Ways to analyze a 1M rows dataset of search queries

[briacg]

Hi,

I have this large dataset, about 1 million search queries with visits, bounce rate and a few other metrics. I'm trying to explore this data to find keyword "buckets" (such as include product name, location name, transactional objective, informational, etc.), as well as explore the density of certain keywords (keywords as in instances of a single word amongst all queries)

My idea was to use Excel and a macro to split all queries in separate words (also clearing punctuation and uppercase/lowercase), then storing this word in a new worksheet, adding to another column the visit counts from the row where the word was extracted (as to give a sense of weight). Before adding the word to the new worksheet, the script will look if the word already existed, if so it would just add the current value of visits to the existing visit counts etc.

In the end it will create sort of a "dictionary" of all the keywords in all search queries ranked by weight (= visits from search query including this keyword)

This would help me get started I believe, because I can't segment and analyze 1M raw search queries...

My issue is: this VBA has been running on my (fast) PC for the last 24hr and it doesn't seem to get to an end. Obviously excel+VBA is not the best way to do text mining and manipulation in such a large dataset (although it's just a 30mb file)

What would you do if you had this dataset and would like to mine the text/semantic as I am doing? Any idea of tools? process?

I'm considering dumping this data into a MySQL db and doing the processing through PHP (the only backend language I'm versed in), and getting the "summified" data stored into another table, which I'll then be able to export to a Excel for analysis. But I'm afraid that I'll be facing memory limit issues and such...

In the meantime, I'm definitely interested into knowing what you guys would do if you had this data and wanted to simply start exploring its constituencies

Thanks!

[john4math]

Yeah, Access can process any number of rows.  It's Microsoft's database program.  You can upload data, and then create queries.  They have a design view where you can construct queries in a WYSIWYG fashion, or if you want, you can write your own SQL.

[briacg]

Thanks a lot John!

I'm going to try this out tonight!

So, I assume, Access won't have the same processing limitations with 1 million rows, will it?

Once I'll be done with the "discovery phase" I'm going through with this keyword list, I'll definitely use Advanced filters (in Excel) as you recommend to understand keyword groups in details

[john4math]

I had a similar problem going through my search query reports.  If you're already familiar with VB you could do this with a Microsoft Access database rather than setting up a MySQL one w/PHP.  I've been working on creating an Access database that I can import my data into, and have it spit out all sorts of useful info (for example negative keywords and placements), but it's only in its early stages right now.

If you just want to see it for a few terms and don't mind doing it one at a time, in the past I've filtered data like this in Excel without VB using advanced filters.  I found that using advanced filters rather than VB sped up the process quite a bit; I'd imagine because it's an innate Excel function. Using 4 filters you can match whole words in the queries.  For example, to find queries with "blah", you'd set a filter for "blah", "* blah", "blah " and " blah *".  Then you can use the Subtotal command to do calculations over the visible rows and calculate the data.

More about advanced filters: http://office.microsoft.com/en-us/excel-help/filter-by-using-advanced-criteria-HP005200178.aspx