Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GOT4Rec: Graph of Thoughts for Sequential Recommendation

Zewen Long, Liang Wang, Shu Wu, Qiang Liu, Liang Wang

TL;DR

GOT4Rec tackles the challenge of leveraging external knowledge in sequential recommendation by introducing a graph of thoughts (GoT) that decomposes user reasoning into short-term, long-term, and collaborative components. It enables LLMs to generate and aggregate multiple thoughts through generation and aggregation transformations, guided by a multi-level retrieval module that uses embedding-based retrieval to refine candidate items. The approach achieves state-of-the-art results across three real-world datasets, with notable gains on food-related items due to effective short-term preference capture, and demonstrates improved novelty by reducing popularity bias. Overall, GOT4Rec provides a scalable, interpretable framework that combines rich user-sequence information with LLM reasoning to enhance recommendation quality and explainability.

Abstract

With their vast open-world knowledge and reasoning abilities, large language models (LLMs) have become a promising tool for sequential recommendation. Researchers have explored various methods to harness these capabilities, but most existing approaches rely on simple input-output prompting, failing to effectively bridge the gap between LLMs' general knowledge and the specific needs of recommendation tasks. While reasoning strategies like chain-of-thought (CoT) have been introduced to enhance performance, they often produce inaccurate recommendations due to underutilized user preference information and insufficient reasoning depth. To address these challenges, we propose GOT4Rec, a novel sequential recommendation method leveraging the graph of thoughts (GoT) reasoning strategy. Our method focuses on three key types of information in user histories: short-term interests, long-term interests and collaborative information from other users. It enables LLMs to reason independently and generate recommendations, subsequently aggregating results to derive final items. This method allows LLMs, with enhanced reasoning capabilities, to better utilize the user sequence information, producing more accurate recommendations and comprehensive explanations. Extensive experiments on real-world datasets demonstrate the effectiveness of GOT4Rec, outperforming existing state-of-the-art baselines with an average improvement of 37.11%. Our code is available at https://anonymous.4open.science/r/GOT4Rec.

GOT4Rec: Graph of Thoughts for Sequential Recommendation

TL;DR

GOT4Rec tackles the challenge of leveraging external knowledge in sequential recommendation by introducing a graph of thoughts (GoT) that decomposes user reasoning into short-term, long-term, and collaborative components. It enables LLMs to generate and aggregate multiple thoughts through generation and aggregation transformations, guided by a multi-level retrieval module that uses embedding-based retrieval to refine candidate items. The approach achieves state-of-the-art results across three real-world datasets, with notable gains on food-related items due to effective short-term preference capture, and demonstrates improved novelty by reducing popularity bias. Overall, GOT4Rec provides a scalable, interpretable framework that combines rich user-sequence information with LLM reasoning to enhance recommendation quality and explainability.

Abstract

With their vast open-world knowledge and reasoning abilities, large language models (LLMs) have become a promising tool for sequential recommendation. Researchers have explored various methods to harness these capabilities, but most existing approaches rely on simple input-output prompting, failing to effectively bridge the gap between LLMs' general knowledge and the specific needs of recommendation tasks. While reasoning strategies like chain-of-thought (CoT) have been introduced to enhance performance, they often produce inaccurate recommendations due to underutilized user preference information and insufficient reasoning depth. To address these challenges, we propose GOT4Rec, a novel sequential recommendation method leveraging the graph of thoughts (GoT) reasoning strategy. Our method focuses on three key types of information in user histories: short-term interests, long-term interests and collaborative information from other users. It enables LLMs to reason independently and generate recommendations, subsequently aggregating results to derive final items. This method allows LLMs, with enhanced reasoning capabilities, to better utilize the user sequence information, producing more accurate recommendations and comprehensive explanations. Extensive experiments on real-world datasets demonstrate the effectiveness of GOT4Rec, outperforming existing state-of-the-art baselines with an average improvement of 37.11%. Our code is available at https://anonymous.4open.science/r/GOT4Rec.

Paper Structure

This paper contains 29 sections, 2 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sequential Recommenders
  4. LLM Reasoning Strategies
  5. The Proposed GOT4Rec Method
  6. Overview of Thought Graph
  7. Short-Term & Long-Term Reasoning Process
  8. Collaborative Reasoning Process
  9. Multi-Level Retrieval Module
  10. Sequence-level Retrieval.
  11. Item-level Retrieval.
  12. Inference Latency and Volume Analyzation
  13. Experiments
  14. Experimental Settings
  15. Datasets.
  16. ...and 14 more sections

Figures (4)

  • Figure 1: Comparison of the LLM output and predictions for the next item generated by SLIM and our proposed GOT4Rec.
  • Figure 2: An example of graph decomposition in the proposed GOT4Rec method. The user interaction sequence is divided to facilitate different types of reasoning: the last few items are used for short-term preference reasoning, while the entire sequence informs long-term preference reasoning. Additionally, sequences from similar users are retrieved for collaborative reasoning. In the thought graph, these thoughts are generated and subsequently aggregated to capture and integrate the various aspects of information in the user interaction sequence.
  • Figure 3: Prompt templates, including summarizing, recommendation, collaboration and voting prompts.
  • Figure 4: Analysis of popularity bias, items in the dataset are sorted by their frequency. Compared to CoT, GOT4Rec demonstrates a more consistent ability to recommend long-tail items.