作为该键对应的值，存储在数据库中。The storage unit is used to convert the bitmap array into a hexadecimal string, and use the label name as a key, and the hexadecimal string is stored in the database as a value corresponding to the key.
信息推送模块4，用于根据用户数据及标签对用户进行分群并根据已分配的用户群体进行信息推送。The information push module 4 is used to group users according to user data and tags and push information according to the assigned user groups.
所述信息推送模块4包括：The information push module 4 includes:
消费推送单元，用于针对消费高于设定阈值和消费低于设定阈值的用户进行优惠券信息的推送，并监控其消费情况；The consumption push unit is used to push coupon information for users whose consumption is higher than the set threshold and whose consumption is lower than the set threshold, and monitor their consumption;
流失召回单元，用于针对流失用户及即将流失的用户进行召回信息推送，并监控其召回用户数量；The lost recall unit is used to push recall information for lost users and users who are about to be lost, and monitor the number of recalled users;
监控单元，用于监控信息推送前后用户数量的变化并进行展示。The monitoring unit is used to monitor and display changes in the number of users before and after the information is pushed.
相对于现有技术，本发明提供了一整套系统化的模块共同协作，搭建了一个智能化的用户运营平台，适用于电商、社交网络等多种领域的平台运营需求，所述游戏平台信息推送系统有效地结合了海量用户数据集，全方位、多角度分析了用户特征，通过对用户进行分群并根据已分配的用户群体进行信息推送，实现推送内容与用户喜好之间的快速匹配，提高了匹配的准确性，节省了网络资源。Compared with the prior art, the present invention provides a set of systematic modules to cooperate together to build an intelligent user operation platform, which is suitable for the platform operation requirements in various fields such as e-commerce and social networks. The game platform information The push system effectively combines massive user data sets, analyzes user characteristics in an all-round and multi-angle manner. The matching accuracy is improved and network resources are saved.
本发明还提供一种计算机可读存储介质，其上储存有计算机程序，该计算机程序被处理器执行时实现如上述游戏平台信息推送方法的步骤。The present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of the above method for pushing game platform information.
本发明可采用在一个或多个其中包含有程序代码的存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。计算机可读储存介质包括永久性和非永久性、可移动和非可移动媒体，可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括但不限于：相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。The present invention may take the form of a computer program product embodied on one or more storage media having program code embodied therein, including but not limited to disk storage, CD-ROM, optical storage, and the like. Computer-readable storage media includes both persistent and non-permanent, removable and non-removable media, and storage of information can be implemented by any method or technology. Information may be computer readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Flash Memory or other memory technology, Compact Disc Read Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
本发明还提供一种计算机设备，包括储存器、处理器以及储存在所述储存器中并可被所述处理器执行的计算机程序，所述处理器执行所述计算机程序时实现如上述游戏平台信息推送方法的步骤。The present invention also provides a computer device, comprising a storage, a processor, and a computer program stored in the storage and executable by the processor, when the processor executes the computer program, the above-mentioned game platform is implemented The steps of the information push method.
本发明并不局限于上述实施方式，如果对本发明的各种改动或变形不脱离本发明的精神和范围，倘若这些改动和变形属于本发明的权利要求和等同技术范围之内，则本发明也意图包含这些改动和变形。The present invention is not limited to the above-mentioned embodiments. If various changes or modifications of the present invention do not depart from the spirit and scope of the present invention, and if these changes and modifications belong to the claims of the present invention and the equivalent technical scope, then the present invention is also Intended to contain these alterations and variants.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910491711.6ACN110222267B (en) | 2019-06-06 | 2019-06-06 | Game platform information pushing method, system, storage medium and equipment |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910491711.6ACN110222267B (en) | 2019-06-06 | 2019-06-06 | Game platform information pushing method, system, storage medium and equipment |
| Publication Number | Publication Date |
|---|---|
| CN110222267Atrue CN110222267A (en) | 2019-09-10 |
