A suite of tools to scrape and parse search engine results
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
My posts are usually R only. But in this post, I want to talk about a suite of tools developed by my colleagues and me that goes beyond R only. This suite of tools helps to gather results from different search engines and includes a browser extension to scrape the results, and a Python library and an R package to parse the results.
The browser extension
The core tool is a browser extension for Mozilla and Chrome that simulates a user searching (at least) 50-top main, news, images and videos search results of up to 8 different search engines.
The repository is well documented and walks you through the steps of setting up the extension to scrape data. Always make sure to deactivate the extension once you are done since it interferes with your normal browsing.
The extension can store local snapshots of the html files of the search results for later analysis. This is done via the web extension SingleFile. To parse these html files efficiently for important information, we a Python library as well as an R package.
The Python library
To use the Python library, simply clone the
repository and either add webbotparser/webbotparser.py to your working directory or navigate to the folder, and run
pip install -e .
The webbotparser is then available in your Python installation.
(The following is an excerpt of the repository README)
Usage
For the search engines and result types supported out of the box, simply run
from webbotparser import WebBotParser
and initialize the WebBotParser for the search engine and result type your are investigating, for example
parser = WebBotParser(engine = 'DuckDuckGo News')
Then, you can obtain the search results as a pandas DataFrame and metadata as a Python dictionary with
metadata, results = parser.get_results(file='path/to/the/result_page.html')
Furthermore, parser.get_metadata(file) can be used to only extract the metadata. parser.get_results_from_dir(dir) allows to directly extract search results spread over multiple pages, as Google text result are provided for instance. For examples also see example.ipynb.
Extracting images
WebBot archives images inline in the html file of the search results, i.e., they are neither external files on your drive nor fetched from the original source on viewing the downloaded search results page. This allows us to extract the images directly from the html file for further analysis. The engines and result types supported out of the box with WebBotParser allow for extracting images as well. Simply initialize WebBotParser as follows:
parser = WebBotParser(engine = 'Google Video', extract_images=True)
You can optionally specify extract_images_prefix, extract_images_format, and extract_images_to_dir. See example.ipynb for more details, including preview in Jupyter Notebooks.
Custom result types
WebBotParser out of the box only provides support for some search engines and result types. Even these parsers might stop working if the search engine providers decide to change their layout. However, WebBotParser can still be used in these cases by defining a custom result_selector, queries, and optionally a metadata_extractor function. In this case, a WebBotParser is initiated with these instead of with the engine attribute
parser = WebBotParser(queries, result_selector, metadata_extractor)
Under the hood, WebBotParser uses BeautifulSoup to
- Parse the search result page’s HTML via LXML
- Disciminate the individual results on each page using a CSS selector called
result_selectorthat matches a list of search results - For each of those results, extract available information through a list of queries
See the below example for available types of queries and their usage
queries = [
# extract the text from inside a matched element, getting all the text over all its children
{'name': 'abc', 'type': 'text', 'selector': 'h3'},
# extract the value of an attribute of a matched element
{'name': 'def', 'type': 'attribute', 'selector': 'a', 'attribute': 'href'},
# whether or not a CSS selector matches, returns a Boolean
{'name': 'ghi', 'type': 'exists', 'selector': 'ul'},
# extract inline images and name them by a title
{'name': 'jkl', 'type': 'image', 'selector': 'g-img > img', 'title_selector': 'h3'}
# pass a custom query function
{'name': 'mno', 'type': 'custom', 'function': my_function},
]
You can optionally provide a metadata_extractor(soup, file) function to extract metadata alongside the search results, or import one of the existing extractors, e.g. with
from webbotparser import GoogleParser metadata_extractor = GoogleParser.google_metadata
The R package
The R package can be installed from GitHub
remotes::install_github("schochastics/webbotparseR")
library(webbotparseR)
The package contains an example html from a google search on climate change.
ex_file <- system.file("www.google.com_climatechange_text_2023-03-16_08_16_11.html", package = "webbotparseR")
Such search results can be parsed via the function parse_search_results(). The parameter engine is used to specify the
search engine and the search type.
output <- parse_search_results(path = ex_file,engine = "google text") output ## # A tibble: 10 × 10 ## title link text image page posit…¹ searc…² type query date ## <chr> <chr> <chr> <chr> <chr> <int> <chr> <chr> <chr> <dttm> ## 1 What… http… Clim… data… 1 1 www.go… text clim… 2023-03-16 08:16:11 ## 2 Home… http… Vita… data… 1 2 www.go… text clim… 2023-03-16 08:16:11 ## 3 Vita… http… “Cli… data… 1 3 www.go… text clim… 2023-03-16 08:16:11 ## 4 Clim… http… In c… data… 1 4 www.go… text clim… 2023-03-16 08:16:11 ## 5 IPCC… http… The … data… 1 5 www.go… text clim… 2023-03-16 08:16:11 ## 6 Clim… http… Comp… data… 1 6 www.go… text clim… 2023-03-16 08:16:11 ## 7 Clim… http… Clim… <NA> 1 7 www.go… text clim… 2023-03-16 08:16:11 ## 8 UNFC… http… What… data… 1 8 www.go… text clim… 2023-03-16 08:16:11 ## 9 Clim… http… Clim… data… 1 9 www.go… text clim… 2023-03-16 08:16:11 ## 10 Caus… http… This… data… 1 10 www.go… text clim… 2023-03-16 08:16:11 ## # … with abbreviated variable names ¹position, ²search_engine
Note that images are always returned base64 encoded.
output$image[1] ## [1] "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAIAAACQkWg2AAAABnRSTlMAAAAAAABupgeRAAAAMklEQVR4AWMAgYYG4hEdNJAHGoCIABvBJayhgcYaIAwaakCwydUA52MKYeeSCgZh4gMAXrJ9ASggqqAAAAAASUVORK5CYII="
The function base64_to_img() can be used to decode the image and save it in an appropriate format.
Caveats
Given that search engines change their frontpage from time to time, the extension can break and needs to be adjusted. As of writing (22/03/2022), the search engine Bing is not supported due to some changes on their frontpage.
Both the Python and R library rely on css selectors to extract the relevant information from the html files. This is even more fragile than what the browser extension does and may require more frequent updates. Both libraries though offer the possibility to use custom selectors. This can be useful in cases where search engines updated their css classes and those changes have not yet been incorporated into the libraries.
R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.