Web corpus construction involves a significant number of design decisions and turning points in data processing. Depending of the purpose of data collection, it may also require a substantial filtering and quality assessment. While some large-scale algorithms can be expected to smooth out irregularities, uses requiring a low margin of error and close reading approaches (such as the search for examples in lexicographic research) imply constant refinements and improvements with respect to the building and processing of the dataset.

Interest

Because of the vastly increasing variety of corpora, text types and use cases, it becomes more and more difficult to assess the adequacy and quality of certain web data for given research objectives. A central operation in corpus construction consists in retaining the desired content while discarding the rest, a task which has many names referring to peculiar subtasks or to the whole: web scraping, boilerplate removal or boilerplate detection, web page template detection, web page cleaning, or web content extraction – for a recent overview see Lejeune & Zhu (2018).

Recently, approaches using the CommonCrawl have flourished, as they allow for faster download and processing by skipping (or more precisely outsourcing) the crawling phase. While I think that finding one’s …

Description

Metadata extraction

Diverse content extraction and scraping techniques are routinely used on web document collections by companies and research institutions alike. Being able to better qualify the contents allows for insights based on metadata (e.g. content type, authors or categories), better bandwidth control (e.g. by knowing when webpages have been updated), or optimization of indexing (e.g. language-based heuristics, LRU cache, etc.).

In short, metadata extraction is useful for different kinds of purposes ranging from knowledge extraction and business intelligence to classification and refined visualizations. It is often necessary to fully parse the document or apply robust scraping patterns, there are for example webpages for which neither the URL nor the server response provide a reliable way to date the document, that is find when it was written.

Context

I regularly work on improving the extraction methods for the web collections at my home institutions. They are unique as they combine both the quantity resulting from broad web crawling and the quality obtained by carefully extracting text and metadata as well as rejecting documents that do not match certain criteria. In that sense, I already published work on methods to derive metadata from web documents in order …

I am currently working on a project for which I need to identify WordPress blogs as fast as possible, given a list of URLs. I decided to write a review on this topic since I found relevant but sparse hints on how to do it.

First of all, let’s say that guessing if a website uses WordPress by analysing HTML code is straightforward if nothing was been done to hide it, which is almost always the case. As WordPress is one of the most popular content management systems, downloading every page and performing a check afterward is an option that should not be too costly if the amount of web pages to analyze is small. However, downloading even a reasonable number of web pages may take a lot of time, that is why other techniques have to be found to address this issue.

The way I chose to do it is twofold, the first filter is URL-based whereas the final selection uses HTTP HEAD requests.

URL Filter

There are webmasters who create a subfolder named “wordpress” which can be seen clearly in the URL, providing a kind of K.O. victory. If the URLs points to a non-text …

The Microblog Explorer project is about gathering URLs from social networks (FriendFeed, identi.ca, and Reddit) to use them as web crawling seeds. At least by the last two of them a crawl appears to be manageable in terms of both API accessibility and corpus size, which is not the case concerning Twitter for example.

Hypotheses:

1. These platforms account for a relative diversity of user profiles.
2. Documents that are most likely to be important are being shared.
3. It becomes possible to cover languages which are more rarely seen on the Internet, below the English-speaking spammer’s radar.
4. Microblogging services are a good alternative to overcome the limitations of seed URL collections (as well as the biases implied by search engine optimization techniques and link classification).

Characteristics so far:

• The messages themselves are not being stored (links are filtered on the fly using a series of heuristics).
• The URLs that are obviously pointing to media documents are discarded, as the final purpose is to be able to build a text corpus.
• This approach is ‘static’, as it does not rely on any long poll requests, it actively fetches the required pages.
• Starting from the main public timeline, the scripts aim at …

Let’s pretend you recently installed R (a software to do statistical computing), you have a text collection you would like to analyze or classify and some time to lose. Here are a few quick commands that could get you a little further. I also write this kind of cheat sheet in order to remember a set of useful tricks and packages I recently gathered and from which I thought they could help others too.

Letter frequencies

In this example I will use a series of characteristics (or features) extracted from a text collection, more precisely the frequency of each letter from a to z (all lowercase). By the way, it goes as simple as that using Perl and regular expressions (provided you have a $text variable):

my @letters = ("a" .. "z");
foreach my $letter (@letters) {
    my $letter_count = () = $text =~ /$letter/gi;
    printf "%.3f", (($letter_count/length($text))*100);
}

First tests in R

After having started R (‘R’ command), one usually wants to import data. In this case, my file type is TSV (Tab-Separated Values) and the first row contains only describers (from ‘a’ to ‘z’), which comes at hand later. This is done using the read.table command.

alpha <- read.table("letters_frequency …

In his professorial thesis (or habilitation thesis), which is about to be made public (the defence takes place next week), Ludovic Tanguy explains why and on what conditions data visualization could help linguists. In a previous post, I showed a few examples of visualization applied to the field of readability assessment. Tanguy’s questioning is more general, it has to do with what is to include in the disciplinary field of linguistics.

