The data management problem, whether solved by a machine or a personal secretary, can be divided into the following stages:
Finding new information.
Evaluating new information.
Linking new information to old information.
Detecting old, unused information.
Displaying linked information and interacting with the user.
Learning how to better display linked information, detect unused information, link current information, evaluate new information, and find new information based on the user's actions.
These stages can be couched in terms of
the actions of a number of simpletons comprising a primitive mind:
Collectors find, retrieve, or accept new pages likely to interest the user (from the web and ftp sites, mail and news, and the desktop). They answer the question: "What's out there?" They are like effectors.Information Clustering:
Examiners parse collected pages to determine their detectable characteristics. They answer the question: "What properties does it have?" They are like sensors.
Clusterers group entities into clusters of related entities. They answer the question: "What's it like?" (or perhaps "Where do I put it?"). They are like memory.Information Analysis:
Cluster mergers decide whether two clusters should be merged into one.
Cluster splitters decide whether a cluster should be split into two new clusters.
Cluster creators decide whether a new cluster should be created.
Cluster killers decide whether a cluster should die. (Note that a cluster can die yet the entities it contains could survive the death.)
Comparators determine the characteristics of entities that distinguish them from other entities. They answer the question: "What makes it special?" They are like perception.User Analysis:
Evaluators test entities for desirable characteristics. They answer the question: "Will my user like it?" They are like emotions.
Mappers map entities into a virtual space. They answer the question: "How do I display it?"
Pollsters determine what entities most interest the user. They answer the question: "What does my user like?"Meta-Analysis:
Purgers mark old, unused entities. They answer the question: "What does my user dislike?"
Modelers order entities based on which ones seem to be most useful to the user now. They answer the question: "What is my user likely to be searching for now? (or soon?)"
Orderers order entity attributes based on which ones seem to be most predictive of strong user interest.
Advisors I: figure out how to better determine which entities the user is likely to be interested in.
Advisors II: figure out which entities are likely to be representative of the current most interesting entities.
Cluster Advisors figure out how to better cluster related entities.
Filter Advisors figure out tests to tell whether a new page might be interesting. They answer the question: "What is likely to interest my user in future?"
Analyst Advisors figure out what new entity characteristics to look for. They answer the question: "What shall I look for next?" They are like curiosity (?)
Collector Advisors figure out where to look for new interesting entities. They answer the question: "Where shall I look next?" (or perhaps "What should I focus on next?") They are like attention.