Big Data and Data Science: Opportunities and Challenges of iSchools

Davenport and Patil (2012) state that the term “data scientist” was coined in 2008 by D. J. Patil and J. Hammerbacher. Some popular definitions of data science include “the study of the generalizable extraction of knowledge from data” (Dhar, 2013) and “an emerging area of work concerned with the collection, preparation, analysis, visualization, management, and preservation of large collections of information” (Stanton, 2012). Provost and Fawcett (2013) stated that “data science involves principles, processes, and techniques for understanding phenomena via the (automated) analysis of data.” Davenport and Patil (2012) discuss that “data scientists are high-ranking professionals with the training and curiosity to make discoveries in the world of big data; they make discoveries while swimming in data; they communicate what they’ve learned and suggest its implications for new business directions.” Based on these definitions, there are three important dimensions of data scientists. Data scientists should be able to (1) understand roles of big data and big data technologies, (2) work with all the steps of a data science lifecycle—discover problems, solve problems, and communicate solutions, and (3) use a range of tools to solve big data problems. We discuss how iSchools should approach the data science education in the following sections.

Industries are struggling to hire talented data scientists to handle increasingly complex data-processing demands and related challenges. Although many universities have begun to offer data science education programs at various levels (Song & Zhu, 2016), big data technologies are relatively new in traditional disciplines’ curricula. Most data scientists have acquired big data skills outside the university. Due to its broad scope, incorporating big data technologies seamlessly into curriculum is a huge challenge to traditional educational departments. Broadly, we summarize those challenges as follows: (1) how to seamlessly incorporate big data revolution in data science education, (2) how to educate students to create tools/value/information/knowledge from big data by utilizing data science skills, (3) how to expand curriculum to newer topics such as intelligent augmentation and cognitive computing, and (4) how to put these emerging data management topics into curriculum.

According to our survey (Song & Zhu, 2016), iSchools are one of pioneering communities in data science education. For over 100 years, iSchools have been dedicated to advancing the information field, beyond library science, connecting the two ends of data and knowledge. Modernized iSchools have trained their students as versatile researchers and professionals who are skilled in information management and data analysis. What made this possible is the multidisciplinary characteristics of iSchools whose faculty members come from diverse educational backgrounds, such as including Library and Information Science, Computer Science, Education, Communication, Management, History, Information Systems, English, Psychology, Philosophy, and Industrial Engineering (Zhu, Yan, & Song, 2016). This multidisciplinary human resource asset would enable iSchools to educate successful data scientists equipped with diverse skills with broad perspectives, compared to other departments that have faculty members with focused expertise. In this regard, iSchools are probably the most ideal institutions for teaching user-based and application-focused data science education.

In summary, big data drives the fourth Industrial Revolution, and iSchools should educate successful data scientists who are prepared for this revolution (Figure 1).

User-based data science education implies the training of a data scientist who solves data science problems in some application domains or who can use data science products effectively, rather than those who develop data science products. We argue that iSchools should focus on educating data science problem-solvers at application levels, rather than developers. There are several different types of data scientists, such as those with hardcore programming backgrounds, in-depth statistical knowledge, or extensive backgrounds in business issues. While all these types of data scientists have the same goal of creating values from data, their foci are somewhat different. Data scientists with hardcore programming skills are good at developing and implementing new algorithms, tools, and systems; they also try to make programs/systems effective and efficient. The data science education offered by computer science departments will do an appropriate job in this direction. Data scientists with strong statistical backgrounds are good at interpreting/analyzing data with advanced statistical techniques. Data scientists with business backgrounds try to validate business models from data with a focus on business analytics. Data science education or its cousin business analytics offered by many business schools focus on the latter two directions.

