Fluent Essays

Describe your opinion OR Summarize it* Word limit: minimum 100 wordsDescribe your opinion OR Summarize it* Word limit: minimum 100 words participation.pdf Unformatted Attachment Preview “The Discipline of Organizing” Meets “Data Science” Position Paper for “Information Science to Data Science” Workshop at iConference 2017 ABSTRACT Many books and articles about data science, machine learning, and predictive analytics make bold predictions that these emerging fields will fundamentally change the world. Most of the most provocative claims forecast radical changes to how information is organized and used, two activities that are at the core of the iSchools. The goal of this short article is to propose bridges between the concepts and methods of resource organization, description, and interactions with resources as they are typically taught in iSchools with the concepts and methods of data science. These bridges should make it easier for iSchools to view data science as a complementary discipline rather than a contrasting and competitive one. KEYWORDS Organizing; classification; data science; machine learning; sensemaking 1 INTRODUCTION Wikipedia begins its definition of “Data Science” with this puzzling note (Figure 1) that should provoke anyone associated with an iSchool to reflect on their own and their institution’s perspective on data science. Many books and articles about data science, machine learning, and predictive analytics make bold predictions that these emerging fields will fundamentally change the world ([1], [2], [3], etc.). Most of the most provocative claims forecast radical changes to how information is organized and used, two activities that are at the core of the iSchools. Figure 1: Wikipedia Defines “Data Science”. It is undeniable that the new methods and tools of data science and machine learning let us organize more information, to do it faster, and to make predictions based on what people have clicked on, bought, or said. Data science introduces new considerations of scale and speed when massive computational power and new statistical techniques are harnessed to organize and act on information. Some data science enthusiasts go so far as to say that because machine learning algorithms can discover interesting patterns in data, there is much less need for systematic resource organization and description. Many iSchools have responded strongly to data science, and they are doing mostly the same things: adding new courses, modifying existing ones, hiring new faculty, and even changing the names of their schools or departments to include “data.” Some iSchools teach some of the algorithms and technology of data science, but they are unlikely to dislodge computer science and statistics departments as their primary home. iSchools differ in where they view data science on the continuum from “opportunity” to “threat,” but an impartial observer can easily see concerns at iSchools that data science will concentrate its academic resources in computer science and statistics and possibly undermine their own missions and appeal to students and faculty. This article suggests that iSchools can and should assert themselves as first class citizens with respect to data science because at its core, data science is about how resources are selected, described, and organized; concepts and methods with a long tradition in information and library science and that remain foundational topics in iSchool curricula. Instead of organizing and describing the books in a library or the products in a physical warehouse, a data scientist might create a data warehouse by organizing information about books or products into massive data tables, treating each resource as a row and its descriptive properties as the columns. After people might have organized books or products into categories, machine learning techniques might classify new books or products using those categories, or perhaps discover new categories based on access or purchasing behaviors. But while the techniques of data science are new, many of the challenges are not; data scientists need to select resources wisely and decide how best to describe them; they need to understand that resource description and categorization can be biased; they need to understand the tradeoffs and complements between people and computers; and, they need to test the discoveries that algorithms make with controlled experiments. The goal of this short article is to propose bridges between the concepts and methods of resource organization, description, and interactions with resources as they are typically taught in iSchools with the concepts and methods of data science. These bridges should make it easier for iSchools to view data science as a complementary discipline rather than a contrasting and competitive one. 2 THE DISCIPLINE OF ORGANIZING The Discipline of Organizing (TDO) [4], first published in 2013, is a novel synthesis of insights and perspectives from the fields of Information Science to Data Science Workshop, iConference 2017 library and information science, computer science, informatics, business, law, cognitive science, among several others in which concepts and methods about organizing resources are foundational. Computer scientists and library/information science professionals often address the same questions without realizing it. Instead of emphasizing the specific types of organizing systems that each of those fields is known for – libraries, museums, data repositories, business information systems, human resource organizations, and so on – TDO recasts organizing as a design activity to answer questions about what is organized and why, how much, when, by what means, and where the organizing is accomplished. This “design dimensions” or “design patterns” perspective is forwardlooking and generative, and its more abstract concepts and vocabulary enable interdisciplinary collaboration and innovation not previously possible because of the unifying concept of an “organizing system” – an intentionally arranged collection of resources and a set of supported interactions. In 2014, TDO was named an “Information Science Book of the Year” by the Association for Information Science and Technology. TDO is now used as a primary or secondary text for (mostly) graduate and undergraduate courses in more than 80 schools (mostly iSchools) in over 20 countries. TDO has been substantially revised each year since it was first published, which is unusual but which ratifies the idea that multiple perspectives can reinforce a shared focus on organizing, while at the same time highlighting the concepts, technologies, and methods that distinguish those points of view. The 4th edition [5], published in August 2016, builds bridges between organizing and data science, and introduces a new “data science” category of disciplinespecific content. 