Scraping R-bloggers with Python – Part 2

In my previous post I showed how to write a small simple python script to download the pages of R-bloggers.com. If you followed that post and ran the script, you should have a folder on your hard drive with 2409 .html files labeled post1.html , post2.html and so forth. The next step is to write a small script that extract the information we want from each page, and store that information in a .csv file that is easily read by R. In this post I will show how to extract the post title, author name and date of a given post and store it in a .csv file with a unique id.

To do this open a document in your favorite python editor (I like to use aquamacs) and name it: extraction.py. As in the previous post we start by importing the modules that we will use for the extraction:

from BeautifulSoup import BeautifulSoup

import os
import re

As in the previous post we will be using the BeautifulSoup module to extract the relevant information from the pages. The os module is used to get a list of file from the directory where we have saved the .html files, and finally the re module allows us to use regular expressions to format the titles that include a comma value or a newline value (\n). We need to remove these as they would mess up the formatting of the .csv file.

After having read in the modules, we need to get a list of files that we can iterate over. First we need to specify the path were the files are saved, and then we use the os module to get all the filenames in the specified directory:

path = "/Users/thomasjensen/Documents/RBloggersScrape/download"

listing = os.listdir(path)

It might be that there are other files in the given directory, hence we apply a filter, in shape of a list comprehension, to weed out any file names that do not match our naming scheme:

listing = [name for name in listing if re.search(r"post\d+\.html",name) != None]

Notice that a regular expression was used to determine whether a given name in the list matched our naming scheme. For more on regular expressions have a look at this site.

The final steps in preparing our extraction is to change the working directory to where we have our .html files, and create an empty dictionary:

os.chdir(path)
data = {}

Dictionaries are one of the great features of Python. Essentially a dictionary is a mapping of a key to a specific value, however the fact that dictionaries can be nested within each other, allows us to create data structures similar to R’s data frames.

Now we are ready to begin extracting information from our downloaded pages. Much as in the previous post, we will loop over all the file names, read each file into Python and create a BeautifulSoup object from the file:

for page in listing:
    site = open(page,"rb")
    soup = BeautifulSoup(site)

In order to store the values we extract from a given page, we update the dictionary with a unique key for the page. Since our naming scheme made sure that each file had a unique name, we simply remove the .html part from the page name, and use that as our key:

key = re.sub(".html","",page)

data.update({key:{}})

This will create a mapping between our key and an empty dictionary, nested within the data dictionary. Once this is done we can start extract information and store it in our newly created nested dictionary. The content we want is located in the main column, which has the id tag “leftcontent” in the html code. To get at this we use find() function on soup object created above:

content = soup.find("div", id = "leftcontent")

The first “h1” tag in our content object contains the title, so again we will use the find() function on the content object, to find the first “h1” tag:

title = content.findNext("h1").text

To get the text within the “h1” tag the .text had been added to our search with in the content object.

To find the author name, we are lucky that there is a class of “div” tags called “meta” which contain a link with the author name in it. To get the author name we simply find the meta div class and search for a link. Then we pull out the text of the link tag:

author = content.find("div",{"class":"meta"}).findNext("a").text

Getting the date is a simple matter as it is nested within div tag with the class “date”:

date = content.find("div",{"class":"date"}).text

Once we have the three variables we put them in dictionaries that are nested within the nested dictionary we created with the key:

data[key]["title"] = title
data[key]["author"] = author
data[key]["date"] = date

Once we have run the loop and gone through all posts, we need to write them in the right format to a .csv file. To begin with we open a .csv file names output:

output = open("/Users/thomasjensen/Documents/RBloggersScrape/output.csv","wb")

then we create a header that contain the variable names and write it to the output.csv file as the first row:

variables = unicode(",".join(["id","date","author","title"]))
header = variables + "\n"
output.write(header.encode("utf8"))

Next we pull out all the unique keys from our dictionary that represent individual posts:

keys = data.keys()

Now it is a simple matter of looping through all the keys, pull out the information associated with each key, and write that information to the output.csv file:

for key in keys:
    print key
    id = key
    date = re.sub(",","",data[key]["date"])
    author = data[key]["author"]
    title = re.sub(",","",data[key]["title"])
    title = re.sub("\\n","",title)
    linelist = [id,date,author,title]
    linestring = unicode(",".join(linelist))
    linestring = linestring + "\n"
    output.write(linestring.encode("utf-8"))