Compared with the aforementioned types, the suggested user-based data science education tries to produce data scientists who can think from the perspective of end users of data science products or problem-solvers in applications with broad perspectives on data science lifecycles from requirements to model-building, evaluation, deployments, and data curation. Most iSchool curricula emphasize concepts such as requirement modeling, metadata management, data and software lifecycles, human-centered computing and design, software or system evaluation, social aspects of information such as ethics of big data and data science, security and privacy issues of big data, visualization, and data curation. The coverage of these topics in computer science or business school curricula is not as extensive as those of iSchools, however. The iSchool curriculum has strong advantages in the user-based data science education by training students who understand the importance of requirement modeling, know the roles of metadata and utilize them, design and develop systems with human-centered usability in mind, consider security and privacy of data in all the stages of the data science lifecycle, validate analysis outcomes, perform appropriate story-telling for stakeholders, manage projects with keen insights on data science lifecycles, protect data and generated insights, utilize data and outcome ethically, and know how to set up strategies for data archiving and curation.

Examples of questions these data scientists address from the user-perspectives are (1) what are the major problems that online marketers want to solve using a data science approach and what are specific questions they need to ask to solve the problems? (2) what aspects of a specific data science product do the end users feel comfortable and uncomfortable with? and (3) what kinds of final reports/insights/ visualizations are most helpful in delivering key information to each stakeholder, and how should they be presented to them as a story? These questions can be effectively tackled only if we focus on those users’ needs, requirements, validations, and effectiveness of visualizations and presentations.

Based on these considerations, iSchools’ methods of teaching data science education should broadly emphasize actual problem-solvers or users who are going to use data science products. From the practical point of view, however, with the limited number of courses in any data science degree programs, there will be no luxury of having separate courses for the subjects listed above. Hence, the main ideas of those subjects should be strategically injected into several different hardcore data science courses such as Introduction to Data Science, Data Science Project Management, Communicating with Data, Visualization, the Capstone Project, and so forth.

One pitfall we need to guard against is to be careful not to turn our students into generalists who have no strong expertise of their own in specific areas. Even though students should have a broad understanding on various issues they will meet while they work on data science projects, it is imperative that each student develops his/ her own expertise in some aspects or areas in a data science lifecycle. In order to cultivate one’s own expertise, students could work with established data scientists and engage in real-world projects led by capable mentors. As the Capstone Project course will help achieve this goal, it should be a mandatory core course in any data science curriculum.

The purpose of the tool-based data science education at iSchools is to emphasize the importance of automated tools and utilization of available libraries. This toolbased data science education does not mean students just need to learn tools without learning programming languages. On the contrary, we claim that iSchool students should learn both high-level analytical languages and data science tools. Learning a programming language helps students think logically and computationally (which we discuss further in Section 4.2), which is a critical skill in any IT education. We recommend high-level scripting languages such as Python and R, instead of Java or C++. Compared to Java and C++, both Python and R are relatively easy to learn, and they have a wide collection of libraries and packages, which can be easily mixed with big data analytics frameworks such as Hadoop and Spark. The key point is to develop applications by utilizing existing libraries or Web services, not to implement algorithms, which is done in computer science education. Students should be exposed to a wide collection of available libraries and services and learn how to effectively use them, without having to “reinvent the wheel.” This approach will give students ample experience in developing new applications with existing libraries and help them focus on problem-solving rather than low-level coding.

In addition to high-level languages, students should also learn one or more major data science tools. There are three important reasons we emphasize the use of automated tools in data science education. First, students will more quickly understand the flow of data science projects by just clicking, without writing a code. Automated tools will handle data input, some data pre-processing, executing algorithms, and visualizing output. With automated tools, students can focus on business problems, the nature of input data, strengths and weaknesses of the algorithms executed, interpreting the output, and validating the results. These tools would allow them to quickly appreciate the value and importance of data science approaches, and support the idea of learning by doing.

Second, once students graduate from college, most will use tools or high-level data science languages in industry to work on real-world data science problems in some application domains. Hence, it makes sense to let them gain experience with popular automated data science tools and understand their strengths and limitations.

