Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Taxonomy-Guided Zero-Shot Recommendations with LLMs

Yueqing Liang, Liangwei Yang, Chen Wang, Xiongxiao Xu, Philip S. Yu, Kai Shu

TL;DR

This paper tackles the challenge of deploying large language models in zero‑shot recommender systems by introducing TaxRec, a taxonomy‑guided framework. TaxRec first generates an in‑domain taxonomy with a one‑time prompt, categorizes the candidate items into a structured pool, and then uses taxonomy‑aware prompts to elicit categorized recommendations from an LLM, followed by a feature‑based ranking in a zero‑shot setting. Across MovieLens and BookCrossing datasets, TaxRec demonstrates substantial improvements over strong zero‑shot baselines, with GPT‑4 yielding the largest gains, and its effectiveness arising from taxonomy‑driven information compression, structured prompting, and parsing/mapping of LLM outputs to a categorized item pool. The work highlights the importance of taxonomy‑informed prompting for aligning LLM capabilities with recommender tasks and provides practical evidence that structured domain knowledge can unlock zero‑shot performance without fine‑tuning.

Abstract

With the emergence of large language models (LLMs) and their ability to perform a variety of tasks, their application in recommender systems (RecSys) has shown promise. However, we are facing significant challenges when deploying LLMs into RecSys, such as limited prompt length, unstructured item information, and un-constrained generation of recommendations, leading to sub-optimal performance. To address these issues, we propose a novel method using a taxonomy dictionary. This method provides a systematic framework for categorizing and organizing items, improving the clarity and structure of item information. By incorporating the taxonomy dictionary into LLM prompts, we achieve efficient token utilization and controlled feature generation, leading to more accurate and contextually relevant recommendations. Our Taxonomy-guided Recommendation (TaxRec) approach features a two-step process: one-time taxonomy categorization and LLM-based recommendation, enabling zero-shot recommendations without the need for domain-specific fine-tuning. Experimental results demonstrate TaxRec significantly enhances recommendation quality compared to traditional zero-shot approaches, showcasing its efficacy as personal recommender with LLMs. Code is available at https://github.com/yueqingliang1/TaxRec.

Taxonomy-Guided Zero-Shot Recommendations with LLMs

TL;DR

This paper tackles the challenge of deploying large language models in zero‑shot recommender systems by introducing TaxRec, a taxonomy‑guided framework. TaxRec first generates an in‑domain taxonomy with a one‑time prompt, categorizes the candidate items into a structured pool, and then uses taxonomy‑aware prompts to elicit categorized recommendations from an LLM, followed by a feature‑based ranking in a zero‑shot setting. Across MovieLens and BookCrossing datasets, TaxRec demonstrates substantial improvements over strong zero‑shot baselines, with GPT‑4 yielding the largest gains, and its effectiveness arising from taxonomy‑driven information compression, structured prompting, and parsing/mapping of LLM outputs to a categorized item pool. The work highlights the importance of taxonomy‑informed prompting for aligning LLM capabilities with recommender tasks and provides practical evidence that structured domain knowledge can unlock zero‑shot performance without fine‑tuning.

Abstract

With the emergence of large language models (LLMs) and their ability to perform a variety of tasks, their application in recommender systems (RecSys) has shown promise. However, we are facing significant challenges when deploying LLMs into RecSys, such as limited prompt length, unstructured item information, and un-constrained generation of recommendations, leading to sub-optimal performance. To address these issues, we propose a novel method using a taxonomy dictionary. This method provides a systematic framework for categorizing and organizing items, improving the clarity and structure of item information. By incorporating the taxonomy dictionary into LLM prompts, we achieve efficient token utilization and controlled feature generation, leading to more accurate and contextually relevant recommendations. Our Taxonomy-guided Recommendation (TaxRec) approach features a two-step process: one-time taxonomy categorization and LLM-based recommendation, enabling zero-shot recommendations without the need for domain-specific fine-tuning. Experimental results demonstrate TaxRec significantly enhances recommendation quality compared to traditional zero-shot approaches, showcasing its efficacy as personal recommender with LLMs. Code is available at https://github.com/yueqingliang1/TaxRec.
Paper Structure (19 sections, 5 equations, 4 figures, 4 tables)

This paper contains 19 sections, 5 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM for Recommendations
  4. Zero-shot Recommendations
  5. Methodology
  6. Problem Formulation
  7. One-time Taxonomy Categorization
  8. LLM-based Recommendation
  9. Experiments
  10. Experimental Setup
  11. Datasets
  12. Baselines
  13. Evaluation Metrics
  14. Implementation Details
  15. Overall Performance (RQ1)
  16. ...and 4 more sections

Figures (4)

  • Figure 1: (a) LLMs are zero-shot recommenders without knowing other user-item interactions when acting as personal recommendation assistants. (b) Three challenges occur when integrating RecSys within the all-task-in-one LLM assistant, i.e., limited prompt length, vague item title and unconstrained generation.
  • Figure 2: The proposed TaxRec for zero-shot LLM-based recommendation. (a) One-time Taxonomy Categorization step aims to generate in-domain taxonomy and enrich/categorize item's title into structured text information. (b) LLM-based Recommendation step provides ranked item lists for users based on the user's historical interactions.
  • Figure 3: Recommendation performance by changing the number of features in taxonomy on both datasets.
  • Figure 4: Recommendation performance by changing the methods for calculating the matching score on both Movie and Book datasets.