Notice that we first create four variables that contain the id, date, author and title information. With regards to the title we use two regular expressions to remove any commas and “\n” from the title, as these would create new columns or new line breaks in the output.csv file. Finally we put the variables together in a list, and turn the list into a string with the list items separated by a comma. Then a linebreak is added to the end of the string, and the string is written to the output.csv file. As a last step we close the file connection:

output.close()

And that is it. If you followed the steps you should now have a csv file in your directory with 2409 rows, and four variables – ready to be read into R. Stay tuned for the next post which will show how we can use this data to see how R-bloggers has developed since 2005. The full extraction script is shown below:

from BeautifulSoup import BeautifulSoup

import os
import re

 path = "/Users/thomasjensen/Documents/RBloggersScrape/download"
 listing = os.listdir(path)

listing = [name for name in listing if re.search(r"post\d+\.html",name) != None]
 os.chdir(path)
 data = {}
 for page in listing:
site = open(page,"rb")
soup = BeautifulSoup(site)
key = re.sub(".html","",page)
print key
data.update({key:{}})
 content = soup.find("div", id = "leftcontent")
title = content.findNext("h1").text
author = content.find("div",{"class":"meta"}).findNext("a").text
date = content.find("div",{"class":"date"}).text
data[key]["title"] = title
data[key]["author"] = author
data[key]["date"] = date

 output = open("/Users/thomasjensen/Documents/RBloggersScrape/output.csv","wb")

 keys = data.keys()
 variables = unicode(",".join(["id","date","author","title"]))
 header = variables + "\n"
 output.write(header.encode("utf8"))
 for key in keys:
print key
id = key
date = re.sub(",","",data[key]["date"])
author = data[key]["author"]
title = re.sub(",","",data[key]["title"])
title = re.sub("\\n","",title)
linelist = [id,date,author,title]
linestring = unicode(",".join(linelist))
linestring = linestring + "\n"
output.write(linestring.encode("utf-8"))
 output.close()

Source:http://www.r-bloggers.com/scraping-r-bloggers-with-python-part-2/

Data Mining and Frequent Datasets

I've been doing some work for my exams in a few days and I'm going through some past papers but unfortunately there are no corresponding answers. I've answered the question and I was wondering if someone could tell me if I am correct.

My question is

    (c) A transactional dataset, T, is given below:
    t1: Milk, Chicken, Beer
    t2: Chicken, Cheese
    t3: Cheese, Boots
    t4: Cheese, Chicken, Beer,
    t5: Chicken, Beer, Clothes, Cheese, Milk
    t6: Clothes, Beer, Milk
    t7: Beer, Milk, Clothes

    Assume that minimum support is 0.5 (minsup = 0.5).

    (i) Find all frequent itemsets.

Here is how I worked it out:

    Item : Amount
    Milk : 4
    Chicken : 4
    Beer : 5
    Cheese : 4
    Boots : 1
    Clothes : 3

Now because the minsup is 0.5 you eliminate boots and clothes and make a combo of the remaining giving:

    {items} : Amount
    {Milk, Chicken} : 2
    {Milk, Beer} : 4
    {Milk, Cheese} : 1
    {Chicken, Beer} : 3
    {Chicken, Cheese} : 3
    {Beer, Cheese} : 2

Which leaves milk and beer as the only frequent item set then as it is the only one above the minsup?

data mining

Nanor

3 Answers

There are two ways to solve the problem:

    using Apriori algorithm
    Using FP counting

Assuming that you are using Apriori, the answer you got is correct.

The algorithm is simple:

First you count frequent 1-item sets and exclude the item-sets below minimum support.

Then count frequent 2-item sets by combining frequent items from previous iteration and exclude the item-sets below support threshold.

The algorithm can go on until no item-sets are greater than threshold.

In the problem given to you, you only get 1 set of 2 items greater than threshold so you can't move further.

There is a solved example of further steps on Wikipedia here.

You can refer "Data Mining Concepts and Techniques" by Han and Kamber for more examples.

