14 How Much Is It Being Organized?
“It is a general bibliographic truth that not all documents should be accorded the same degree of organization.”
Not all resources should be accorded the same degree of organization. In this section we will briefly unpack this notion of degree of organization into three important and related dimensions: the amount of description detail or organization applied to each resource, the amount of organization of resources into classes or categories, and the overall extent to which interactions in and between organizing systems are shaped by resource description and arrangement.
It is important to note that this section is not asking the question “how much stuff is being organized?” but rather to what degree is the stuff being organized. Another way to ask the same question is “how many organizing principles are at work?” in this organizing system. Your closet might be arranged only by body part covered and season; an online music store will organize resources by genre, artist name, band name, album name, popularity, date released, and maybe others. So we would say that the online music store is organized much more than the closet, because more organizing principles are at work.
(Resource Description and Metadata and Categorization: Describing Resource Classes and Types, more thoroughly address these questions about the nature and extent of description in organizing systems.)
Not all resources in a collection require the same degree of description for the simple reason we discussed in “Why Is It Being Organized?”: Organizing systems exist for different purposes and to support different kinds of interactions or functions. Let us contrast two ends of the “degree of description” continuum. Many people use “current events awareness” or “news feed” applications that select news stories whose titles or abstracts contain one or more keywords (Google Alert is a good example). This exact match algorithm is easy to implement, but its all-or-none and one-item-at-a-time comparison misses any stories that use synonyms of the keyword, that are written in languages different from that of the keyword, or that are otherwise relevant but do not contain the exact keyword in the limited part of the document that is scanned. However, users with current events awareness goals do not need to see every news story about some event, and this limited amount of description for each story and the simple method of comparing descriptions are sufficient.
On the other hand, this simple organizing system is inadequate for the purpose of comprehensive retrieval of all documents that relate to some concept, event, or problem. This is a critical task for scholars, scientists, inventors, physicians, attorneys and similar professionals who might need to discover every relevant document in some domain. Instead, this type of organizing system needs rich bibliographic and semantic description of each document, most likely assigned by professional catalogers, and probably using terms from a controlled vocabulary to enforce consistency in what descriptions mean.
Similarly, different merchants or firms might make different decisions about the extent or granularity of description when they assign SKUs because of differences in suppliers, targeted customers, or other business strategies. If you take your car to the repair shop because windshield wiper fluid is leaking, you might be dismayed to find that the broken rubber seal that is causing the leak cannot be ordered separately and you have to pay to replace the “wiper fluid reservoir” for which the seal is a minor but vital part. Likewise, when two business applications try to exchange and merge customer information, integration problems arise if one describes a customer as a single “NAME” component while the other separates the customer’s name into “TITLE,” “FIRSTNAME,” and “LASTNAME.”
Even when faced with the same collection of resources, people differ in how much organization they prefer or how much disorganization they can tolerate. A classic study by Tom Malone of how people organize their office workspaces and desks contrasted the strategies and methods of “filers” and “pilers.” Filers maintain clean desktops and systematically organize their papers into categories, while pilers have messy work areas and make few attempts at organization. This contrast has analogues in other organizing systems and we can easily imagine what happens if a “neat freak” and “slob” become roommates.
An equally wide range, from a little organization to a lot, can be seen in the organizing systems for businesses, armies, governments, or any other institutional organizing systems for people. Organizations with broad scope and many people usually have deep hierarchies and explicit reporting relationships with the CEO, general, or president at the top with numerous layers of vice presidents, directors, department heads, and managers (or colonels, majors, captains, lieutenants, and sergeants). Smaller organizations are more varied, with some embodying multi-layered management, and some embracing a flatter arrangement with fewer management levels, wider spans of authority, and more autonomy for individual workers. Many start-up firms try to grow without any management structure at all in the belief that it makes them more innovative and nimble, but evidence suggests that when no one is responsible for making decisions, the lack of accountability results in poor decisions, or in no decisions at all even when some were sorely needed.
In any case, when people have to do it, describing and organizing resources is work. Stakeholders in an organizing system often have disagreements among about how much organization is necessary because of the implications for who performs the work and who derives the benefits, especially the economic ones. Physicians prefer narrative descriptions and broad classification systems because they make it easier to create patient notes. In contrast, insurance companies and researchers want fine-grained “form-filling” descriptions and detailed classifications that would make the physician’s work more onerous.
The cost-effectiveness of creating systematic and comprehensive descriptions of the resources in an information collection has been debated for nearly two centuries, beginning in 1841 when Sir Anthony Panizzi proposed rules for cataloguing the British Library. In the last half century, the scope of the debate grew to consider the role of computer-generated resource descriptions.
The amount of resource description is always shaped by the currently available technology for capturing, storing, and making use of it. Nineteenth century geologists and paleontologists typically recorded only general information about the depth and surrounding geological features when they found fossils because they had no technology for making more precise measurements and everything they noted they had to record by hand. Today, vastly more detailed information is recorded by instruments and exploited by sophisticated techniques for carbon dating and 3D reconstruction.
Automatically generated descriptions are increasingly an alternative or complement to those created by people. “Smart” resources use sensors to capture information about themselves and their environments (see “Identity and Active Resources”). Our own computers and phones record information about our keystrokes, clicks, communications, and locations. Business and government computers analyze and index most of the text and speech content that flows through and between our personal phones and computers. These indexes typically assign weights to the terms according to calculations that consider the frequency and distribution of the terms in both individual documents and in the collection as a whole to create a description of what the documents are about. These descriptions of the documents in the collection are more consistent than those created by human organizers. They allow for more complex query processing and comparison operations by the retrieval functions in the organizing system. For example, query expansion mechanisms can automatically add synonyms and related terms to the search. Additionally, retrieved documents can be arranged by relevance, while “citing” and “cited-by” links can be analyzed to find related relevant documents.
