Mix Named Inputs

Combines results from several inputs into a single output. Multiple merging/aggregation-techniques can be selected using the parameters. Most common scenario: if an object occurs in multiple inputs, it is an indication that this object is more relevant.

Compared to the normal Mix block, this one allows to define the input coefficient by name.

Input COEFF expects a list of scored input names, for example:

0.9 "productname"
0.1 "category"

Block parameter Input Mapping is a 2-value TUPLE_LIST, which maps these coefficients to the corresponding numbered inputs:

Score        Value 1        Value 2
-----------------------------------
    1              1    productname
    1              2       category

Where Value 1 corresponds to the input identifier of the Mix block and Value 2 corresponds to the one of the names in Input COEFF.

The final coefficients correspond to:

0.9 (Input 1)
0.1 (Input 2)

Note that the Score column of parameter Input Mapping is ignored.

• MIXLIST [OBJ]: one or more inputs that should be combined
• COEFF [STRING]: list of named inputs with coefficients

• RESULT [OBJ]: the combined output of all inputs in a single column

• Input Mapping: table mapping input names to input numeric identifiers

• Input Score normalisation

  • None: don’t change probabilities as they come in
  • Max: divide by the maximum score in each input
  • Standard: performs standard score normalisation "combSUM score combination" (J.H.Lee, 1997)

• Score combination

  • Disjunctive: objects must occur in at least 1 input (default)
  • Conjunctive : objects must occur in all inputs

• Score aggregation: defines how the output scores are calculated (when an instance occurs in multiple inputs)

  • SUM: sums the weights of the individual inputs. Use SUM if the signals are completely independent (default)
  • MAX: selects the highest weight of the object. Often used in combination with Conjunctive.
  • MIN: selects the lowest weight of the object
  • PROD: multiplies of the individual inputs. Often used in combination with Conjunctive.
  • AVG: selects the average weight of the object (only over inputs that contain the input)
  • DISTINCT: aggregates evidence (formula: 1 - prod(1 - p))
  • ALL: don't aggregate, and output all tuples from all inputs as is