ajayvikram commited on
Commit
b09e2a1
·
verified ·
1 Parent(s): 7942b30

Apply explicitly selected dataset corrections

Browse files
README.md CHANGED
@@ -141,6 +141,17 @@ configs:
141
  - split: test
142
  path:
143
  - data/test/DuRecDial-2.0.parquet
 
 
 
 
 
 
 
 
 
 
 
144
  - config_name: EmpatheticDialogues
145
  data_files:
146
  - split: train
@@ -668,7 +679,7 @@ Roles are already oriented for user-simulator training:
668
  <!-- available-datasets:start -->
669
  ## Available Datasets
670
 
671
- **Dataset count: 56**
672
 
673
  Select an individual dataset by its name below, or use `default` for the combined collection.
674
 
@@ -690,7 +701,7 @@ Datasets are grouped by behavioral category:
690
  "StudyChat"
691
  ],
692
  "Emotional & Psychological Support": [
693
- "EmpatheticDialogues"
694
  ],
695
  "General Conversation & Social Interaction": [
696
  "DailyDialog", "Humanual-Chat", "LMSYS-Chat-1M", "MSC", "Topical-Chat", "chitchat-dataset"
 
141
  - split: test
142
  path:
143
  - data/test/DuRecDial-2.0.parquet
144
+ - config_name: ESConv
145
+ data_files:
146
+ - split: train
147
+ path:
148
+ - data/train/ESConv.parquet
149
+ - split: validation
150
+ path:
151
+ - data/validation/ESConv.parquet
152
+ - split: test
153
+ path:
154
+ - data/test/ESConv.parquet
155
  - config_name: EmpatheticDialogues
156
  data_files:
157
  - split: train
 
679
  <!-- available-datasets:start -->
680
  ## Available Datasets
681
 
682
+ **Dataset count: 57**
683
 
684
  Select an individual dataset by its name below, or use `default` for the combined collection.
685
 
 
701
  "StudyChat"
702
  ],
703
  "Emotional & Psychological Support": [
704
+ "ESConv", "EmpatheticDialogues"
705
  ],
706
  "General Conversation & Social Interaction": [
707
  "DailyDialog", "Humanual-Chat", "LMSYS-Chat-1M", "MSC", "Topical-Chat", "chitchat-dataset"
SOURCES.md CHANGED
@@ -549,3 +549,11 @@ Removed 121 previously published records from individually reviewed complete sou
549
  - Deterministic episode-level splits: 49/3/3 episodes and 3,286/164/182 records for train/validation/test. All exchanges from an episode stay together; this is not an interviewer-disjoint evaluation. Supplied context timing and paraphrase fidelity have not been independently verified against the original recordings.
550
  - Source/catalog-level overlap assessment only, as requested; no exhaustive comparison with existing collection records or semantic deduplication claim. Sampled review found some abbreviated references and incomplete-looking answers, which remain as supplied under the approved preservation policy.
551
  - No explicit redistribution license is recorded in the supplied snapshot. No blanket or inferred license is assigned; original source terms remain applicable.
 
 
 
 
 
 
 
 
 
549
  - Deterministic episode-level splits: 49/3/3 episodes and 3,286/164/182 records for train/validation/test. All exchanges from an episode stay together; this is not an interviewer-disjoint evaluation. Supplied context timing and paraphrase fidelity have not been independently verified against the original recordings.
550
  - Source/catalog-level overlap assessment only, as requested; no exhaustive comparison with existing collection records or semantic deduplication claim. Sampled review found some abbreviated references and incomplete-looking answers, which remain as supplied under the approved preservation policy.
551
  - No explicit redistribution license is recorded in the supplied snapshot. No blanket or inferred license is assigned; original source terms remain applicable.
552
+
553
+ ## ESConv
554
+
555
+ Original [ESConv](https://github.com/thu-coai/Emotional-Support-Conversation/tree/f262d062ad74cb39b17ea476facc81568ddcba24), Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang and Minlie Huang; CoAI Group, Tsinghua University; Towards Emotional Support Dialog Systems, ACL 2021. [Source LICENSE: CC BY-NC 4.0](https://github.com/thu-coai/Emotional-Support-Conversation/blob/f262d062ad74cb39b17ea476facc81568ddcba24/LICENSE). The source README also states academic research use; original notices are retained in local source documentation. No blanket collection license is assigned.
556
+
557
+ The 1,300 original ESConv conversations yield two human participant perspectives each (2,600 records). Target speaker becomes assistant; the other speaker becomes user. Only the seeker receives the original emotion and situation as system context. Supporter context omits that private information. Source wording, whitespace, utterance boundaries and ordering remain unchanged, including conversational study-control remarks. Strategy labels, ratings and post-chat surveys are excluded from rendered messages. No OdysSim personas or records and no FailedESConv supplement are imported.
558
+
559
+ Splits reproduce the authors' codes_zcj/_reformat/process.py recipe for the identical 1,300-record source: seed 13, 70% train / 15% validation / 15% test by conversation (1,820 / 390 / 390 perspective rows). These are not the older 1,053-dialogue experiment assignments. Both perspectives stay together; actual participant disjointness is not established. No exact or qualifying complete-prefix overlap was found internally or against the publication baseline. Comparison is not exhaustive semantic deduplication. Source context timing and every support response's quality are not independently verified; semantic review covers seeded samples.
data/test/ESConv.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea26b053a50679fe1f99f5bbf95c729302f7afeb327bbd3bb60f58c232906dd0
3
+ size 243021
data/train/ESConv.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a1574c2b94ce73c7f9e989f0213d5587e765a5d8f664104ffd0c8a0d8e697e2
3
+ size 1076962
data/validation/ESConv.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:02daaea97dcdb600dacbf7f6d1ccbc7496e64dc4ebdf05848d4197c75eceb6a8
3
+ size 240452