Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

InBox: Recommendation with Knowledge Graph using Interest Box Embedding

Zezhong Xu, Yincen Qu, Wen Zhang, Lei Liang, Huajun Chen

TL;DR

This work introduces a novel embedding-based model called InBox, which significantly outperforms state-of-the-art methods like HAKG and KGIN on recommendation tasks and proposes that an interest comprises diverse basic concepts, and box intersection naturally supports concept combination.

Abstract

Knowledge graphs (KGs) have become vitally important in modern recommender systems, effectively improving performance and interpretability. Fundamentally, recommender systems aim to identify user interests based on historical interactions and recommend suitable items. However, existing works overlook two key challenges: (1) an interest corresponds to a potentially large set of related items, and (2) the lack of explicit, fine-grained exploitation of KG information and interest connectivity. This leads to an inability to reflect distinctions between entities and interests when modeling them in a single way. Additionally, the granularity of concepts in the knowledge graphs used for recommendations tends to be coarse, failing to match the fine-grained nature of user interests. This homogenization limits the precise exploitation of knowledge graph data and interest connectivity. To address these limitations, we introduce a novel embedding-based model called InBox. Specifically, various knowledge graph entities and relations are embedded as points or boxes, while user interests are modeled as boxes encompassing interaction history. Representing interests as boxes enables containing collections of item points related to that interest. We further propose that an interest comprises diverse basic concepts, and box intersection naturally supports concept combination. Across three training steps, InBox significantly outperforms state-of-the-art methods like HAKG and KGIN on recommendation tasks. Further analysis provides meaningful insights into the variable value of different KG data for recommendations. In summary, InBox advances recommender systems through box-based interest and concept modeling for sophisticated knowledge graph exploitation.

InBox: Recommendation with Knowledge Graph using Interest Box Embedding

TL;DR

This work introduces a novel embedding-based model called InBox, which significantly outperforms state-of-the-art methods like HAKG and KGIN on recommendation tasks and proposes that an interest comprises diverse basic concepts, and box intersection naturally supports concept combination.

Abstract

Knowledge graphs (KGs) have become vitally important in modern recommender systems, effectively improving performance and interpretability. Fundamentally, recommender systems aim to identify user interests based on historical interactions and recommend suitable items. However, existing works overlook two key challenges: (1) an interest corresponds to a potentially large set of related items, and (2) the lack of explicit, fine-grained exploitation of KG information and interest connectivity. This leads to an inability to reflect distinctions between entities and interests when modeling them in a single way. Additionally, the granularity of concepts in the knowledge graphs used for recommendations tends to be coarse, failing to match the fine-grained nature of user interests. This homogenization limits the precise exploitation of knowledge graph data and interest connectivity. To address these limitations, we introduce a novel embedding-based model called InBox. Specifically, various knowledge graph entities and relations are embedded as points or boxes, while user interests are modeled as boxes encompassing interaction history. Representing interests as boxes enables containing collections of item points related to that interest. We further propose that an interest comprises diverse basic concepts, and box intersection naturally supports concept combination. Across three training steps, InBox significantly outperforms state-of-the-art methods like HAKG and KGIN on recommendation tasks. Further analysis provides meaningful insights into the variable value of different KG data for recommendations. In summary, InBox advances recommender systems through box-based interest and concept modeling for sophisticated knowledge graph exploitation.
Paper Structure (22 sections, 29 equations, 5 figures, 3 tables)

This paper contains 22 sections, 29 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. PROBLEM FORMULATION
  3. METHODOLOGY
  4. Item Point and Relation, Tag Box
  5. Basic Pretraining Step
  6. Box intersection
  7. Interest Box Recommendation
  8. Model Prediction
  9. EXPERIMENTS
  10. Experimental Settings
  11. Datasets Description.
  12. Evaluation Metrics.
  13. Baselines.
  14. Parameter Settings.
  15. RQ1: Overall Performance
  16. ...and 7 more sections

Figures (5)

  • Figure 1: An example illustrating a specific interest is the combination of several concepts.
  • Figure 2: (a) An example of the knowledge-aware recommendation data. (b) An illustration of the item points and concept boxes.
  • Figure 3: Framework of the proposed model InBox: The recommendation task is completed through three training steps (not all required). The initial two steps focus solely on the KG data to obtain suitable representations for items, tags, and relations. In the third step, the objective is to leverage the user's interest box to compute the matching score, which serves as the recommendation result.
  • Figure 4: The geometric intuition of the different distances in 2-dimensional space. (A) The point-to-point distance for IRI triplets. (B) The box-to-box distance for TRT triplets. (C) The inside and outside distance for IRT triplets.
  • Figure 5: Four cases from Last-FM. The red points are the items connected to the relation-tag pair, and the blue points are randomly sampled items.