Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Wizard of Shopping: Target-Oriented E-commerce Dialogue Generation with Decision Tree Branching

Xiangci Li, Zhiyu Chen, Jason Ingyu Choi, Nikhita Vedula, Besnik Fetahu, Oleg Rokhlenko, Shervin Malmasi

TL;DR

This work tackles data scarcity in conversational product search by introducing TRACER, a framework that grounds dialogue generation in product plans via a decision-tree search. It generates Wizard of Shopping (WoS), a large-scale, natural, target-oriented CPS dataset across three domains, validated by human evaluations and downstream experiments. TRACER combines customer preference sampling, tree-guided dialogue planning, and verbalization (interactive or single-pass) to produce coherent conversations and realistic search trajectories that minimize user effort. The WoS dataset demonstrates clear benefits for downstream tasks in Conversational Query Generation and Conversational Product Ranking, and the work provides extensive analysis and comparisons with prior datasets, highlighting its significance for CPS research and application.

Abstract

The goal of conversational product search (CPS) is to develop an intelligent, chat-based shopping assistant that can directly interact with customers to understand shopping intents, ask clarification questions, and find relevant products. However, training such assistants is hindered mainly due to the lack of reliable and large-scale datasets. Prior human-annotated CPS datasets are extremely small in size and lack integration with real-world product search systems. We propose a novel approach, TRACER, which leverages large language models (LLMs) to generate realistic and natural conversations for different shopping domains. TRACER's novelty lies in grounding the generation to dialogue plans, which are product search trajectories predicted from a decision tree model, that guarantees relevant product discovery in the shortest number of search conditions. We also release the first target-oriented CPS dataset Wizard of Shopping (WoS), containing highly natural and coherent conversations (3.6k) from three shopping domains. Finally, we demonstrate the quality and effectiveness of WoS via human evaluations and downstream tasks.

Wizard of Shopping: Target-Oriented E-commerce Dialogue Generation with Decision Tree Branching

TL;DR

This work tackles data scarcity in conversational product search by introducing TRACER, a framework that grounds dialogue generation in product plans via a decision-tree search. It generates Wizard of Shopping (WoS), a large-scale, natural, target-oriented CPS dataset across three domains, validated by human evaluations and downstream experiments. TRACER combines customer preference sampling, tree-guided dialogue planning, and verbalization (interactive or single-pass) to produce coherent conversations and realistic search trajectories that minimize user effort. The WoS dataset demonstrates clear benefits for downstream tasks in Conversational Query Generation and Conversational Product Ranking, and the work provides extensive analysis and comparisons with prior datasets, highlighting its significance for CPS research and application.

Abstract

The goal of conversational product search (CPS) is to develop an intelligent, chat-based shopping assistant that can directly interact with customers to understand shopping intents, ask clarification questions, and find relevant products. However, training such assistants is hindered mainly due to the lack of reliable and large-scale datasets. Prior human-annotated CPS datasets are extremely small in size and lack integration with real-world product search systems. We propose a novel approach, TRACER, which leverages large language models (LLMs) to generate realistic and natural conversations for different shopping domains. TRACER's novelty lies in grounding the generation to dialogue plans, which are product search trajectories predicted from a decision tree model, that guarantees relevant product discovery in the shortest number of search conditions. We also release the first target-oriented CPS dataset Wizard of Shopping (WoS), containing highly natural and coherent conversations (3.6k) from three shopping domains. Finally, we demonstrate the quality and effectiveness of WoS via human evaluations and downstream tasks.

Paper Structure

This paper contains 77 sections, 2 equations, 1 figure, 21 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Conversational Product Search.
  4. Difference with Prior Work in Conversational Recommendation & Search.
  5. Dialogue Generation Approach
  6. Customer Preference Sampling
  7. Dialogue Planning with Decision Tree
  8. Verbalization
  9. Interactive Generation.
  10. Single-Pass Generation.
  11. Enhancing Conversational Naturalness
  12. How conversation starts.
  13. How conversation ends.
  14. Improving naturalness by adding hints.
  15. Improving naturalness by LLM knowledge.
  16. ...and 62 more sections

Figures (1)

  • Figure 1: An example of generated shopping conversation using TRACER, which leverages the LLM with different roles and decision tree-based planning.