Third, automated tools are getting more powerful and diverse, with features such as data pre-processing, model-building, machine learning, validation, visualization, and report generation. Examples are IBM Watson Analytics

http://www.ibm.com/analytics/watson-analytics/

The shortage of qualified data scientists is already widely known as a challenging problem in industry and education

www.mckinsey.com/features/big_data/

The key points of application-based data science education at iSchools are threefold, developing (a) project-based education, (b) the ability to work with domain experts, and (c) the expertise in one application domain. First, students should be involved in data science projects in diverse forms and levels. In earlier courses, they should understand big data use cases to which data science disciplines are applied for digital transformation. Students should experience several complete case studies with exposure to each step of a data science lifecycle using an automated tool in the very early stage of the degree program. This will give them a big picture of a data science project without coding, allow them to think about a domain they want to work on further, and help them choose a specific area they want to develop their own expertise during the rest of their degree program. Then, there should be a capstone course that students work on as a team to address real-world data science projects, perhaps with a domain expert.

Second, students should learn how to work with domain experts and students of other disciplines as a team. Team work is generally important in any discipline, but this requirement cannot be emphasized too much in data science education as most data science problems are complex and highly interdisciplinary, and require a team effort. Data scientists should frequently communicate with each other, try to understand each other’s domain specific knowledge, work to make the data and data science methods understood by all parties, and try to visualize and present the results in a clear and jargon-free manner. Therefore, the ability to collaborate with domain experts or other members of a team is an essential skill.

Third, it is desirable for students to develop their own aptitude or expertise in one domain area. Completely and fully understanding a single domain requires tremendous time and effort, and it is thus impossible for a single data scientist to have in-depth knowledge of many domains. Students should be encouraged to develop expertise and in-depth knowledge in one domain area, as most data science problems are complex and require extensive experience. Students should also be encouraged to work on non-traditional domains to create innovative applications. For example, it is reported that IBM Watson can create recipes by merging multiple food ingredients, create a piece of music after learning an artist’s style, produce paintings, and create movie previews and scenarios. These applications are beyond traditional experiential avenues such as sales, marketing, healthcare, traveling, and so forth. Having an aptitude for innovation will give students opportunities to make contributions to the domain, and possibly open a new field.

Overall, iSchools’ data science educational strategies should focus on teaching user-based perspectives as a problem-solver in application domains, and tending to specific needs of end users of data science products. iSchools can teach both coding using a high-level scripting language and automated data science tools. The curriculum should provide several project-based experiences so that students can explore their own options and find a preferred domain.

Just as Information Science schools and their recent successors, iSchools, have focused on people, technology, and information, we also recommend people, technology, and data as the baseline of data science education of iSchools (Song & Zhu, 2016). The only difference in data science education is that the focus is on data instead of information in order to emphasize the data-driven paradigm we are currently working under. Based on these three pillars of data science, we propose to enforce computational thinking as a basic problem-solving framework. Data science courses should then be designed to broadly cover issues and techniques present in a data science lifecycle. The ways that data science courses are taught should accommodate the information computing principles in the forms of user-based, tool-based, and application-based, as noted above.

After being prepared with necessary background knowledge, students need to know how to solve big data problems with the use of computational thinking. Wing (2006) defines computational thinking as “a way of solving problems, designing systems, and understanding human behavior that draws on concepts fundamental to computer science. It represents a universally applicable attitude and skill set everyone, not just computer scientists, would be eager to learn and use” (Wing, 2006, p.33). Computational thinking is a set of thought processes necessary for solving problems using computers, used to develop all software products. As there are many software products to be applied by end users, it is regarded as a fundamental skill for everyone, especially as software becomes widespread and more important than hardware. Computational thinking is known as analytical thought that shares in many ways with mathematical thinking (solving problems), engineering thinking (designing systems), and scientific thinking (understanding human behavior) (Wing, 2008). With core concepts such as abstraction, decomposition, and patterns, Wing (2006) propose that a large, complex, and difficult problem can be reformulated into problems that we know, and these problems can be further solved using algorithms and data. Bundy (2007) stated that “computational thinking is influencing research in nearly all disciplines, both in the sciences and the humanities” (Bundy, 2007, p.67). For example, Wing (2008) noted several disciplines such as statistics, biology, economics, chemistry, and physics that have been influenced by computational thinking. As coding is the best way to learn computational thinking, students must learn coding as a means for solving problems systematically and thinking logically.

