Four Misunderstandings in Game Analysis in the big data cloud Era

In recent years, with the popularity of big data and the rapid development of big data technology, data analysis and data mining have received increasing attention from enterprises, especially in the gaming industry, more and more data analysis and views have been put forward in a spurt. "data-driven refined operation" and "quantitative research on gamer groups ", the introduction of the "mobile game data analysis system" and other concepts shows that in the game R & D and operation processes, the demand for data analysis is endless and relevant, the discussion of data analysis is also different. The loss caused by blind and excessive dependence on data or subjective interpretation of data cannot be estimated. In the following sections, I will describe my work experience to illustrate several mistakes in data indicator fraud and game data analysis.

1. blindly compare data without understanding data definitions

Colleagues are happy to pay attention to benchmark about various types of games in the industry, such as the day-to-day retention of S-level games, ARPU of S-level games, ACU/PCU of S-level games, and other common operational indicators, in my opinion,Comparison is one of the values of data. It is the most direct method used to measure product differences. data comparison is based on the same data collection method and data indicator calculation method. Therefore, before comparing data, you must first understand benchmark's computing standards and data collection methods to demonstrate the significance of data comparison.

2. Over-reliance on analysis methods, indulge in the Data Modeling Process

During his college career, the author majored in Statistics. He once participated in the mathematical modeling competition and obtained a good ranking. He also performed a series of projects, such as BP neural network, Bayesian decision tree, and cluster analysis, I was very excited when I first came into contact with game data analysis. I used various methods to analyze the data. Gradually, I found that in actual work, data analysis is not as rigorous as academic research, and it is necessary to make a quick judgment on data performance, you do not need to verify whether the sample population meets certain statistical distribution before each analysis, or use "Artificial Neural Networks" or other "high-tech means" to predict the number of future users of the product, even when the conclusion "A> B" is given, there is no need to make a "significance test". The test is more about understanding the business. Therefore, in the process of data analysis, do not rely too much on the analysis method, but pay attention to the grasp of the game business.

3. data exists objectively. Do not misunderstand the data.

For colleagues who have been working in the first line for a while,Data analysis is often carried out in such a strange circle. When we extract data, we can see that some data tables exist, in addition, I have come to some conclusions on my own understanding of various phenomena. Under the guidance of such ideas, there is always a way to use data to verify my conclusions. In my opinion, data exists objectively. to interpret data, we also need to maintain an objective and neutral attitude. We need to avoid interpreting a piece of data for our own points of view.

4. If the purpose of data analysis is unclear, the requirements for fuzzy analysis are vague, and the analysis is incomplete, a 300% analysis report should be prepared.

Clarify the purpose and needs of the analysis. For example, Do not mistaken core user research for active user analysis. Manager Liu of wanglong once shared this case with me. The product manager submitted a CoC activity data analysis report to you to measure the activity effect. Generally, you will take out the game macro data in the early, middle, and later stages of the activity, draw a picture of the performance of each stage, and then make a judgment. Then I was excited and handed the Report to the product manager. From the perspective of a data analyst, such a report is cheap. When someone asks for analysis, he may have 10 questions, but only described three questions for us. We cannot simply solve these three problems, we should be more neutral and objective in thinking about such a problem from multiple perspectives, and then from the product itself, product players, product operations, and so on, comprehensively measure such a problem, discover potential opportunities, and then make a 300% analysis report instead of 100%.