| CN110222267B CN110222267B (en) | 2023-07-25 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910491711.6AActiveCN110222267B (en) | 2019-06-06 | 2019-06-06 | Game platform information pushing method, system, storage medium and equipment |
| Country | Link |
|---|---|
| CN (1) | CN110222267B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111083211A (en)* | 2019-12-09 | 2020-04-28 | 广州探途网络技术有限公司 | User touch method and system based on big data platform |
| CN111127426A (en)* | 2019-12-23 | 2020-05-08 | 山东大学齐鲁医院 | A method and system for evaluating the cleanliness of gastric mucosa based on deep learning |
| CN111191151A (en)* | 2019-12-20 | 2020-05-22 | 上海淇玥信息技术有限公司 | Method and device for pushing information based on POI (Point of interest) tag and electronic equipment |
| CN111224731A (en)* | 2019-12-26 | 2020-06-02 | 支付宝(杭州)信息技术有限公司 | Content pushing method, device and equipment based on voice broadcast |
| CN111291816A (en)* | 2020-02-17 | 2020-06-16 | 支付宝(杭州)信息技术有限公司 | Method and device for carrying out feature processing aiming at user classification model |
| CN111428129A (en)* | 2020-02-28 | 2020-07-17 | 海南和方信息科技有限公司 | Game information pushing method, system and storage medium thereof |
| CN111737575A (en)* | 2020-06-19 | 2020-10-02 | 北京字节跳动网络技术有限公司 | Content distribution method and device, readable medium and electronic equipment |
| CN111773732A (en)* | 2020-09-04 | 2020-10-16 | 完美世界(北京)软件科技发展有限公司 | Method, device and device for detecting target game users |
| CN111953763A (en)* | 2020-08-06 | 2020-11-17 | 腾讯科技（深圳）有限公司 | Business data pushing method and device and storage medium |
| CN112016961A (en)* | 2020-08-26 | 2020-12-01 | 北京字节跳动网络技术有限公司 | Push method, apparatus, electronic device, and computer-readable storage medium |
| CN112015975A (en)* | 2020-07-15 | 2020-12-01 | 北京淇瑀信息科技有限公司 | Financial user-oriented information pushing method and device based on Newton's cooling law |
| CN112426724A (en)* | 2020-11-30 | 2021-03-02 | 北京达佳互联信息技术有限公司 | Game user matching method and device, electronic equipment and storage medium |
| CN112446763A (en)* | 2020-11-27 | 2021-03-05 | 广州三七互娱科技有限公司 | Service recommendation method and device and electronic equipment |
| CN112619165A (en)* | 2020-12-18 | 2021-04-09 | 咪咕互动娱乐有限公司 | Game user selection method, system, server and storage medium |
| CN112651433A (en)* | 2020-12-17 | 2021-04-13 | 广州锦行网络科技有限公司 | Abnormal behavior analysis method for privileged account |
| CN112734463A (en)* | 2020-12-30 | 2021-04-30 | 咪咕音乐有限公司 | Service information sending method and device, electronic equipment and storage medium |
| CN113014476A (en)* | 2021-03-17 | 2021-06-22 | 维沃移动通信有限公司 | Group creation method and device |
| CN113082725A (en)* | 2021-03-08 | 2021-07-09 | 杭州电魂网络科技股份有限公司 | Game user grouping method and device, electronic equipment and storage medium |
| CN113329058A (en)* | 2021-04-30 | 2021-08-31 | 青岛以萨数据技术有限公司 | Data pushing method and device and storage medium |
| CN113413610A (en)* | 2021-06-21 | 2021-09-21 | 网易（杭州）网络有限公司 | Game item recommendation method, equipment and computer-readable storage medium |
| CN113572753A (en)* | 2021-07-16 | 2021-10-29 | 北京淇瑀信息科技有限公司 | User equipment authentication method and device based on Newton's cooling law |
| CN113706202A (en)* | 2021-08-31 | 2021-11-26 | 杭州群核信息技术有限公司 | Recall strategy generating method based on low-steady-state user identification and early warning |
| CN113742388A (en)* | 2020-05-29 | 2021-12-03 | 北京顺源开华科技有限公司 | Data pushing method and device, storage medium and electronic equipment |
| CN113837780A (en)* | 2020-06-23 | 2021-12-24 | 上海莉莉丝科技股份有限公司 | Information delivery method, system, device and medium |
| CN114140159A (en)* | 2021-11-30 | 2022-03-04 | 深圳市风行趋势科技有限公司 | Method and device for analyzing user consumption behavior based on oil product order data |
| CN114637826A (en)* | 2020-12-16 | 2022-06-17 | 中移动信息技术有限公司 | User behavior classification audit method, device, equipment and computer storage medium |
| CN114832386A (en)* | 2022-04-26 | 2022-08-02 | 江苏果米文化发展有限公司 | Game user intelligent management system based on big data analysis |
| TWI776742B (en)* | 2021-11-29 | 2022-09-01 | 愛酷智能科技股份有限公司 | System for analyzing user behavior in information exchange platform |