The fundamental principle below big data and data science is the data-driven paradigm (DDP). It is a belief that data is an asset and data tells the story, leading to the data-driven decision-making. iSchools have a unique position to motivate students to understand the DDP. Students should try to understand where big data leads us, how data is used in data science methods, what the impacts of the data-driven paradigm are to science, engineering, industry, and the society. Changes introduced by big data to the society are extensively discussed by Mayer-Schonberger and Cukier (2013).

One important point discussed by Mayer-Schonberger and Cukier (2013) is the notion of datatification in DDP. Datatification implies the creation of machine-processable data by digitizing the thoughts or behaviors of humans, machines, or objects. Some important activities here could include quantification and measuring of granular data, and transforming a phenomenon into quantified data for tabulation and analysis. An example could be datafication of personal lives, i.e. datafy user behaviors such as which music they listen to and how often and when, how often a user turns heaters on, and how often people go food shopping and when, etc. Creating a new data set via datatification will be a basis for innovative applications in many industries and disciplines.

Students should clearly understand how DDP enables new opportunities that were not possible in the past. Google is the most prominent example of a company that implemented data as a service model. Along with maintaining its position as the premiere search engine, Google has datafied our search keywords into Google Trends, and used collected data to translate one language to another. Big data enables us to ask new business questions and derive supporting decisions. As every new data source appears, we have to think about how they can be used and with which data they can be integrated to create synergy. For example, social network data, which has datafied users’ behaviors, is widely used for many different purposes—social marketing, mobile sales, opinion mining, election analysis, identifying most influential users, etc. Traditional points of sale data are analyzed together with weather data to see the impacts of weather on sales. And Twitter data could be combined with weather data and electronic healthcare data to predict the number of asthma patients coming to the emergency rooms, etc. When a data source is combined with another data source, new insights can be generated.

The DDP also leads to data-driven scientific discovery and data-driven economy. Evidence-based medicine has long been practiced in medical fields, which continue to invest heavily in data science management. In the science, applications of data science techniques to big scientific data are now a new paradigm in astronomy, physics, and biology. Data science is a fast-changing field and one needs to pay attention to the state-of-the-art scientific research to be on top of new developments. This helps data science projects be equipped with the best available and effective technologies based on past experiences and cutting-edge trends.

In addition to data-driven scientific discovery, understanding of data-driven markets, data-driven services, and data-driven economy is also necessary for student learning. Major IT companies such as Google, Facebook, IBM, and Amazon are rapidly expanding their data assets by acquiring companies that have accumulated data on markets, professional relationships (e.g. LinkedIn acquired by Microsoft), weather, news, and other topics, while collecting and analyzing massive amounts of the data to find new opportunities for innovative applications. Nowadays, many customers give more value to services that are delivered by products than the products themselves, and data science helps create high quality data-driven services. New types of businesses such as Uber and Airbnb are good examples of the data-driven economy, which utilizes data to sell business services, seriously cutting into the business of taxis and hotels.

Students should also learn many big data use cases, including how the DDP is applied in those use cases. Understanding existing cases would help them create a new big data use case and datatification ideas. When students know how to identify an important big data problem by combining multiple big data sources and applying systematic thinking to create new knowledge, products, or services, they will be able to join the field of big data specialists.

The specific steps used to solve data problems can be summarized as a data science lifecycle or data analytics lifecycle. We proposed a data analytics lifecycle in our previous study (Figure 6 in Song & Zhu (2016)), which we call it a data science lifecycle in this paper. As shown in Figure 4, the eight important steps include business understanding, data understanding, data preparation, model planning, model building, evaluation, deployment, and review and monitoring (Table 1).

Table 1

Eight steps of a data science lifecycle.

iSchools can embed all the steps of the data science lifecycle into a data science curriculum and structure courses based on it. The data science capstone project course can be a good platform for students to practise all the steps of the lifecycle.

In this section, we discuss two areas where iSchools have strengths over other disciplines—data science project management and big data curation (For modeling aspects of big data, we refer to Storey & Song, 2017).