Research map / Label-efficient and robust learning / Recognition and long-tail
Short Text Representation for Detecting Churn in Microblogs
Hadi Amiri, Hal Daume III
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 neighboring-word features to unigrams (ablation) | BOW churn classifier (hinge) · Churn detection, macro-avg | F1: 65.30 -> 73.63 | Improved |
| add dependency-path features to unigrams (ablation) | BOW churn classifier (hinge) · Churn detection, macro-avg | F1: 65.30 -> 72.40 | Improved |
| use RNN vector for dependency path instead of unigram+Dep | churn classifier (hinge) · Churn detection, macro-avg | F1: 72.40 -> 75.66 | Improved |
| add context (thread) indicators to unigram+Nb (ablation) | BOW churn classifier (hinge) · Churn detection, macro-avg | F1: 73.63 -> 74.27 | Improved |
| combine Nb, Dep and Cntx features (ablation) | BOW churn classifier (hinge) · Churn detection, macro-avg | F1: 73.63 -> 77.03 | Improved |
| combine all RNN features (twRep+NbRep+DepRep+CntxRep) over twRep+NbRep (ablation) | RNN representation (logistic) · Churn detection, macro-avg | F1: 73.90 -> 77.56 | Improved |
| combine BOW and RNN representations (ablation) | churn classifier (hinge) · Churn detection, macro-avg | F1: 77.03 -> 78.30 | Improved |
| concatenate tRep and wRep (twRep) instead of unigrams (ablation) | tweet representation · Churn detection, macro-avg | F1: 65.30 -> 66.13 | Improved |
| tRep alone instead of unigrams (128 dims) (ablation) | tweet representation · Churn detection, macro-avg | F1 | No clear effect |
| train RNN on balanced instead of imbalanced development data (ablation) | twRep+NbRep (logistic) · Churn detection, macro-avg | F1: 75.89 vs 73.90 (BOW-comparison cell); balanced twRep 68.27 | Mixed |
| use sentiment classifiers instead of churn-specific features | churn classification · Churn detection, Verizon | F1: 40.06 -> 65.17 (TargetSnt vs Unigram) | Hurt |
| concatenate neighboring-word embeddings instead of averaging (ablation) | NbRep · Churn detection | F1 | Improved |
Closest papers by meaning
- TopicRNN: A Recurrent Neural Network with Long-Range Semantic Dependency (ICLR 2017 · similarity 0.81)
- Causal Customer Churn Analysis with Low-rank Tensor Block Hazard Model (ICML 2024 · similarity 0.81)
- Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamics (NeurIPS 2019 · similarity 0.81)
- The geometry of integration in text classification RNNs (ICLR 2021 · similarity 0.81)
- How recurrent networks implement contextual processing in sentiment analysis (ICML 2020 · similarity 0.81)
- SMTPD: A New Benchmark for Temporal Prediction of Social Media Popularity (CVPR 2025 · similarity 0.79)
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.
