Machine Learning

You might think of Now & Then as a "cognitive computer" that learns and adapts by monitoring your queries, the changing content and context of your journal,  and your unique journal use patterns. It also learns from feedback about your world, your life experiences, through "active" teaching, and your rating of query relevance. You jump start its education by:

• "initializing" its understanding of how you use language using a "term attribute scaling" exercise where you "rate" (from 1 to 10) attributes of words in your vocabulary,

• "constructing" rules using the RuleBuilder, and

• "populating" user-specific ontologies (for example, places: USA―California―San Francisco area―Muir Woods)

Two exercises establish Now & Then's initial understanding of your worldview.

1. Attribute analysis
Using the Attribute Scaling Utility you initialize, personalize, and update your journal's understanding about how you personally use language. The Utility prompts you to rank, or scale, the attribute-specific relationship between many sets of two words. An example of an attribute scaling is to rate, between 1 to 10, how closely you associate the word "Redwood City" with each of the terms: with "work" "home" "client" "Florida" "Golden Gate" "pizza" "mountains"

In Attribute Scaling analysis, the degree in which terms are associated is expressed using a variance and correlation matrix. The matrix is actually a multi-dimensional mathematical representation of a "space" or map of the many-to-many relationships of the terms you use in your journal. The coordinates for the terms and phrases used in your journal are mapped into hundreds of dimensions. Your queries are filtered through this multi-dimensional "map" of your "attribute-scaled" vocabulary to chart the similarities or dissimilarities (distances) between terms and phrases―and to detect meaningful underlying language relationships and structures.

To simplify the concept, reduce "multi-dimensions" down to just three―with the coordinates for the words "dolphin" "Florida" and "football"  plotted on X, Y, Z axis. Their relationships are defined by their coordinates in three―rather than N dimensions.     
3 dimensions Florida - dolphin- football
                                                                
2. Worldview Ontologies                                
Similarly, you provide the initial values of "knowledge maps" or ontologies that capture the relative relationships and manage descriptions about the people, places, things, and ideas described by you in your journal―this is knowledge about knowledge, or the meta-knowledge of your worldview.

Using Now & Then's ontology entry utility, you can add more information to your ontologies at any time. For example, you might create groups called "neighbors", "work associates", or "best friends" and then add the names and / or nicknames of individuals. Or, you might create "places" concept groups such as "local", "regional", or "business trips" and populate each with the "place" names that you reference in your journal. You can also manage your ontologies to add detail and clarify ambiguities―for example, "for the years 1998 through 2010, "regional" is the San Francisco area. For the years 1980 -1984 "regional" is metro Santa Barbara.

home