3 THE BRIDGES BETWEEN ORGANIZING AND DATA SCIENCE The bridges in “The Discipline of Organizing” to data science include broadening the concept of “usage metadata,” reframing descriptive statistics as organizing techniques, expanding the treatment of classification to include computational methods, and emphasizing examples of data-driven resource selection, organization, interaction, and maintenance. The following sections of this short article are adapted excerpts from the book. R science has long discussed how records of a user choice in accessing, browsing, buying, highlighting, linking, and other interactions become “usage metadata” that can inform the design and operation of information systems. When organizing systems contain digital resources, or physical resources that have sensing, recording, or communication capabilities, these interaction traces can be made predictable, persistent, and consistent. TDO introduces the concept of “interaction resource” to broaden the traditional information science concept of “usage metadata” to include data collected from the “Internet of Things” and other dataproducing resources. The most common use of interaction resources in organizing systems is in search engines to adjust the order of search hits, select ads, or personalize the content of web pages. 3.2 Common activities for organizing “data” and everything else Every organizing system requires four activities: selecting resources, organizing them, designing interactions with them, and maintaining the resources and the interactions. We can see these activities in the organization of a kitchen, a library, a museum, or an enterprise information system. How explicit these activities are depends on the scope, the breadth or variety of the resources, and the scale, the number of resources that the organizing system encompasses. In the organizing systems for the “memory institutions” studied in many iSchools, many of these common activities have domain-specific vocabulary like “acquisition,” “accessioning,” and “ingesting” but using more neutral language like “adding resources to a collection” enables more interdisciplinary collaboration and innovation. These same four activities can be seen in data science, and doing so makes it straightforward to view the field of data science as a subdiscipline of information organization. Data is selected, described to enable its organization, and then interacted with” using algorithms that enable classification, prediction, recommendation, inference, and hypothesis testing. Finally, the data in an organizing system is maintained over time, with discipline-specific words like “governance,” “de-duplication,” and “cleansing,” but it is easy to recognize the commonalities with maintenance activities for “nondata” resources like “curation” and “restoration” for physical resources, and “layoff” or “retirement” for human resources. 2 . Glushko 3.1 The Concept of “Interaction Resource” (“usage metadata”) Data science obviously relies on data, but where does the data come from? A great deal of it is created when people interact with organized resources or when resources with agency interact with each other or with their environments. But this data that provides much of the fodder for data science is similar in concept to interaction traces like fingerprints, a coffee cup stain on a newspaper, or the erosion on a shortcut path across a lawn, where the record of the interaction is less persistent and consistent only because there is no separate technology for capturing it. Information One undeniable impact of data science on organizing systems of all types is that these four activities have becoming increasingly datadriven. 3.2.1 Data-driven Resource Selection Selection methods and criteria vary across resource domains, but their common purpose is to determine how well the resource satisfies the specifications for the properties or capabilities that enable a person or nonhuman agent to perform the intended activities. Data science has enabled the selection of human-resources to become highly data-intensive; employers hire people after assessing the match between their competencies and capabilities Information Science to Data Science Workshop, iConference 2017 (expressed verbally or in a resume, or demonstrated in some qualification test) and what is needed to do the required activities. The popular LinkedIn site, which has hundreds of millions of resumes that it data mines to find statistically superior job candidates, is literally a gold mine for the company because it makes money by referring those candidates to potential employers. Data-intensive hiring practices in baseball are entertainingly presented in the book entitled Moneyball [6] or the 2011 movie starring Brad Pitt. 3.2.2 Data-driven Organizing Organizing systems arrange resources according to many different principles. Most organizing systems use principles that are based on specific resource properties or properties derived from the collection as a whole. What properties are significant and how to think about them depends on the number of resources being organized, the purposes for which they are being organized, and on the experiences and implicit or explicit biases of the intended users of the organizing system. Data science also needs to choose resource properties (although it usually calls them “features” rather than “properties”). Another way to bridge data science with organizing is to recognize that using descriptive statistics about a collection or dataset is a way to summarize it concisely and identify the properties that might be most useful as organizing principles. Statistical descriptions can be created for any resource property, with the simplest being the number of resources that have the property or some particular value of it, such as the number of times a particular word occurs in a document or the number of copies a book has sold. Descriptive statistics and associated visualizations can suggest which properties make good organizing principles because they exhibit enough variation to distinguish resources in their most useful interactions. For example, it probably isn’t useful to organize books according to their weight because almost all books weigh between ½ and 2 pounds, unless you are in the business of shipping books and paying according to how much they weigh. .. Glushko No matter how measurements are distributed, it can be useful to