Model Specification | Attribution | Teradata Vantage - Model Specification - Analytics Database

Database Analytic Functions

Deployment
VantageCloud
VantageCore
Edition
Enterprise
IntelliFlex
VMware
Product
Analytics Database
Release Number
17.20
Published
June 2022
Language
English (United States)
Last Update
2024-04-06
dita:mapPath
gjn1627595495337.ditamap
dita:ditavalPath
ayr1485454803741.ditaval
dita:id
jmh1512506877710
Product Category
Teradata Vantageā„¢

Model Types and Specification Definitions

Row 0: Model Type Row 1, ..., n: Distribution Model Specification Additional Information
SIMPLE MODEL:PARAMETERS Distribution model for all events. For MODEL and PARAMETER definitions, see following table.
EVENT_REGULAR EVENT:WEIGHT:MODEL:PARAMETERS Distribution model for a regular event.

EVENT cannot be a conversion, excluded, or optional event.

For MODEL and PARAMETER definitions, see following table.

Sum of WEIGHT values must be 1.0.

For example, suppose that model table has these specifications:

  email:0.19:LAST_CLICK:NA

  impression:0.81:UNIFORM:NA

Within WindowSize of a conversion event, 19% of conversion event is attributed to last email event and 81% is attributed uniformly to all impression events.

EVENT_OPTIONAL EVENT:WEIGHT:MODEL:PARAMETERS Distribution model for an optional event.

EVENT must be in optional event table.

For MODEL and PARAMETER definitions, see following table.

Sum of WEIGHT values must be 1.0.

SEGMENT_ROWS Ki:WEIGHT:MODEL:PARAMETERS Distribution model by row. Sum of Ki values must be value K specified by 'rows:K' in WindowSize syntax element.

Function considers rows from most to least recent. For example, suppose that function call has these syntax elements:

WindowSize ('rows:10')

Model1 ('SEGMENT_ROWS',

'3:0.5:UNIFORM:NA',

'4:0.3:LAST_CLICK:NA',

'3:0.2:FIRST_CLICK:NA')

Attribution for a conversion event is divided among attributable events in 10 rows immediately preceding conversion event. If conversion event is in row 11, first model specification applies to rows 10, 9, and 8; second applies to rows 7, 6, 5, and 4; and third applies to rows 3, 2, and 1.

Half attribution (5/10) is uniformly divided among rows 10, 9, and 8; 3/10 to last click in rows 7, 6, 5, and 4 (that is, in row 7), and 2/10 to first click in rows 3, 2, and 1 (that is, in row 1).

SEGMENT_SECONDS Ki:WEIGHT:MODEL:PARAMETERS Distribution model by time in seconds. Sum of Kivalues must be value K specified by 'seconds:K' in WindowSize syntax element.

Function considers rows from most to least recent. For example, suppose that function call has these syntax elements:

WindowSize ('seconds:20')

Model1 ('SEGMENT_SECONDS',

'6:0.5:UNIFORM:NA',

'8:0.3:LAST_CLICK:NA',

'6:0.2:FIRST_CLICK:NA')

Attribution for a conversion event is divided among attributable events in 20 seconds immediately preceding conversion event. If conversion event is at second 21, first model specification applies to seconds 20-15 (counting backward); second applies to seconds 14-7; and third applies to seconds 6-1.

Half attribution (5/10) is uniformly divided among seconds 20-15; 3/10 to last click in seconds 14-7, and 2/10 to first click in seconds 6-1.

MODEL Values and Corresponding PARAMETER Values

MODEL values are case-sensitive. Attributable events are those whose types are not specified in excluding events table.

MODEL PARAMETERS Description
'LAST_CLICK' 'NA' Conversion event is attributed entirely to most recent attributable event.
'FIRST_CLICK' 'NA' Conversion event is attributed entirely to first attributable event.
'UNIFORM' 'NA' Conversion event is attributed uniformly to preceding attributable events.
'EXPONENTIAL' 'alpha,type' where alpha is a decay factor in range (0, 1) and type is ROW, MILLISECOND, SECOND, MINUTE, HOUR, DAY, MONTH, or YEAR.

When alpha is in range (0, 1), sum of series wi=(1-alpha)*alphai is 1. Function uses wi as exponential weights.

Conversion event is attributed exponentially to preceding attributable events (the more recent the event, the higher the attribution).
'WEIGHTED' You can specify any number of weights. If there are more attributable events than weights, extra (least recent) events are assigned zero weight. If there are more weights than attributable events, then function renormalizes weights. Conversion event is attributed to preceding attributable events with weights specified by PARAMETERS.SEGMENT_SECONDS (when you specify 'rows:K&seconds:K' in WindowSize syntax element)

Allowed FirstModelTable/SecondModelTable Combinations

FirstModelTable Type SecondModelTable Type
SIMPLE Not allowed
EVENT_REGULAR  
EVENT_REGULAR EVENT_OPTIONAL (when you specify optional events table)
SEGMENT_ROWS SEGMENT_SECONDS (when you specify 'rows:K&seconds:K' in WindowSize syntax element)
SEGMENT_ROWS  
SEGMENT_SECONDS Not allowed