He gives a few reasons to use the methods from the emerging field of visual analytics and mentions some of its upholders (like Daniel Keim or Jean-Daniel Fekete). But he also states that they are not well adapted to the prevailing models of scientific evaluation.

Why use visual analytics in linguistics ?

His main point is the (fast) growing size and complexity of linguistic data. Visualization comes at hand when selecting, listing or counting phenomena does not prove useful anymore. There is evidence from the field of cognitive psychology that an approach based on form recognition may lead to an interpretation. Briefly, new needs come forth when calculations come short.

Tanguy gives to main examples of cases where it is obvious : firstly the analysis of networks, which can be …

It is not only a matter of scale : the perspective one chooses is crucial when it comes to visualize how difficult a text is. Two main options can be taken into consideration:

• An overview in form of a summary which enables to compare a series of phenomena for the whole text.
• A visualization which takes the course of the text into account, as well as the possible evolution of parameters.

I already dealt with the first type of visualization on this blog when I evoked Amazon’s text stats. To sum up, their simplicity is also their main problem, they are easy to read and provide users with a first glimpse of a book, but the kind of information they deliver is not always reliable.

Sooner or later, one has to deal with multidimensional representations as the number of monitored phenomena keeps increasing. That is where a real reflexion on finding a visualization that is faithful and clear at the same time. I would like to introduce two examples of recent research that I find to be relevant to this issue.

An approach inspired by computer science

The first one is taken from an article by Oelke et al. (2010 …

I recently read that Microsoft was planning to analyze several social networks in order to know more about users, so that the search engine could deliver more appropriate results. See this article on geekwire.com : Microsoft idea: Analyze social networks posts to deduce mood, interests, education.

Among the variables that are considered, the ‘sophistication and education level’ of the posts is mentionned. This is highly interesting, because it assumes a double readability assessment, on the reader’s side and on the side of the search engine. More precisely, this could refer to a classification task.

Here is an extract of a patent describing how this is supposed to work.

[0117] In addition to skewing the search results to the user’s inferred interests, the user-following engine 112 may further tailor the search results to a user’s comprehension level. For example, an intelligent processing module 156 may be directed to discerning the sophistication and education level of the posts of a user 102. Based on that inference, the customization engine may vary the sophistication level of the customized search result 510. The user-following engine 112 is able to make determinations about comprehension level several ways, including from a user’s …

I recently updated the blogroll section and I also would like to share a few links:

• a huge links selection on pattern matching: Pattern Matching Pointers
• the comp.text Frequently Asked Questions (Usenet archive)
• links on annotation (page in French, links in English) collected by Karën Fort (Paris 13)

As I will be teaching LaTeX soon the LaTeX links section of the blog has expanded.

Last but not least, here is an E-Book, Mining of Massive Datasets, by A. Rajaraman and J. D. Ullmann. It was made of classes taught at Stanford and is now free to use (available chapter by chapter or as a whole), very up-to-date and informative on this hot topic. It seems to be a good introduction as well. That said I cannot really review it since I am not an expert of this research field.

Here is the reference:

• A. Rajaraman and J. D. Ullmann, Mining of Massive Datasets, Stanford, Palo Alto, CA: e-book, 2010.

Here is my selection of introductory courses given by well-known specialists in Computer Science or Natural Language Processing and recorded so that they can be followed at home.

1. Artificial Intelligence | Natural Language Processing, Christopher D. Manning, Stanford University.
More than 20 hours, 18 lectures.
Introduction to the key topics of NLP, summary of existing models.
Lecture 12 : Dan Jurafsky as a guest lecturer.
Requires the Silverlight plugin (no comment). Transcripts available.

2. Bits, Harry R. Lewis, Harvard University.
A general overview of information as quantity and quantitative methods.
Very comprehensive lecture (data theories, internet protocols, encryption, copyright issues, laws…), cut in small pieces for you to pick a focused topic.
Several formats available, links to blog posts.

3. Search Engines: Technology, Society, and Business, various lecturers, UC Berkeley.
Fall 2007, 13 lectures.
Overview of the topic.
Requires iTunes (no comment).