1) Big data is not a quantitative pile, but has a strong correlation structure
For example, if there is a kind of data that records how much every big tree in the world grows taller every year, this kind of data is not valuable because it is just a simple pile.
If the data becomes, every big tree records its, location, climatic conditions, tree species, tree age, the surrounding flora and fauna ecology, and the height of the growth each year, then this data has a structure. Structured data has a strong research value first, and a strong commercial value second.
In the case of Taobao data, for example, if you only record information about buyers, sellers, items sold, and prices of a transaction, then the commercial value is very limited. If Taobao includes the social relationships between buyers and other behaviors before and after a purchase, then this data will be very valuable.
So only three-dimensional, highly structured data can be called big data, only valuable, otherwise it can only be called large-scale data.
2) The scale of big data must be large, and larger than the scale of large-scale data
To do some predictive modeling requires a lot of data, training corpus, if the data is not large enough, a lot of mining is difficult to do, such as click-through rate prediction. The most straightforward example, if you can know a user's long-term whereabouts data, the behavior of surfing the Internet, read operations and write operations. Then you can almost make very accurate predictions about this person, and all kinds of recommended work can be done very accurately.