Event-based semantic search and retrieval
A technique for semantic search and retrieval that is event-based, wherein is event is composed of a sequence of observations that are user speech or physical actions. Using a first set of conversations, a machine learning model is trained against groupings of utterances therein to generate a speech act classifier. Observation sequences therein are organized into groupings of events and configured for subsequent event recognition. A set of second (unannotated) conversations are then received. The set of second conversations is evaluated using the speech act classifier and information retrieved from the event recognition to generate event-level metadata that comprises, for each utterance or physical action within an event, one or more associated tags. In response to a query, a search is performed against the metadata. Because the metadata is derived from event recognition, the search is performed against events learned from the set of first conversations. One or more conversation fragments that, from an event-based perspective, are semantically-relevant to the query, are returned.
1 . A method of semantic search and retrieval with respect to a set of first conversations that have been annotated to identify speech acts, physical acts, and events, wherein a speech act is a labeled grouping of semantically-similar utterances, wherein a physical act is a non-linguistic action taken by an actor, and wherein an event is composed of a sequence of observations that are user speech or physical actions, and wherein observation sequences are organized into groupings of events, comprising:
training a machine learning model against groupings of utterances in the set of first conversations to generate a classifier of speech acts;
using the trained machine learning model and information obtained from recognizing events from the grouping of events to generate a set of event-level data;
responsive to receipt of a query, perform a search again the event-level data; and
returning a response to the query.
2 . The method as described in claim 1 wherein the response comprises one of: a list of one or more events as identified from the set of first conversations that are associated to the query, and a conversational fragment retrieved from the set of first conversations, the conversational fragment having an information structure that is semantically-similar to the query.
3 . The method as described in claim 1 further including filtering the response according to at least one filter condition.
4 . The method as described in claim 1 further including performing one of: event pattern matching, and event classification, to generate the information.
5 . The method as described in claim 1 wherein the event-level data comprises, for a given utterance or physical action within an event, one or more tags.
6 . The method as described in claim 6 wherein the one or more tags comprise a moment of interest tag.
7 . The method as described in claim 1 wherein the query comprises one of: an utterance, and an ungrammatical collection of words.
8 . The method as described in claim 1 wherein at least some of the conversations in the set of first and second conversations have at least one or more turns, wherein a turn captures all consecutive utterances from a same conversational entity.
9 . The method as described in claim 1 wherein at least some of the conversations in the set of first and second conversations are derived from one of: a human-to-human interaction, and a human-to-conversational bot interaction.
10 . The method as described in claim 1 wherein the set of second conversations are received as a data stream in real-time or near real-time.
11 . The method as described in claim 1 wherein the set of second conversations comprise an historical corpus of conversational transcripts.
12 . The method as described in claim 1 wherein the query represents a conversational moment of interest.