Publication Date

Spring 2022

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Engineering

Advisor

Jorieta G. Jetcheva

Subject Areas

Computer engineering

Abstract

CRS strives to return the most relevant recommendations to users through a multi-turn interactive conversation. The Covid-19 pandemic has undoubtedly accelerated the pace of adoption of conversational AI systems by many online platforms needing to provide highly available customer support. This ongoing demand calls for the need to implement more generic and efficient conversational agents and recommendation engines that can provide customers with the required information at every stage of the user’s interaction with the e-commerce platform. This study presents a Smart Conversational AI-based recommendation system - SMARTREC, which enables a multi-turn conversation with the user to understand the context and semantics behind their product requirements and generate appropriate recommendations in real-time. Several ongoing studies investigate how to evolve CRS, there are many open research problems in this space. Current CRS suffers from four major issues. First, a lack of proper contextual understanding of the user intention; second, inaccurate semantic mapping of user preferences in natural language to the interested item attributes; third, rely only on current conversation and suffer from data sparsity; fourth, trained on open-domain crowd-sourced conversational data preventing the system from learning the user intentions accurately. This study implements a novel real-time CRS by curating large-scale domain data, further combining them with a common-sense semantic network to build an intelligent domain knowledge graph. Finally, this study conducted extensive experiments to demonstrate the efficiency of SMARTREC in yielding better performance as a CRS.

Download

Share

COinS