Ranking- and Intent-based Response Selector

Description

Response Selector is a component that selects the final response among hypotheses provided by different skills. The Ranking- and Intent-based Response Selector utilizes floating point annotations by ranking hypotheses with a candidate annotator (e.g., Sentence Ranker), scaling the resulting scores with heuristics based on entities and intents, and finally selecting the best ranked one.

Parameters

Utilizes annotations by SENTENCE_RANKER_ANNOTATION_NAME candidate annotator. In case of absence of these annotations, utilizes provided SENTENCE_RANKER_SERVICE_URL to annotate hypotheses according to N_UTTERANCES_CONTEXT last utterances. Parameter FILTER_TOXIC_OR_BADLISTED defines whether it filers out toxic hypotheses or not.

Dependencies

• either candidate annotations by SENTENCE_RANKER_ANNOTATION_NAME or service SENTENCE_RANKER_SERVICE_URL,
• intents annotations,
• entities annotations.