| CN115034830A (en)* | 2022-06-29 | 2022-09-09 | 腾云天宇科技(苏州)有限公司 | Method, apparatus, device, storage medium and program product for data processing |
| CN115129809A (en)* | 2022-06-06 | 2022-09-30 | 网易（杭州）网络有限公司 | Method, device, electronic device and storage medium for determining user activity |
| CN115168740A (en)* | 2022-09-06 | 2022-10-11 | 网娱互动科技(北京)股份有限公司 | Method and system for generating marketing task based on big data analysis |
| CN115531886A (en)* | 2022-10-08 | 2022-12-30 | 广州易幻网络科技有限公司 | User and equipment data management method, system and storage medium |
| CN116492693A (en)* | 2023-06-28 | 2023-07-28 | 北京乐盟互动科技有限公司 | Method and device for promoting online game data, electronic equipment and storage medium |
| CN116757750A (en)* | 2023-06-05 | 2023-09-15 | 广州盈风网络科技有限公司 | Operation pushing method, device, equipment and medium based on loss rate prediction |
| CN117520410A (en)* | 2023-11-03 | 2024-02-06 | 华青融天(北京)软件股份有限公司 | Service data processing method, device, electronic equipment and computer readable medium |
| CN119904277A (en)* | 2025-01-10 | 2025-04-29 | 德州聚安特安全服务有限公司 | A game promotion system and method based on data analysis |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105912729A (en)* | 2016-05-19 | 2016-08-31 | 华南理工大学 | Three-dimensional time sequence dynamic model based mobile phone game recommendation system |
| CN106097114A (en)* | 2016-06-27 | 2016-11-09 | 江苏易乐网络科技有限公司 | A kind of game public sentiment monitoring and player's behavior analysis cloud service platform |
| CN107203899A (en)* | 2017-05-03 | 2017-09-26 | 北京小米移动软件有限公司 | Information-pushing method and equipment |
| CN107609708A (en)* | 2017-09-25 | 2018-01-19 | 广州赫炎大数据科技有限公司 | A kind of customer loss Forecasting Methodology and system based on mobile phone games shop |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105912729A (en)* | 2016-05-19 | 2016-08-31 | 华南理工大学 | Three-dimensional time sequence dynamic model based mobile phone game recommendation system |
| CN106097114A (en)* | 2016-06-27 | 2016-11-09 | 江苏易乐网络科技有限公司 | A kind of game public sentiment monitoring and player's behavior analysis cloud service platform |
| CN107203899A (en)* | 2017-05-03 | 2017-09-26 | 北京小米移动软件有限公司 | Information-pushing method and equipment |
| CN107609708A (en)* | 2017-09-25 | 2018-01-19 | 广州赫炎大数据科技有限公司 | A kind of customer loss Forecasting Methodology and system based on mobile phone games shop |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111083211A (en)* | 2019-12-09 | 2020-04-28 | 广州探途网络技术有限公司 | User touch method and system based on big data platform |
| CN111083211B (en)* | 2019-12-09 | 2023-08-11 | 广州探途网络技术有限公司 | User touch method and system based on big data platform |
| CN111191151A (en)* | 2019-12-20 | 2020-05-22 | 上海淇玥信息技术有限公司 | Method and device for pushing information based on POI (Point of interest) tag and electronic equipment |
| CN111191151B (en)* | 2019-12-20 | 2023-08-25 | 上海淇玥信息技术有限公司 | Method and device for pushing information based on POI (point of interest) tag and electronic equipment |
| CN111127426B (en)* | 2019-12-23 | 2020-12-01 | 山东大学齐鲁医院 | A method and system for evaluating the cleanliness of gastric mucosa based on deep learning |
| CN111127426A (en)* | 2019-12-23 | 2020-05-08 | 山东大学齐鲁医院 | A method and system for evaluating the cleanliness of gastric mucosa based on deep learning |
| CN111224731A (en)* | 2019-12-26 | 2020-06-02 | 支付宝(杭州)信息技术有限公司 | Content pushing method, device and equipment based on voice broadcast |
| CN111224731B (en)* | 2019-12-26 | 2021-06-08 | 支付宝(杭州)信息技术有限公司 | Content pushing method, device and equipment based on voice broadcast |
| CN111291816B (en)* | 2020-02-17 | 2021-08-06 | 支付宝(杭州)信息技术有限公司 | Method and device for carrying out feature processing aiming at user classification model |
| CN111291816A (en)* | 2020-02-17 | 2020-06-16 | 支付宝(杭州)信息技术有限公司 | Method and device for carrying out feature processing aiming at user classification model |
| CN111428129A (en)* | 2020-02-28 | 2020-07-17 | 海南和方信息科技有限公司 | Game information pushing method, system and storage medium thereof |
| CN113742388A (en)* | 2020-05-29 | 2021-12-03 | 北京顺源开华科技有限公司 | Data pushing method and device, storage medium and electronic equipment |
