2/14/00 abstract
Thompson, B., and Thompson, B. (1991). Overturning the category bucket. Byte, 16(1), 249-255.
Categorization and categorization are important topics for AI. Classical AI methods learn to classify based on a set of training examples. Conceptual clustering is a categorization method based on finding discriminating features between objects.
2/14/00 abstract
Jacob, E. K. (1991). Classification and categorization: drawing the line. In B. H. Kwasnik and R. Fidel (Eds.), Advances in classification research, vol. 2 (p. 67-83). Washington D.C.: American Society for Information Science.
Contrasts classification and categorization. Classical categorization theory states that a category is a set of features whose union is necessary and sufficient for determining membership. New reseach indicates that the categories people use are fuzzy and dynamic. Categorization refers to this fuzzy, mental partitioning of the world. Classification refers to a rigid class structure based on the classical theory.
2/14/00 abstract
Parsons, J., and Wand, Y. (1997). Choosing classes in conceptual modeling. Communications of the ACM 40 (6), 63-69.
Discusses methods for creating classes in AI systems. Fewer classes provide less information about their members. More classes increase the cognitive workload necessary to deal with them. Classes (concepts) should be formed to support cognitive economy and inference. A class should be nonempty, and its definition should contain all properties common to its members. Every property in the universe should be used in the definition of a class. Classes should not be defined in terms of other classes.
2/14/00 abstract
Shera, J. H. (1965/1950). Classification as the basis of bibliographic organization. In Libraries and the organization of knowledge (p. 77-96). Hamden, CT: Archon.
Argues that classification is the heart of organizing books. Summarizes the history of classification. Hierarchical classification isn't sufficient, since document content is poly-dimensional. A ``natural'' classification is dependent on a purpose, and may change over time. Computers allow us to separate the knowledge from the physical document, so that one book may appear in several places of the classification scheme.
2/15/00 misc
(Alan asked about AI being able to categorize in a human way)
First, a disclaimer: As you mentioned, this article is a bit dated, and even then (being in Byte magazine rather than a research journal) it was hardly on the cutting edge. AI, being a relatively new field, changes quickly.
No, AI isn't human, and making it human doesn't seem to be on the horizon. But AI can learn from the ways humans do things. Human categorization is affected by the experiences, context, and goals of the categorizer.
If we want AI systems to categorize like humans, they must be able to experience things in the real world. Some AI people do this by giving their systems eyes, ears, legs, etc. and letting the systems form their own mental models of the world. Others believe that an AI system can be fed knowledge (through a keyboard), and corrected when it makes mistakes, so that it will have a head start at creating a mental model.
On the other hand, if we want AI systems to categorize for humans, they must have access to the experience, context, and goals of the humans they're working for. This is tricky. Humans don't want to spend hours typing this information, but sometimes the system already has access to it. Your web browser keeps track of your browsing history. It should be able to use that information to classify and/or categorize new pages for you. Your word processor knows about the documents you're working on. It should be able to tell your web browser, so the browser can do its job better.
2/16/00 abstract
Soergel, D. (1999). The rise of ontologies or the reinvention of classification. Journal of the American Society for Information Science, 50(12), 1119-1120.
Discusses the different purposes for making classifications (or ontologies). Library science, AI, linguistics, natural language processing, and software engineering are all building ontologies. These disciplines would benefit from working with each other.
2/16/00 assigment
Return to the local grocery store that you visited for Assignment 5. In your journal, write an analysis of the store's organization that focuses on: (1) the explicit and/or implicit categories indicated by the organization of merchandise; (2) why you think this particular organizational structure was adopted; and (3) how this organizational scheme accords with a typical shopper's mental model of a grocery store. Bring your analysis of the store's organization to class.
The categorization scheme looks like this:
perishables
-- floral??
-- deli
-- baked
-- bread
-- made at Kroger
-- premade
-- veggies
-- packaged
-- common
-- round
-- uncommon
-- fruit
-- seafood
-- frozen
-- fresh
-- meat
-- fresh
-- frozen
-- packaged/lunchmeat
-- dairy
-- milk
-- yogurt
-- cheese
-- butter
-- ice cream
nonperishables
-- health/gormet
-- canned on sale
-- books
-- pharmacy
-- bath/hygene products
-- seasonal
-- misc
-- canned veggies
-- baking (flour, sugar, pie pans...)
-- canned fruit
-- breakfast (cereal, bananas, oatmeal...)
-- soup
-- italian (pasta, sauces, canned tomato goods...)
-- mexican
-- canned meat
-- international (rice, spice mixes, soy sauce...)
-- paper products
-- snack food
-- liquor
-- frozen
-- prepared
-- quick meals
-- veggies
-- video
-- greeting cards
Kroger seems to be arranged mostly for ease of use by the people who work there. Like many grocery stores, perishable items are placed around the edges of the store, probably to facilitate inspection and restocking of items. Therefore, the perishable category above is wrapped around the nonperishable in the actual store. The frozen items are all together, presumably to save money by having limited freezer space. Sale and seasonal items are grouped together for easy replacement. The bakery and pharmacy are placed back-to-back, which probably saves money by requiring fewer walls.
The nonperishable misc items, which take up a third of the store, don't place any particular constraints on the employees, so they may be categorized for the customers. This section is a strange mix of categorizing by use and categorizing by type of food. The canned goods (soups, veggies, fruits) tend to be organized by food type, while the more strangely shaped items (flour, cereal, spices, pasta) tend to be organized by use.
Some items appear in more than one place. Bananas can be found in the fruit section as well as the breakfast section. Yeast is often in the baking section, but also in the prepared meat section (by the pepperoni). Other items that (in my mind) should be in several places are not. Canned tomatos are in the italian section, but not the canned fruit or veggie section.
Since the overall arrangement is for the benefit of the store rather than the customer, I doubt that it is a ``natural'' arrangement according to anyone's mental model. However, many grocery stores are arranged in a similar manner, so experienced shoppers may have developed a mental model that closely matches the store layout.
2/20/00 class notes
Jacob Says:
``We don't discover reality, we create it [through our category models]''
2/20/00 group work
Our group project was to design a grocery story for a single woman, age 18.
The group decided that the ``normal'' staples weren't very necessary, since this group would tend to cook less than older people or people with family members in the house.
We came up with the following sections:
party chips cake mix quick-cook pasta create-a-meal frozen dinners non-food lightbulbs eat now fresh fruit deli items dating makeup cards condoms
Our arrangement placed the get-quick items (party and dating sections) near the doors. This allows customers who only want a few things to get in and out quickly. It also makes sure that customers who want other items know that these items are available. The non-food items are placed in the center of the store, and the other categories are arranged around the edges. Though we didn't explicitly try, this arrangment is a good approximation of the ``power alley'' found in other grocery stores.