For King of Glory, let’s talk about personalized precision marketing based on big data

Today I will talk about attracting new users, preventing the loss of paying users , and recommending game modes based on big data operations .

1. Prevent loss

For paying players

• 1.1 Factors that affect the loss of paying players: frequency of item purchases, payment amount, game duration, game frequency, game level, win rate, etc.
• 1.2 Users who have churned in the past three months (customizable) are divided into a test set and a validation set. Through the factor analysis algorithm, strong correlation factors are obtained. Through the logistic regression algorithm, the regression equation of lost paying players is obtained. These churn factors of non-churned users are put into the regression equation for calculation.
• 1.3 Predict whether paying users are approaching the critical point of churn and issue churn warnings. Conduct targeted recovery marketing activities through qualitative analysis of common churn and user interviews

For free players

• 1.4 Factors that cause free players to churn: game duration, game frequency, game level, win rate, etc. Through the factor analysis algorithm, strong correlation factors are obtained. Through the logistic regression algorithm, the regression equation of lost free players was obtained. These churn factors of non-churned users are put into the regression equation for calculation.
• 1.5 Predict free users and see if they are close to the critical point of churn, and issue churn warnings. Conduct targeted recovery marketing activities through qualitative analysis of common churn and user interviews

Some negative feedback factors also need to be considered in preventing loss. For example, some users are often disconnected, the network latency is high, the regional server often crashes, the client often crashes, and so on. In addition, time decay also needs to be considered when dealing with game frequency. The frequency of a player's activity is different if he played the game 30 times in the first few days of a month than if he played the game once a day for a month. Maybe as time goes by, the previous frequency needs to be attenuated.

2. Game Mode Recommendations

• 2.1 Based on the player’s game mode, game duration, frequency, etc. Through clustering algorithm, the player groups are classified. Calculate the mode preference of each overall user, or the mode preference of a single user
• 2.2 Analyze the hero, skin or other item preferences of players with different mode preferences, and recommend heroes (skins) that players have not purchased based on similarity, that is, player-specific items for targeted marketing of related activities

3. Attract new customers

Theoretically speaking, for a MMOB like Honor of Kings where everyone competes against each other, on Tencent’s social relationship chain, a QQ tip s out does not need to worry too much about attracting new users. But when the new user growth curve does not look good, there is still something that can be done.

Early Game

• 3.1 Observe the user retention rate , churn rate, return rate, etc. for different channels , regions, genders, etc.
• 3.2 For user groups with high retention rates of new users, analyze their strongly correlated characteristics: income, game time, frequency, channel, region, gender, etc. Use the decision tree algorithm to analyze and find out the probability of users with certain behavioral characteristics becoming high-retention users in certain channels and regions.

Game maturity

• 3.3 The mainstream target users have been saturated. Analyze the basic attribute characteristics of mainstream target users. Through decision tree algorithms, etc., we can analyze what basic characteristics the mainstream target users meet and then attract them. Or acquire new users through known user attribute characteristics from channels where users of competing products or similar products gather