employ descriptive statistics to organize resources or observations into categories or quantiles that have the same number of them. Quartiles (4 categories), deciles (10), and percentiles (100) are commonly used partitions. Descriptive statistics do not identify the categories they create by giving them familiar cultural or institutional labels the way that human organizers do. Instead, they create implicit categories of items according to how much they differ from the most typical or frequent ones. For example, in any dataset where the values follow the normal distribution, statistics of central tendency and dispersion serve as standard reference measures for any observation. These statistics identify categories of items that are very different or statistically unlikely outliers, which could be signals of measurement errors, poorly calibrated equipment, employees who are inadequately trained or committing fraud, or other problems. Categories that people create and label also can be used more explicitly in computational algorithms and applications. In machine R learning [2], a program that can assign an item or instance to one or more existing categories is called a classifier. Numerous techniques for creating classifiers do so via training them with already correctly categorized examples. This is called “supervised learning” because it starts with instances labeled by category, and it involves learning because over time the classifier improves its performance by adjusting the weights for features that distinguish the categories. An important difference between categories created by people and those created computationally is that the former can almost always be inspected and reasoned about by other people, but only some of the latter can. A computational model that categorizes loan applicants as good or poor credit risks probably uses properties like age, income, home address, and marital status, so that a banker can understand and explain a credit decision. However, many other computational categories, especially those that created by clustering and deep learning techniques, are inseparable from the mathematical model that learned to use them, and as a result are uninterpretable by people. 3.2.3 Data-driven Interactions Organizing systems can use stored or computed information about user preferences or past interactions to anticipate user needs or personalize interactions. This has the effect of substituting information for interaction to make interactions unnecessary or simpler. For example, drivers no longer need to stop at toll booths, because most human toll-takers have been replaced by smart “toll tags” that broadcast their identity when the car they are in passes a radio receiver at a tolling location. Each interaction resource created identifies an account and credit card with which to pay the toll; taken together, the collection of these interaction resources can be used as the primary resources in other organizing systems that 3 manage traffic congestion, make recommendations to drivers about detours, or that support road design. 3.2.4 Data-driven Maintenance Recent developments in sensor technology enable very extensive data collection about the state and performance of machines, engines, equipment, and other types of physical resources, including human ones. (Are you wearing an activity tracker right now?) When combined with historical information about maintenance activity, predictive analytics techniques can use this data to determine normal operating ranges and indicators of coming performance degradation or failures [3]. Predictive maintenance can maximize resource lifetimes while minimizing maintenance and inventory costs. 4 BUILDING BIGGER BRIDGES: FROM SENSEMAKING TO SCIENCE Bridging from organizing to data science is an assertion that the disciplines are complementary rather than competitive, and we can strengthen the connection by situating both in a larger context of human activity that includes sensemaking and science. Information Science to Data Science Workshop, iConference 2017 4.1 Sensemaking as “the Mother of Organizing” For thousands of years, even before the invention of written language, people have systematically collected things, information about those things, and observations of all kinds to understand how their world works. Paleolithic humans made cave paintings depicting the results of hunts and animal migrations; ancient Egyptians recorded the annual floods of the Nile River in stone carvings; and Babylonian, Egyptian, Chinese, and Mesoamerican astronomers organized lunar, solar, and planetary observations as calendars starting about five thousand years ago. These diverse efforts to impose meaning on experience by recording, analyzing, organizing, and reorganizing observations can be collectively described as sensemaking ([7], [8]). R and refine models of any type by systematically varying the conditions under which observations are made to discover how the results change in different situations. 4.4 From Data Science to Science A fundamental challenge in sensemaking and modeling is balancing the competing goals of understanding a particular collection or dataset and applying that understanding to new instances. Models can differ in the number of resource descriptions they use as parameters, and it is easy and tempting to overfit a model by using more parameters that capture random variations in observations. Overfitting produces spurious accuracy in reproducing the original observations, but it makes models less generalizable. Some sensemaking is hard-wired by evolution in perceptual mechanisms that simplify and organize the information our senses collect (like Gestalt mechanisms), but that sort of automatic sensemaking is dominated by intentional sensemaking. The highest level of sensemaking is the creation of scientific theories, with a preference for “more organized” or simpler explanations for the observations, the principle known as Occam’s razor. This is the bridge between data science and science. 4.2 From Sensemaking to Organizing REFERENCES Making sense of a single resource collection involves identifying the properties and principles that contrast the instances or observations. This is the bridge between sensemaking and organizing. Organizing systems use those principles and properties to enable more welldefined and functional interactions. Sensemaking becomes more intentional when systematic statistical, experimental, and scientific methods are consciously followed to extract and organize knowledge from collections of samples, observations, or measurements. 4 ... Purchase answer to see full attachment