| CN111737575A (en)* | 2020-06-19 | 2020-10-02 | 北京字节跳动网络技术有限公司 | Content distribution method and device, readable medium and electronic equipment |
| CN111737575B (en)* | 2020-06-19 | 2023-11-14 | 北京字节跳动网络技术有限公司 | Content distribution method, content distribution device, readable medium and electronic equipment |
| CN113837780A (en)* | 2020-06-23 | 2021-12-24 | 上海莉莉丝科技股份有限公司 | Information delivery method, system, device and medium |
| CN112015975B (en)* | 2020-07-15 | 2023-11-14 | 北京淇瑀信息科技有限公司 | Information pushing method and device for financial users based on Newton's law of cooling |
| CN112015975A (en)* | 2020-07-15 | 2020-12-01 | 北京淇瑀信息科技有限公司 | Financial user-oriented information pushing method and device based on Newton's cooling law |
| CN111953763B (en)* | 2020-08-06 | 2021-06-25 | 腾讯科技（深圳）有限公司 | Business data pushing method and device and storage medium |
| CN111953763A (en)* | 2020-08-06 | 2020-11-17 | 腾讯科技（深圳）有限公司 | Business data pushing method and device and storage medium |
| CN112016961A (en)* | 2020-08-26 | 2020-12-01 | 北京字节跳动网络技术有限公司 | Push method, apparatus, electronic device, and computer-readable storage medium |
| CN111773732A (en)* | 2020-09-04 | 2020-10-16 | 完美世界(北京)软件科技发展有限公司 | Method, device and device for detecting target game users |
| CN112446763B (en)* | 2020-11-27 | 2024-11-19 | 广州三七互娱科技有限公司 | Service recommendation method, device and electronic equipment |
| CN112446763A (en)* | 2020-11-27 | 2021-03-05 | 广州三七互娱科技有限公司 | Service recommendation method and device and electronic equipment |
| CN112426724A (en)* | 2020-11-30 | 2021-03-02 | 北京达佳互联信息技术有限公司 | Game user matching method and device, electronic equipment and storage medium |
| CN112426724B (en)* | 2020-11-30 | 2023-07-25 | 北京达佳互联信息技术有限公司 | Matching method and device for game users, electronic equipment and storage medium |
| CN114637826B (en)* | 2020-12-16 | 2025-09-09 | 中移动信息技术有限公司 | User behavior classification auditing method, device, equipment and computer storage medium |
| CN114637826A (en)* | 2020-12-16 | 2022-06-17 | 中移动信息技术有限公司 | User behavior classification audit method, device, equipment and computer storage medium |
| CN112651433B (en)* | 2020-12-17 | 2021-12-14 | 广州锦行网络科技有限公司 | Abnormal behavior analysis method for privileged account |
| CN112651433A (en)* | 2020-12-17 | 2021-04-13 | 广州锦行网络科技有限公司 | Abnormal behavior analysis method for privileged account |
| CN112619165A (en)* | 2020-12-18 | 2021-04-09 | 咪咕互动娱乐有限公司 | Game user selection method, system, server and storage medium |
| CN112734463A (en)* | 2020-12-30 | 2021-04-30 | 咪咕音乐有限公司 | Service information sending method and device, electronic equipment and storage medium |
| CN113082725A (en)* | 2021-03-08 | 2021-07-09 | 杭州电魂网络科技股份有限公司 | Game user grouping method and device, electronic equipment and storage medium |
| CN113014476B (en)* | 2021-03-17 | 2023-04-07 | 维沃移动通信有限公司 | Group creation method and device |
| CN113014476A (en)* | 2021-03-17 | 2021-06-22 | 维沃移动通信有限公司 | Group creation method and device |
| CN113329058A (en)* | 2021-04-30 | 2021-08-31 | 青岛以萨数据技术有限公司 | Data pushing method and device and storage medium |
| CN113329058B (en)* | 2021-04-30 | 2022-10-04 | 青岛以萨数据技术有限公司 | Data pushing method and device and storage medium |
| CN113413610A (en)* | 2021-06-21 | 2021-09-21 | 网易（杭州）网络有限公司 | Game item recommendation method, equipment and computer-readable storage medium |
| CN113572753B (en)* | 2021-07-16 | 2023-03-14 | 北京淇瑀信息科技有限公司 | User equipment authentication method and device based on Newton's cooling law |
| CN113572753A (en)* | 2021-07-16 | 2021-10-29 | 北京淇瑀信息科技有限公司 | User equipment authentication method and device based on Newton's cooling law |
| CN113706202A (en)* | 2021-08-31 | 2021-11-26 | 杭州群核信息技术有限公司 | Recall strategy generating method based on low-steady-state user identification and early warning |
| TWI776742B (en)* | 2021-11-29 | 2022-09-01 | 愛酷智能科技股份有限公司 | System for analyzing user behavior in information exchange platform |
| CN114140159B (en)* | 2021-11-30 | 2024-12-13 | 深圳市风行趋势科技有限公司 | Method and device for analyzing user consumption behavior based on oil product order data |
| CN114140159A (en)* | 2021-11-30 | 2022-03-04 | 深圳市风行趋势科技有限公司 | Method and device for analyzing user consumption behavior based on oil product order data |