141

There is more than two algorithms to solve this problem. I will just mention a few of them: Apriori, FPGrowth, Eclat, HMine, DCI, Relim, AIM, etc. –  Phil Mar 5 '13 at 7:18

OK to start, you must first understand, data mining (sometimes called data or knowledge discovery) is the process of analyzing data from different perspectives and summarizing it into useful information - information that can be used to increase revenue, cuts costs, or both. Data mining software is one of a number of analytical tools for analyzing data. It allows users to analyze data from many different dimensions or angles, categorize it, and summarize the relationships identified. Technically, data mining is the process of finding correlations or patterns among dozens of fields in large relational databases.

Now, the amount of raw data stored in corporate databases is exploding. From trillions of point-of-sale transactions and credit card purchases to pixel-by-pixel images of galaxies, databases are now measured in gigabytes and terabytes. (One terabyte = one trillion bytes. A terabyte is equivalent to about 2 million books!) For instance, every day, Wal-Mart uploads 20 million point-of-sale transactions to an A&T massively parallel system with 483 processors running a centralized database.

Raw data by itself, however, does not provide much information. In today's fiercely competitive business environment, companies need to rapidly turn these terabytes of raw data into significant insights into their customers and markets to guide their marketing, investment, and management strategies.

Now you must understand that association rule mining is an important model in data mining. Its mining algorithms discover all item associations (or rules) in the data that satisfy the user-specified minimum support (minsup) and minimum confidence (minconf) constraints. Minsup controls the minimum number of data cases that a rule must cover. Minconf controls the predictive strength of the rule.

Since only one minsup is used for the whole database, the model implicitly assumes that all items in the data are of the same nature and/or have similar frequencies in the data. This is, however, seldom the case in real- life applications. In many applications, some items appear very frequently in the data, while others rarely appear. If minsup is set too high, those rules that involve rare items will not be found. To find rules that involve both frequent and rare items, minsup has to be set very low.

This may cause combinatorial explosion because those frequent items will be associated with one another in all possible ways. This dilemma is called the rare item problem. This paper proposes a novel technique to solve this problem. The technique allows the user to specify multiple minimum supports to reflect the natures of the items and their varied frequencies in the database. In rule mining, different rules may need to satisfy different minimum supports depending on what items are in the rules.

Given a set of transactions T (the database), the problem of mining association rules is to discover all association rules that have support and confidence greater than the user-specified minimum support (called minsup) and minimum confidence (called minconf).

I hope that once you understand the very basics of data mining that the answer to this question shall become apparent.

1

The Apriori algorithm is based on the idea that for a pair o items to be frequent, each individual item should also be frequent. If the hamburguer-ketchup pair is frequent, the hamburger itself must also appear frequently in the baskets. The same can be said about the ketchup.

So for the algorithm, it is established a "threshold X" to define what is or it is not frequent. If an item appears more than X times, it is considered frequent.

The first step of the algorithm is to pass for each item in each basket, and calculate their frequency (count how many time it appears). This can be done with a hash of size N, where the position y of the hash, refers to the frequency of Y.

If item y has a frequency greater than X, it is said to be frequent.

In the second step of the algorithm, we iterate through the items again, computing the frequency of pairs in the baskets. The catch is that we compute only for items that are individually frequent. So if item y and item z are frequent on itselves, we then compute the frequency of the pair. This condition greatly reduces the pairs to compute, and the amount of memory taken.

Once this is calculated, the frequencies greater than the threshold are said frequent itemset.

Source: http://stackoverflow.com/questions/14164853/data-mining-and-frequent-datasets?rq=1

Using Kimono Labs to Scrape the Web for Free

Historically, I have written and presented about big data—using data to create insights, and how to automate your data ingestion process by connecting to APIs and leveraging advanced database technologies.

Recently I spoke at SMX West about leveraging the rich data in webmaster tools. After the panel, I was approached by the in-house SEO of a small company, who asked me how he could extract and leverage all the rich data out there without having a development team or large budget. I pointed him to the CSV exports and some of the more hidden tools to extract Google data, such as the GA Query Builder and the YouTube Analytics Query Builder.

However, what do you do if there is no API? What do you do if you want to look at unstructured data, or use a data source that does not provide an export?