It is important to recognize the potential downside to automated resource description. A detailed description produced by sensors or computers can seem more accurate or authoritative than a simpler one created by a human observer, even if the latter would be more useful for the intended purposes. Moreover, the more detailed the description, the greater the opportunity to use it for new purposes. This might be desirable, as when a company realizes that it can cross- and up-sell because it has been tracking every click in a web store to create a collection of interaction resources. But it could be undesirable, because detailed transaction data can be used to violate privacy and civil rights. It depends on who controls the collected information and their incentives for using it or not using it.
A second constraint on the degree of organization comes from the size of the collection within the scope of the organizing system. Organizing more resources requires more descriptions to distinguish any particular resource from the rest, and more constraining organizing principles. Similar resources need to be grouped or classified to emphasize the most important distinctions among the complete set of resources in the collection. A small neighborhood restaurant might have a short wine list with just ten wines, arranged in two categories for “red” and “white” and described only by the wine’s name and price. In contrast, a gourmet restaurant might have hundreds of wines in its wine list, which would subdivide its “red” and “white” high-level categories into subcategories for country, region of origin, and grape varietal. The description for each wine might in addition include a specific vineyard from which the grapes were sourced, the vintage year, ratings of the wine, and tasting notes.
At some point a collection grows so large that it is not economically feasible for people to create bibliographic descriptions or to classify each separate resource, unless there are so many users of the collection that their aggregated effort is comparably large; this is organizing by “crowdsourcing.” This leaves two approaches that can be done separately or in tandem.
The second approach is to rely on automated and more general-purpose organizing technologies that organize resources through computational means. Search engines are familiar examples of computational organizing technology, and “Computational Classification” describes other common techniques in machine learning, clustering, and discriminant analysis that can be used to create a system of categories and to assign resources to them.
Finally, we must acknowledge the ways in which information processing and telecommunications technologies have transformed and will continue to transform organizing systems in every sphere of economic and intellectual activity. A century ago, when the telegraph and telephone enabled rapid communication and business coordination across large distances, these new technologies enabled the creation of massive vertically integrated industrial firms. In the 1920s, the Ford Motor Company owned coal and iron mines, rubber plantations, railroads, and steel mills so it could manage every resource needed in automobile production and reduce the costs and uncertainties of finding suppliers, negotiating with them, and ensuring their contractual compliance. Adam’s Smith’s invisible hand of the market as an organizing mechanism had been replaced by the visible hand of hierarchical management to control what Ronald Coase in 1937 termed “transaction costs” in The Nature of the Firm.
In recent decades, a new set of information and computing technologies enabled by Moore’s law—unlimited computing power, effectively free bandwidth, and the Internet—have turned Coase upside down, leading to entirely new forms of industrial organization made possible as transaction costs plummet. When computation and coordination costs drop dramatically, it becomes possible for small firms and networks of services (provided by people or by computational processes) to out-compete large corporations through more efficient use of information resources and services, and through more effective information exchange with suppliers and customers, much of it automated. Herbert Simon, a pioneer in artificial intelligence, decision making, and human-computer interaction, recognized the similarities between the design of computing systems and human organizations and developed principles and mechanisms applicable to both.
The Forms of Resource Descriptions, focuses on the representation of resource descriptions, taking a more technological or implementation perspective. Interactions with Resources, discusses how the nature and extent of descriptions determines the capabilities of the interactions that locate, compare, combine, or otherwise use resources in information-intensive domains.
(Malone 1983) is the seminal research study, but individual differences in organizing preferences were the basis of Neil Simon’s Broadway play The Odd Couple in 1965, which then spawned numerous films and TV series.
See Grudin’s classic work on non-technological barriers to the successful adoption of collaboration technology (Grudin 1994).
Panizzi is most often associated with the origins of modern library cataloging. He (Panizzi 1841) published 91 cataloging rules for the British Library that defined authoritative forms for titles and author names, but the complexity of the rules and the resulting resource descriptions were widely criticized. For example, the famous author and historian Thomas Carlyle argued that a library catalog should be nothing more than a list of the names of the books in it. Standards for bibliographic description are essential if resources are to be shared between libraries. See (Denton 2007), (Anderson and Perez-Carballo 2001a, 2001b).
Information theory was developed to attack the technical problem of packing the maximum amount of data into the signal carrying telephone calls, but it quickly provided an essential statistical foundation in language analysis and computational linguistics. (Shannon 1948). Company organization and other examples applying information theory to the analysis of organizing systems can be found in (Levitin 2014, Chapter 7).
Coase won the 1991 Nobel Prize in economics for his work on transaction costs, which he first published as a graduate student (Coase 1937). Berkeley business professor Oliver Williamson received the prize in 2009 for work that extended Coase’s framework to explain the shift from the hierarchical firm to the network firm (Williamson 1975, 1998). The notion of the “visible hand” comes from (Chandler 1977). Simon won the Nobel Prize in economics in 1978, but if there were Nobel Prizes in computer science or management theory he surely would have won them as well. Simon was the author or co-author of four books that have each been cited over 10,000 times, including (Simon 1997, 1996) and (Newell and Simon 1972).