SparseMult: A Sparse Tensor Decomposition Model for Knowledge Graph Link Prediction
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
ABSTRACT Knowledge graphs (KGs) have shown great power in many downstream natural language processing (NLP) tasks, such as recommendation system and question answering. Despite the large amount of knowledge facts in KGs, KGs still suffer from an issue of incompleteness, namely, lots of relations between entities are missing. Link prediction, also known as knowledge graph completion (KGC), aims to predict missing relations between entities. The models based on tensor decomposition, such as Rescal and DistMult, are promising to solve the link prediction task. However, previous Rescal model lacks the ability to scale to large KGs due to the large amount of parameters. DistMult simplifies Rescal by using diagonal matrices to represent relations, while it suffers from the limitation of dealing with antisymmetric relations. To address these problems, in this paper, we propose a SparseMult model, which is a novel tensor decomposition model based on sparse relation matrix. Specifically, we view KGs as 3D tensors and decompose them as entity vectors and relation matrices. To reduce the number of parameters in relation matrices, we represent each relation matrix as a sparse block diagonal matrix. Thus, the complexity of relation matrices grow linearly with the embedding size, making it able to scale up to large KGs. Moreover, we analyze the ability of modeling different relation patterns and show that our SparseMult is capable to model symmetry, antisymmetry, and inversion relations. We conduct extensive experiments on three widely used benchmark datasets FB15k‐237, WN18RR, and CCKS2021 KGs. Experimental results demonstrate that our SparseMult model outperforms most of the state‐of‐the‐art methods.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it