For today's analytics pros, the world of scraping—or content extraction (sounds less black hat)—has evolved a lot, and there are lots of great technologies and tools out there to help solve those problems. To do so, many companies have emerged that specialize in programmatic content extraction such as Mozenda, ScraperWiki, ImprtIO, and Outwit, but for today's example I will use Kimono Labs. Kimono is simple and easy to use and offers very competitive pricing (including a very functional free version). I should also note that I have no connection to Kimono; it's simply the tool I used for this example.

Before we get into the actual "scraping" I want to briefly discuss how these tools work.

The purpose of a tool like Kimono is to take unstructured data (not organized or exportable) and convert it into a structured format. The prime example of this is any ranking tool. A ranking tool reads Google's results page, extracts the information and, based on certain rules, it creates a visual view of the data which is your ranking report.

Kimono Labs allows you to extract this data either on demand or as a scheduled job. Once you've extracted the data, it then allows you to either download it via a file or extract it via their own API. This is where Kimono really shines—it basically allows you to take any website or data source and turn it into an API or automated export.

For today's exercise I would like to create two scrapers.

A. A ranking tool that will take Google's results and store them in a data set, just like any other ranking tool. (Disclaimer: this is meant only as an example, as scraping Google's results is against Google's Terms of Service).

B. A ranking tool for Slideshare. We will simulate a Slideshare search and then extract all the results including some additional metrics. Once we have collected this data, we will look at the types of insights you are able to generate.

1. Sign up

Signup is simple; just go to http://www.kimonolabs.com/signup and complete the form. You will then be brought to a welcome page where you will be asked to drag their bookmarklet into your bookmarks bar.

The Kimonify Bookmarklet is the trigger that will start the application.

2. Building a ranking tool

Simply navigate your browser to Google and perform a search; in this example I am going to use the term "scraping." Once the results pages are displayed, press the kimonify button (in some cases you might need to search again). Once you complete your search you should see a screen like the one below:

It is basically the default results page, but on the top you should see the Kimono Tool Bar. Let's have a close look at that:

The bar is broken down into a few actions:

    URL – Is the current URL you are analyzing.

    ITEM NAME – Once you define an item to collect, you should name it.

    ITEM COUNT – This will show you the number of results in your current collection.

    NEW ITEM – Once you have completed the first item, you can click this to start to collect the next set.

    PAGINATION – You use this mode to define the pagination link.

    UNDO – I hope I don't have to explain this ;)

    EXTRACTOR VIEW – The mode you see in the screenshot above.

    MODEL VIEW – Shows you the data model (the items and the type).

    DATA VIEW – Shows you the actual data the current page would collect.

    DONE – Saves your newly created API.

After you press the bookmarklet you need to start tagging the individual elements you want to extract. You can do this simply by clicking on the desired elements on the page (if you hover over it, it changes color for collectable elements).

Kimono will then try to identify similar elements on the page; it will highlight some suggested ones and you can confirm a suggestion via the little checkmark:

A great way to make sure you have the correct elements is by looking at the count. For example, we know that Google shows 10 results per page, therefore we want to see "10" in the item count box, which indicates that we have 10 similar items marked. Now go ahead and name your new item group. Each collection of elements should have a unique name. In this page, it would be "Title".

Now it's time to confirm the data; just click on the little Data icon to see a preview of the actual data this page would collect. In the data view you can switch between different formats (JSON, CSV and RSS). If everything went well, it should look like this:

As you can see, it not only extracted the visual title but also the underlying link. Good job!

To collect some more info, click on the Extractor icon again and pick out the next element.

Now click on the Plus icon and then on the description of the first listing. Since the first listing contains site links, it is not clear to Kimono what the structure is, so we need to help it along and click on the next description as well.

As soon as you do this, Kimono will identify some other descriptions; however, our count only shows 8 instead of the 10 items that are actually on that page. As we scroll down, we see some entries with author markup; Kimono is not sure if they are part of the set, so click the little checkbox to confirm. Your count should jump to 10.

Now that you identified all 10 objects, go ahead and name that group; the process is the same as in the Title example. In order to make our Tool better than others, I would like to add one more set— the author info.

