Game Cloud of the mass game log analysis scheme

We will combine case studies to see how to use data analysis to actually improve the game.

This time we have an example of a mobile end of the card games, simulation board game 21 points. The popular version of the game is free of charge, and users can obtain a version without advertising and additional features after paying a certain fee. The problem is that this 21-point game does not generate the expected revenue. The expectation is to solve this problem, increase user participation and consumption.

If the designer of the game does not have the concept of data analysis, and does not collect enough data for analysis, it can only pat the head to make policy changes. And if the idea of data analysis is deployed at the beginning of design, then every step of the analysis will be substantiated.

We assume that the game design has been considered at the beginning of the data analysis, on the Ali cloud in accordance with the following figure to deploy:

ECS as game server;

RDS, as a business database, maintains transaction-related data and synchronizes the data needed for analysis to ODPs through the Aliyun mining cloud;

SLS, as a log service, periodically imports log data into ODPs

ODPs data analysis as an Open data processing service.

In order to find out why the game did not bring the expected revenue, we take steps to troubleshoot and make the appropriate action:

1. Is there enough people to download the app?

View the total number of downloads and the first installation of the game. Downloads and installation can be maintained in RDS.

If you don't have enough downloads, you need PR and marketing to work around the visibility of the product. For example, add creative elements to the game, collaborate with the operators to promote them, adopt new skin packs or attractive titles. Increase the advertising effort in mobile browsers or search clients.

If the data shows that there is enough user downloads and first-time installation, it does not mean that the problem is in terms of popularity, and that there is no need to add additional PR input, but continue to view step 2.

2. What is the first thing they do in the game for users who install games? What percentage of all users who play too many rounds until they completely end the game?

For the first question, we look at the logged in database log activity (Entry event distribute,eed), which can also be maintained in RDS. For the second question, we automatically send a message to the server to collect the data every time the user finishes the game.

If a lot of users enter the game, but most people do not choose to upgrade to the paid version, it may be interesting to attract people's attention, but not very good monetization. If this is the case, consider the value of the upgrade and find out what it can do to bring more value to the user and let the user know.

If only a few users have completed the game, or if most people quit at the beginning of the game, proceed to step 3.

3. At what step did the user exit?

To get this data, you need to log each step as the user performs the action. Because the exit event can only be determined based on the actions performed prior to the event, you need to add the data points when the user performs each step, and the finer the granularity the better (the volume of data is distribution,exd). Obviously such data is not suitable for the business database, that is, the RDS in the above, and the use of Aliyun Log service SLS, can automatically read the resulting log files, and automatically imported into the ODPs for subsequent analysis. It will not occupy the real-time business resources, but also facilitate the analysis of large amounts of data processing.

Then, because you do not know whether the user will come back to continue the game, you need to filter out those who are not in a certain period of time to log back to the user. The method is to estimate the average logon interval for active users and then multiply the 10来 to determine whether the user is permanently lost. For example, if the average logon interval for most users is 1 days, it can be judged as a loss for users who have not logged in for 10 days.

Here exd can have many branches, for example, how many users to complete the game guide? If most users exit after completing the guide, the guide itself may have some complexity and requires further design of the game guide. If most users do not complete the guide, you need to consider setting the guide to optional on the main page of the game so that users can skip the game guide and enter the game directly.

If most users have completed the game guide, proceed to step 4.

4. When users are ready to start the game, can they successfully match the same objects?

Which game object does the user choose first? Versus another player or a random choice? Playing alone or with a machine? Or with the people around the use of "Hot Seat Mode"? If most users use the game button first to play with friends, when they do, what options do they use to find a partner? Now a lot of people will be in the "micro-letter" to look for. When you select the micro-mail import, a Allow page is popped up asking if you want to allow access to the application. How many users choose to allow access to the application?

This data can be recorded in RDS or in the form of a log by SLS to import ODPs. By setting the map and reduce rules in ODPs, you get the analysis data you need.

If the "not allowed" ratio is high, it may be that the content of the page is allowed to scare the user. In this case, you can remove some common statements such as "Access data at any time", "Disturb my friends", so that the page does not look so disturbing users. In addition, how many users are using this process to find friends and then play games together? You can compare the number of people entering the "micro-mail" option by comparing the numbers that start the game at this point and enter the game immediately. If this ratio is low, it means you need to improve at the game pairing. Of course, there is a need for more data analysis rules.

5. When users enter the game, do they finish the game? How many rounds did they play?

If you quit after one or two rounds of play, if the game is a person and the turn between them is too long, it may be because the length of the turn is prolonged and can cause boredom. This can be queried by the average time elapsed between rounds.

If the round length looks reasonable, but the player still exits after one or two innings, it needs to check the game's gameplay.

6. For those who have played a few games but still lost, consider the following questions:

Do these players always win or always lose in their own game rounds? If they always win or lose, they may find the game too simple or too complex. If the winning and losing results are more random, it requires more precise games to be inspired by playing with the data. If the game is their friend, it may not inspire much. If the game is a machine, more data collection is needed for further research.

These steps are just a simplified example of the data analysis for the game improvement, as shown in the figure, the game process data from RDS or SLS import ODPs, and then from the ODPs to analyze a large number of data to help specify the follow-up decision. It also can be seen that the data analysis in the whole process of the analysis of the silent support, it is "I would like to, shifting quicksand lying lake Embankment." Only accompany you, waiting for the rotation of spring and summer. ”

Learn more about the game architecture >>