Research map / Label-efficient and robust learning / Recognition and long-tail
P-SIF: Document Embeddings Using Partition Averaging
Vivek Gupta, Ankit Saw, Pegah Nokhiz, Praneeth Netrapalli, Piyush Rai, Partha Talukdar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add topic partitioning and concatenation to SIF weighted averaging | SIF document embeddings · 20NewsGroup | accuracy: 82.3 -> 85.4 | Improved |
| Add topic partitioning and concatenation to SIF weighted averaging | SIF document embeddings · Reuters-21578 | F1: 76.78 -> 82.41 | Improved |
| Initialize word vectors with Doc2VecC (ablation) | P-SIF · 20NewsGroup | accuracy: 85.4 -> 86.0 | Improved |
| Initialize word vectors with Doc2VecC (ablation) | P-SIF · Reuters-21578 | F1: 82.41 -> 82.87 | Improved |
| Use GMM overlapping clustering instead of sparse dictionary learning (ablation) | P-SIF · Some STS tasks with short, single-sentence texts | Pearson's r | Improved |
| Use sparse dictionary learning instead of GMM overlapping clustering (ablation) | P-SIF · 20NewsGroup and Reuters classification | F1: ~1.5% | Improved |
| Weight each word embedding by its learned dictionary coefficient (ablation) | P-SIF word-topic vectors · Document embedding tasks | performance | Improved |
| Increase word-vector dimensions from 200 to 500 (ablation) | SIF and P-SIF · 20NewsGroup | accuracy: <0.4% | No clear effect |
| Increase word-vector dimensions beyond 500 (ablation) | SIF and P-SIF · Document classification | performance | No clear effect |
| Use 20 partitions with Doc2VecC-initialized vectors instead of 40 (ablation) | P-SIF · 20NewsGroup | accuracy: 40 partitions; 20 partitions | Improved |
Closest papers by meaning
- Active Discriminative Text Representation Learning (AAAI 2017 · similarity 0.87)
- A Simple but Tough-to-Beat Baseline for Sentence Embeddings (ICLR 2017 · similarity 0.87)
- Representing Mixtures of Word Embeddings with Mixtures of Topic Embeddings (ICLR 2022 · similarity 0.86)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.86)
- Further results on latent discourse models and word embeddings (JMLR 2021 · similarity 0.86)
- Contextual Document Embeddings (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
