Data from users also talk about data analysis

Intermediary transaction http://www.aliyun.com/zixun/aggregation/6858.html ">seo diagnose Taobao guest cloud host technology Hall

Yesterday saw Caoz wrote "Data analysis this matter", very worthy of depth, after reading very touched, also here casually write about the data analysis of personal views.

First of all, in the data analysis I also dare not take the master, not a lot of analysis algorithm, do not use any statistical tools, will be silly to stare at. But I like to look at all kinds of data, university day to look at a variety of hardware evaluation, the graduate stage saw countless cameras, lens evaluation, and then the world every week to think about a variety of game consoles, games sales. Work also particularly like the establishment of a variety of statistical systems, look at all kinds of data, now all the company's statistics are written by myself, the general work every day will spend nearly 30% of the time to study the data, at least can be regarded as a full of data analysis enthusiasts.

About the data analysis, Caoz has said very good, I can only add to my experience.

1, whether to do statistics or look at other people's data, the first step is always the reliability of data acquisition. If you are sampling data, be sure to look at the sampling method to see what kind of error may exist. If it is their own data, but also to see whether the data acquisition itself is scientific, such as statistical user behavior generally use JS callback, if you also use the Apache log to do statistics, the results will not be reliable.

2, get the data, it is necessary to establish statistics, at this time, need to think about, the establishment of what kind of statistical information to better analyze products and user characteristics. Many times, a single feature is often difficult to describe and needs to be integrated in many places to see. For example, Web search, often to see the first Ctr, the first three Ctr, the last click and many other factors, and through a number of different factors combined to make analysis and judgment.

3. Be skeptical about the data, especially if there is a causal relationship between the data itself and the conclusion you are trying to reach. For example, Web search results if Ctr High must be experience, okay? Is it ideal to search for an advertised RPM?

4, the generation of the same data, often can have different statistical methods, if the choice of errors, the conclusions will often be widely divergent. For example, to analyze the site's dependence on the search engine, should be using PV, with the session, or with UV statistics it? If a user visits many times a day, some from the search engine, some are active access, how to calculate it? There is a deep knowledge in it.

5, the data often have a lot of noise, how to filter these noises is also very important. Just as voting has voting machines, some spider will execute your statistics JS, some users will be late, if not very good filtering and processing, will make the reliability of the data greatly compromised.

6, understand the various reasons that may cause fluctuations in the data, and through continuous analysis, validation and elimination to find the real reason. For example, when the search traffic drops, there may be a number of reasons, such as the computer room network failure, competitors with some products disruptive, the line of the code there are major destabilizing factors, operators out of trouble or power rationing and so on, each has a different way of verification, the need from the server log, keynote data, the subregion, A number of dimensions such as user behavior are tracked and tested to find the real possible core cause.

Estimating and judging data requires a feeling that is not innate and requires constant exercise and training. This process may be very long, in general, you need to read a lot of data, to develop their basic knowledge of the data, but also to analyze some events (such as weekends, holidays, or failures, etc.) data changes. And before the product on the line, first exercise the estimate, and then through the actual value of their own prediction to verify and evaluate. Through this continuous study and analysis, and gradually develop their own understanding of the data.

Data comes from users, and this is a lot of time to study and analyze human nature. Just like the ads in different positions on the page, how much can CTR generally achieve? The same position, put the advertisement good or the user product good? How much can CTR to make a new product? Do the internet is mostly high-end users, a lot of things themselves will not use will not point, but it is so, the need for users have a very strong sense of generation, To change your thinking and analyze human nature, you can avoid a lot of overly optimistic predictions and unnecessary trial and error beforehand.

Above, is a little experience of their own.

Welcome to the micro-letter public number "search Engine quest", search micro-letter public number Guoang_search