Resolving Abstract Anaphoric References in Conversational Systems Using Hierarchically Stacked Neural Networks

摘要

#$%^&*AU2019204953A120200130.pdf#####ABSTRACT RESOLVING ABSTRACT ANAPHORIC REFERENCES IN CONVERSATIONAL SYSTEMS USING HIERARCHICALLY STACKED NEURAL NETWORKS Conversational systems are required to be capable of handling more sophisticated interactions than providing factual answers only. Such interactions are handled by resolving abstract anaphoric references in conversational systems which includes antecedent fact references and posterior fact references. The present disclosure resolves abstract anaphoric references in conversational systems using hierarchically stacked neural networks. In the present disclosure, a deep hierarchical maxpool network based model is used to obtain a representation of each utterance received from users and a representation of one or more generated sequences of utterances. The obtained representations are further used to identify contextual dependencies with in the one or more generated sequences which helps in resolving abstract anaphoric references in conversational systems. Further, a response for an incoming sequence of utterances is retrieved based on classification of incoming sequence of utterances into one or more pre-created responses. The proposed model takes lesser time to retrain. [To be published with FIG. 2]rol 200 Receiving, in a multi-turn retrieval chat-bot, a plurality of consecutive utterances comprising of at least a sub- set of 202 utterances indicative of anaphoric reference to specific entities or facts comprised in past utterances, in a multi-turn retrieval chat-bot Generating, (i) one or more sequences of the plurality of consecutive utterances and (ii) obtaining one or more pre- 204 created corresponding responses from the database for each of the one or more generated sequences Training, a Deep Hierarchical Maxpool Network (DHMN) based model, using the one or more generated sequences, and the plurality of consecutive utterances, to (a) obtain a representation for each of (i) the one or more generated 206 sequences, and (ii) the plurality of consecutive utterances, (b) identify contextual dependencies within the one or more generated sequences using each representation to resolve anaphoric references Updating, using a Character to Word Encoder (CWE) network comprised in the DHMN based model, the representation of 208 the plurality of consecutive utterances based on a presence or an absence of discrepancies in one or more utterances comprised in an incoming sequence of utterances Classifying, using the trained DHMN based model and the identified contextual dependencies, the incoming sequence 210 of utterances, based on at least one of (i) the updated o representation, and (ii) the representation of the one or more generated sequences, into at least one of the one or more precreated corresponding answers FIG. 2