Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Excerpt

This recognizer uses a frozen Universal Sentence Encoder TensorFlow model to encode, using sentence embedding vectors, a list of Frequently Asked Questions and tag sentences that match a question/answer pair given a specified threshold of accuracy with the question/answer from the FAQ.

The recognizer also gives the option to use a python model instead of TensorFlow. You'll need the Python Bridge running for this.

Note

This is a plugin recognizer. Uses FAQ Stage

Before adding the FAQ Recognizer, download the tensorflow model from here and unzip it under %SAGA_SERVER%\tf-models

Configuration

Configuration using TensorFlow:

Image RemovedImage Added

  • Parameter
    summaryThe minimum similarity match between the vectors of a sentence and a question/answer to consider the question to tag the input sentence.
    default0.8
    nameMatch threshold
    typedouble
    requiredtrue
  • Parameter
    summaryWhether or not to include the answer vectors when evaluating an input sentence.
    defaultfalse
    nameEval Answer
    typeboolean
    requiredtrue
  • Parameter
    summaryWhether or not to include the vector generated in the metadata.
    defaultfalse
    nameAdd Vectors
    typeboolean
    requiredtrue


Configuration using Python:

  • Parameter
    summaryPython Bridge host name
    defaultlocalhost
    nameHost Name
    requiredtrue
  • Parameter
    summaryPython Bridge host port
    default5000
    namePort
    typeinteger
    requiredtrue
  • Parameter
    summaryPython model to be used by the recognizer
    defaultbert-base-uncased
    nameSelect Model
    requiredtrue
  • Parameter
    summaryThe minimum similarity match between the vectors of a sentence and a question/answer to consider the question to tag the input sentence.
    default0.8
    nameMatch threshold
    typedouble
    requiredtrue
  • Parameter
    summaryVersion of the python model to use
    default1
    nameSelect Version
    typeinteger
    requiredtrue


Adding a Question/Answer pair

Click on the Image Removedthe Image Added button which will popup the "Add new FAQ" dialog

Image RemovedImage Added


  • Parameter
    summaryQuestion to add to the FAQ
    nameWrite question here
    requiredtrue
  • Parameter
    summaryAnswer to the question
    nameWrite answer here
    requiredtrue
  • Parameter
    summaryIf provided, the source of the question/answer. (Usually a link to the HTML FAQ page).
    nameUrl
  • Parameter
    summaryConfidence Adjustment for this entry
    default1
    nameConfidence Adjustment
    typedouble
    requiredtrue

    • 0.0 to < 1.0  decreases confidence value
    • 1.0 confidence value remains the same
    • > 1.0 to  2.0 increases confidence value

Image RemovedImage Added


Load FAQ for evaluation

Clicking on the Image Removed button the Image Added button will encode all question/answer pairs for evaluations.

When Image Removedicon shows in orangeWhen theImage Added icon shows in purple, it means that the vectors have been loaded and its ready to use for matches.

Click onImage RemovedImage Addedicon to copy to the clipboard the sentence embedding vector of the question (top) or answer (bottom) for debugging purposes.


General Settings

Include Page
Generic Recognizer Config
Generic Recognizer Config