Runs - Incident Streams 2021

l3i-ttxth

Participants | Proceedings | Appendix

• Run ID: l3i-ttxth
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 9/28/2021
• Type: auto
• Task: 2021b
• MD5: 5acd45d015e9b67e71dad0b384ca9a2e
• Run description: INFO-PRIORITY-roBERTa-base+2xTransformer+event_type+event_title, 256, hashtags

---

l3i-ttxth.combined

Participants | Proceedings | Appendix

• Run ID: l3i-ttxth.combined
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 9/28/2021
• Type: auto
• Task: 2021b
• MD5: 811fb0199dd0d96abd95006badd05fd2
• Run description: INFO-roBERTa-base+2xTransformer+event_type+event_title, 256, hashtags, PRIORITY-roBERTa-base+2xTransformer+event_type+event_title, 280

---

njit-debly

Participants | Proceedings | Appendix

• Run ID: njit-debly
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 10/15/2021
• Type: auto
• Task: 2021b
• MD5: 35eb33755e5310bafb1daba79567b101
• Run description: UCD ensemble run with different pipelines for information class classification and priority scoring and augmentation with AugLy.

---

njit-EDA

Participants | Proceedings | Appendix

• Run ID: njit-EDA
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 7/23/2021
• Type: auto
• Task: 2021b
• MD5: db534077498f6dc192f27197acccbbcf
• Run description: NJIT RoBERTa run with Easy Data Augmentation (EDA) in place of synonym augmentation.

---

njit-label.prop

Participants | Proceedings | Appendix

• Run ID: njit-label.prop
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 7/1/2021
• Type: auto
• Task: 2021b
• MD5: 7cd1f18d685cce468e3b0ea4a3569ba1
• Run description: Model uses GPT2 to generate tweets and label propagation to label them, using which we train a new transformer model.

---

njit-semi.sup

Participants | Proceedings | Appendix

• Run ID: njit-semi.sup
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 6/27/2021
• Type: auto
• Task: 2021b
• MD5: f7d0c382a635edc6fbb984c487547421
• Run description: Model uses GPT2 to generate tweets and semi-supervision to label them, using which we train a new transformer model.

---

njit-semi_sup_cat2prior

Participants | Proceedings | Appendix

• Run ID: njit-semi_sup_cat2prior
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 7/6/2021
• Type: auto
• Task: 2021b
• MD5: a8ae7ae6af4c4349d2b9ad317049e907
• Run description: Semi-supervised generation pipeline with priority scores using highest average score of information type labels.

---

njit.augly.v2

Participants | Proceedings | Appendix

• Run ID: njit.augly.v2
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 7/19/2021
• Type: auto
• Task: 2021b
• MD5: e9b78ee38da18dd8ad911a628226a2fd
• Run description: NJIT RoBERTa run with AugLy in place of synonym augmentation.

---

njit.deberta

Participants | Proceedings | Appendix

• Run ID: njit.deberta
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 8/2/2021
• Type: auto
• Task: 2021b
• MD5: cc3e81f2229fed90449e377633fdb296
• Run description: NJIT RoBERTa run with DeBERTa instead of RoBERTa.

---

njit.label.prop.cat2prior

Participants | Proceedings | Appendix

• Run ID: njit.label.prop.cat2prior
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 7/8/2021
• Type: auto
• Task: 2021b
• MD5: 8494638c02ee775455b9443e8f47a395
• Run description: Label Propagation generation pipeline with priority scores using highest average score of information type labels.

---

njit.roberta

Participants | Proceedings | Appendix

• Run ID: njit.roberta
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 6/23/2021
• Type: auto
• Task: 2021b
• MD5: 3173af803c5a97f11f0d8ce9557df07b
• Run description: Model that uses a simple text augmentation strategy for expanding training data. Then, we use a pre-trained RoBERTa model to generate text embeddings of tweets and classify them.

---

njit_bert

Participants | Proceedings | Appendix

• Run ID: njit_bert
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 5/4/2021
• Type: auto
• Task: 2021a
• MD5: a8801f839108f184ca1dc18e13957f7b
• Run description: This method uses embeddings generated from BERT and a simple classification strategy to learn information types and priorities for a given social media message.

---

njit_roberta

Participants | Proceedings | Appendix

• Run ID: njit_roberta
• Participant: njit
• Track: Incident Streams
• Year: 2021
• Submission: 5/4/2021
• Type: auto
• Task: 2021a
• MD5: 1d2fa82c8282f7bc9c7bbc9c9d0d451b
• Run description: This method uses embeddings generated from RoBERTa and a simple classification strategy to learn information types and priorities for a given social media message.

---

RB_2T_MT_H_280

Participants | Proceedings | Appendix

• Run ID: RB_2T_MT_H_280
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 5/4/2021
• Type: auto
• Task: 2021a
• MD5: 6604acc15adf03a22fb0cef1244a2513
• Run description: INFO-roBERTa-base+2xTransformer+MultitaskEvent_type, 280, hashtags

---

RB_2T_TT_280_SVM

Participants | Proceedings | Appendix

• Run ID: RB_2T_TT_280_SVM
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: 5ed12eb1115728191261368f99dd9faa
• Run description: PRIORITY-SVM-mixed INFO-roBERTa-base+2xTransformer+event_type+event_title, 280, combined

---

RB_2T_TTH_256_LR5

Participants | Proceedings | Appendix

• Run ID: RB_2T_TTH_256_LR5
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: 9465916a142b81bf541beb33ff91d1e3
• Run description: PRIORITY-LR5 INFO-roBERTa-base+2xTransformer+event_type+event_title, 256, hashtags, combined

