Data mining is not about retrieving the right data. That area is usually referred to as "information retrieval" or "database querying". This is an interesting and important area, but it is not data mining. If you want to understand the relationship between data mining and information retrieval, you can think of data mining as inferring the "right query" based on a set of documents that are marked as either relevant or non-relevant.

Data mining is also not about merging many databases into a single database. That area is called "database integration" or "data fusion". Properly used, database integration can complement data mining, but it is not an essential component.

Finally, the term "data mining" has sometimes been used in a pejorative sense — as a way of referring to the misuse of statistical tools ("That's just data mining"). This practice is also called "data dredging" or "overfitting your model". The problems associated with searching a large space of possible models are well known in the data mining community, and we have many technical methods to deal with those problems (randomization tests, cross-validation, ensemble methods, pruning, and penalized evaluation functions, to name a few). This is an area of my own research, so let me know if you would like more detail. To summarize, however, many modern data mining techniques take adequate steps to address these problems.