Techniques for Sampling Online Text-Based Data Sets

Introduction

Studying social media often involves downloading publically-available textual data. Based on studies of email messages, Facebook, blogs, gaming websites, and Twitter, this essay describes sampling techniques for selecting online data for specific research projects. As previously noted (Webb & Wang, 2013; Wiles, Crow, & Pain, 2011), research methodologies for studying online text tend to follow or adapt existing research methodologies, including sampling techniques. The sampling techniques discussed in this chapter follow well-established sampling practices, resulting in representative and/or purposeful samples; however, the established techniques have been modified to apply to sampling online text—where unusually large populations of messages are available for sampling and the population of messages is a state of constant growth. The sampling techniques discussed in this chapter can be used for both qualitative and quantitative research.

Rapidly advancing internet technologies have altered daily life as well as the academic landscape. Researchers across disciplines are interested in examining the large volumes of data generated on internet platforms, such as social networking sites and mobile devices. Compared to data collected and analyzed through traditional means, big data generated around-the-clock on the internet can help researchers identify latent patterns of human behavior and perceptions that were previously unknown. The richness of the data brings economic benefits to diverse data-intensive industries such as marketing, insurance, and healthcare. Repeated observations of internet data across time amplify the size of already large data sets; data-gathered across time have long interested academics. Vast-sized data sets, typically called “big data,” share at least four shared traits: The data are unstructured, growing at an exponential rate, transformational, and highly complicated.

As more big data sets become available to the researchers through the convenience of internet technologies, ability to analyze the big data sets can weaken. Many factors can contribute to a deficiency in analysis. One major obstacle can be the capability of the analytical systems. Although software developers have introduced multiple analytical tools for scholars to employ with big data (e.g., Hadoop, Storm), the transformational nature of big data requires frequent software updates as well as increases in relevant knowledge. In other words, analyzing big data requires specialized knowledge. Another challenge is selecting an appropriate data-mining process. As Badke (2012, p.47) argued, seeking “specific results for specific queries” without employing the proper mining process can further complicate the project instead of helping manage it. Additionally, data of multi-petabyte which include millions of files from heterogeneous operating systems might be too large to back up through conventional computing methods. In such a case, the choice of the data mining tool becomes critical in determining the feasibility, efficiency, and accuracy of the research project.

Many concerns raised regarding big data collection and analysis duplicate concerns surrounding conventional online data collection:

• •

  Credibility of Online Resources: Authors of the online text often post anonymously. Their responses, comments, or articles are susceptible to credibility critiques;

• •

  Privacy Issues: Internet researchers do not necessarily have permission of the users who originally generated the text. Users are particularly uncomfortable when data generated from personal information, such as Facebook posts or text messages on mobile devices, are examined without their explicit permission. No comprehensive legal system currently exists that draws a clear distinction between publically available data and personal domains;

• •

  Security Issues: When successful online posters, such as bloggers, enjoy the free publicity of the internet, they also can be victimized by co-option of their original work and thus violation of their intellectual property rights. It is difficult for researchers to identify the source of a popular Twitter post that is re-tweeted thousands of times, often without acknowledging the original author. Therefore, data collected from open-access online sources might infringe authors’ copyrights.