---

RB_2Tx2_TTH_280

Participants | Proceedings | Appendix

• Run ID: RB_2Tx2_TTH_280
• Participant: L3i_Rochelle
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: 41344b81ec73e7f2cb117f2ec96ee91a
• Run description: INFO-roBERTa-base+2xTransformer, 280 PRIORITY-roBERTa-base+2xTransformer+event_type+event_title, 280, hashtags, combined

---

Siena2021A

Participants | Proceedings | Appendix

• Run ID: Siena2021A
• Participant: SienaCLTeam
• Track: Incident Streams
• Year: 2021
• Submission: 5/5/2021
• Type: auto
• Task: 2021a
• MD5: 65ab86765727467d4dce48fc51372aa4
• Run description: We used ktrain package to train our system

---

ucdcs-mtl.ens

Participants | Proceedings | Appendix

• Run ID: ucdcs-mtl.ens
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 6/29/2021
• Type: auto
• Task: 2021b
• MD5: 7f62ba10d0ec8cf8f7ea9a3eb4c334a5
• Run description: The ensemble run of multi-task learning with Deberta and eda augmentation, i.e., ucdcs-run4 at 2021a (alignment issue fixed)

---

ucdcs-mtl.ens.fta

Participants | Proceedings | Appendix

• Run ID: ucdcs-mtl.ens.fta
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 10/18/2021
• Type: auto
• Task: 2021b
• MD5: 90a1f79f1890e82791da2639865c12f9
• Run description: UCD ensemble run with extra steps plus few-shot augmentation and noise label annealing

---

ucdcs-mtl.ens.new

Participants | Proceedings | Appendix

• Run ID: ucdcs-mtl.ens.new
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 10/18/2021
• Type: auto
• Task: 2021b
• MD5: 4ef2964940f9d05c4916152149324bd9
• Run description: UCD ensemble run with extra steps

---

ucdcs-mtl.fta

Participants | Proceedings | Appendix

• Run ID: ucdcs-mtl.fta
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 10/14/2021
• Type: auto
• Task: 2021b
• MD5: 0043ea87b8d29c35ffc4bb7b24eda407
• Run description: The run of deberta-base multi-task learning with few-shot augmentations

---

ucdcs-mtl.fta.nla

Participants | Proceedings | Appendix

• Run ID: ucdcs-mtl.fta.nla
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 10/18/2021
• Type: auto
• Task: 2021b
• MD5: 564d5a366c92c5973e56e0b859ee81b0
• Run description: The run of deberta-base multi-task learning with few-shot augmentations plus noise label annealing

---

ucdcs-run1

Participants | Proceedings | Appendix

• Run ID: ucdcs-run1
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: 977891a36db9f07a83c5a96f91e7a949
• Run description: Multi task training using deberta-base without eda augmentation.

---

ucdcs-run2

Participants | Proceedings | Appendix

• Run ID: ucdcs-run2
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: a120ab8bff0b59662bd34037b0a38f6e
• Run description: Multi task training using deberta-base with eda augmentation.

---

ucdcs-run3

Participants | Proceedings | Appendix

• Run ID: ucdcs-run3
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: f1619ce03e1421a3d34337d3eec4892a
• Run description: multi-task learning using deberta-large without eda augmentation

---

ucdcs-run4

Participants | Proceedings | Appendix

• Run ID: ucdcs-run4
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 5/3/2021
• Type: auto
• Task: 2021a
• MD5: d3c7ba1ac09e44fd7647ec2a5840a98c
• Run description: multi-task learning ensembles of run1, 2, and 3

---

ucdcs-strans.nb

Participants | Proceedings | Appendix

• Run ID: ucdcs-strans.nb
• Participant: UCD-CS
• Track: Incident Streams
• Year: 2021
• Submission: 10/10/2021
• Type: auto
• Task: 2021b
• MD5: 9ab1707ff96ec51d46d92d9237659a65
• Run description: The run of using sentence transformers as the fixed features and use GaussianNB as the downstream classifier

---

uogTr-01-pw

Participants | Proceedings | Appendix

• Run ID: uogTr-01-pw
• Participant: uog_trec_team
• Track: Incident Streams
• Year: 2021
• Submission: 5/5/2021
• Type: auto
• Task: 2021a
• MD5: 30853bf778e20438f1a8bf8a20dfc995
• Run description: 25 binary classifier + regression head on BERT base, using weighting schemes for balancing, each task using optimal tuned parameters from pool.

---

uogTr-02-pwcoocc

Participants | Proceedings | Appendix

• Run ID: uogTr-02-pwcoocc
• Participant: uog_trec_team
• Track: Incident Streams
• Year: 2021
• Submission: 5/5/2021
• Type: auto
• Task: 2021a
• MD5: 20d798cfe65d837002ad28d861fc4c1e
• Run description: 25 binary classifier + regression head on BERT base, using weighting schemes for balancing, each task using optimal tuned parameters from pool. Actionable tasks utilise transfer learning techniques based on label co-occurrence to alleviate data scarcity.

---

uogTr-04-coocc

Participants | Proceedings | Appendix

• Run ID: uogTr-04-coocc
• Participant: uog_trec_team
• Track: Incident Streams
• Year: 2021
• Submission: 5/5/2021
• Type: auto
• Task: 2021a
• MD5: 5d3b68a0c927266e58250d4a401c9e77
• Run description: 25 binary classifier + regression head on BERT base, using weighting schemes for balancing, each task using optimal tuned parameters from pool. All tasks utilise transfer learning techniques based on label co-occurrence to alleviate data scarcity.

---