Once again, click the Plus icon to start a new collection and scroll down to click on the author name. Because this is totally unstructured, Google will make a few recommendations; in this case, we are working on the exclusion process, so press the X for everything that's not an author name. Since the word "by" is included, highlight only the name and not "by" to exclude that (keep in mind you can always undo if things get odd).

Once you've highlighted both names, results should look like the one below, with the count in the circle being 2 representing the two authors listed on this page.

Out of interest I did the same for the number of people in their Google+ circles. Once you have done that, click on the Model View button, and you should see all the fields. If you click on the Data View you should see the data set with the authors and circles.

As a final step, let's go back to the Extractor view and define the pagination; just click the Pagination button (it looks like a book) and select the next link. Once you have done that, click Done.

You will be presented with a screen similar to this one:

Here you simply name your API, define how often you want this data to be extracted and how many pages you want to crawl. All of these settings can be changed manually; I would leave it with On demand and 10 pages max to not overuse your credits.

Once you've saved your API, there are a ton of options (too many to review here). Kimono has a great learning section you can check out any time.

To collect the listings requires a quick setup. Click on the pagination tab, turn it on and set your schedule to On demand to pull data when you ask it to. Your screen should look like this:

Now press Crawl and Kimono will start collecting your data. If you see any issues, you can always click on Edit API and go back to the extraction screen.

Once the crawl is completed, go to the Test Endpoint tab to view or download your data (I prefer CSV because you can easily open it in Excel, CSV, Spotfire, etc.) A possible next step here would be doing this for multiple keywords and then analyzing the impact of, say, G+ Authority on rankings. Again, many of you might say that a ranking tool can already do this, and that's true, but I wanted to cover the basics before we dive into the next one.

3. Extracting SlideShare data

With Slideshare's recent growth in popularity it has become a document sharing tool of choice for many marketers. But what's really on Slideshare, who are the influencers, what makes it tick? We can utilize a custom scraper to extract that kind data from Slideshare.

To get started, point your browser to Slideshare and pick a keyword to search for.

For our example I want to look at presentations that talk about PPC in English, sorted by popularity, so the URL would be:

http://www.slideshare.net/search/slideshow?ft=presentations&lang=en&page=1&q=ppc&qf=qf1&sort=views&ud=any

Once you are on that page, pick the Kimonify button as you did earlier and tag the elements. In this case I will tag:

    Title
    Description
    Category
    Author
    Likes
    Slides

Once you have tagged those, go ahead and add the pagination as described above.

That will make a nice rich dataset which should look like this:

Hit Done and you're finished. In order to quickly highlight the benefits of this rich data, I am going to load the data into Spotfire to get some interesting statics (I hope).

4. Insights

Rather than do a step-by-step walktrough of how to build dashboards, which you can find here, I just want to show you some insights you can glean from this data:

    Most Popular Authors by Category. This shows you the top contributors and the categories they are in for PPC (squares sized by Likes)

    Correlations. Is there a correlation between the numbers of slides vs. the number of likes? Why not find out?
    Category with the most PPC content. Discover where your content works best (most likes).

5. Output

One of the great things about Kimono we have not really covered is that it actually converts websites into APIs. That means you build them once, and each time you need the data you can call it up. As an example, if I call up the Slideshare API again tomorrow, the data will be different. So you basically appified Slisdeshare. The interesting part here is the flexibility that Kimono offers. If you go to the How to Use slide, you will see the way Kimono treats the Source URL In this case it looks like this:

The way you can pull data from Kimono aside from the export is their own API; in this case you call the default URL,

http://www.kimonolabs.com/api/YOURPAIID?apikey=YO...

You would get the default data from the original URL; however, as illustrated in the table above, you can dynamically adjust elements of the source URL.

For example, if you append "&q=SEO"

(http://www.kimonolabs.com/api/YOURPAIID?apikey=YOURAPIKEY&q=SEO)

you would get the top slides for SEO instead of PPC. You can change any of the URL options easily.

I know this was a lot of information, but believe me when I tell you, we just scratched the surface. Tools like Kimono offer a variety of advanced functions that really open up the possibilities. Once you start to realize the potential, you will come up with some amazing, innovative ideas. I would love to see some of them here shared in the comments. So get out there and start scraping … and please feel free to tweet at me or reply below with any questions or comments!

Source: http://moz.com/blog/web-scraping-with-kimono-labs