| CN114832386A (en)* | 2022-04-26 | 2022-08-02 | 江苏果米文化发展有限公司 | Game user intelligent management system based on big data analysis |
| CN114832386B (en)* | 2022-04-26 | 2024-05-14 | 江苏果米文化发展有限公司 | Game user intelligent management system based on big data analysis |
| CN115129809A (en)* | 2022-06-06 | 2022-09-30 | 网易（杭州）网络有限公司 | Method, device, electronic device and storage medium for determining user activity |
| CN115129809B (en)* | 2022-06-06 | 2025-10-03 | 网易（杭州）网络有限公司 | Method, device, electronic device and storage medium for determining user activity |
| CN115034830A (en)* | 2022-06-29 | 2022-09-09 | 腾云天宇科技(苏州)有限公司 | Method, apparatus, device, storage medium and program product for data processing |
| CN115168740A (en)* | 2022-09-06 | 2022-10-11 | 网娱互动科技(北京)股份有限公司 | Method and system for generating marketing task based on big data analysis |
| CN115531886A (en)* | 2022-10-08 | 2022-12-30 | 广州易幻网络科技有限公司 | User and equipment data management method, system and storage medium |
| CN116757750A (en)* | 2023-06-05 | 2023-09-15 | 广州盈风网络科技有限公司 | Operation pushing method, device, equipment and medium based on loss rate prediction |
| CN116492693A (en)* | 2023-06-28 | 2023-07-28 | 北京乐盟互动科技有限公司 | Method and device for promoting online game data, electronic equipment and storage medium |
| CN117520410A (en)* | 2023-11-03 | 2024-02-06 | 华青融天(北京)软件股份有限公司 | Service data processing method, device, electronic equipment and computer readable medium |
| CN119904277A (en)* | 2025-01-10 | 2025-04-29 | 德州聚安特安全服务有限公司 | A game promotion system and method based on data analysis |
| Publication number | Publication date |
|---|---|
| CN110222267B (en) | 2023-07-25 |
| Publication | Publication Date | Title |
|---|---|---|
| CN110222267B (en) | Game platform information pushing method, system, storage medium and equipment | |
| CN108550068B (en) | Personalized commodity recommendation method and system based on user behavior analysis | |
| Keneshloo et al. | Predicting the popularity of news articles | |
| CN102394798B (en) | Multi-feature based prediction method of propagation behavior of microblog information and system thereof | |
| KR101452082B1 (en) | Research mission identification | |
| CN106919619A (en) | A kind of commercial articles clustering method, device and electronic equipment | |
| CN107016068A (en) | Knowledge mapping construction method and device | |
| CN111552882B (en) | A kind of news influence calculation method, device, computer equipment and storage medium | |
| US20140143013A1 (en) | System and method for analyzing social media trends | |
| CN110245355B (en) | Text topic detection method, device, server and storage medium | |
| Yang et al. | Measuring topic network centrality for identifying technology and technological development in online communities | |
| CN102402594A (en) | Rich media personalized recommendation method | |
| CN110717089A (en) | User behavior analysis system and method based on weblog | |
| CN108665513A (en) | Drawing practice based on user behavior data and device | |
| Zhang et al. | Author impact: Evaluations, predictions, and challenges | |
| CN104965930A (en) | Big data based emergency evolution analysis method | |
| Utama et al. | Scientific articles recommendation system based on user’s relatedness using item-based collaborative filtering method | |
| CN114528416B (en) | Enterprise public opinion environment monitoring method and system based on big data | |
| CN116894684B (en) | Big data-based computer data processing method and system | |
| CN109063944A (en) | City banking index analysis method and device based on big data analysis technology | |
| Yan et al. | An e-commerce coupon target population positioning model based on random forest and extreme gradient boosting | |
| CN110766429A (en) | Data value evaluation system and method | |
| CN119862334B (en) | C-terminal user data intelligent management system based on large model | |
| CN111460300A (en) | Network content pushing method and device and storage medium | |
| CN116701729B (en) | Network public opinion detection system and detection method |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |