Apply explicitly selected dataset corrections

README.md

@@ -141,6 +141,17 @@ configs:
   - split: test
     path:
     - data/test/FRAMES.parquet
+- config_name: Humanual-Book
+  data_files:
+  - split: train
+    path:
+    - data/train/Humanual-Book.parquet
+  - split: validation
+    path:
+    - data/validation/Humanual-Book.parquet
+  - split: test
+    path:
+    - data/test/Humanual-Book.parquet
 - config_name: ICSI
   data_files:
   - split: train
@@ -521,7 +532,7 @@ Roles are already oriented for user-simulator training:
 <!-- available-datasets:start -->
 ## Available Datasets
 
-**Dataset count: 43**
+**Dataset count: 44**
 
 Select an individual dataset by its name below, or use `default` for the combined collection.
 
@@ -550,7 +561,7 @@ Datasets are grouped by behavioral category:
     "iq2-corpus"
   ],
   "Preferences & Recommendations": [
-    "CCPE-M", "DuRecDial-2.0", "INSPIRED", "OpenDialKG", "Redial"
+    "CCPE-M", "DuRecDial-2.0", "Humanual-Book", "INSPIRED", "OpenDialKG", "Redial"
   ],
   "Role-Playing & Character Simulation": [
     "CRD3"

SOURCES.md

@@ -171,3 +171,39 @@ For MSC and FITS, the downloaded data files and inspected project pages do not
 provide a separate data-license declaration. The ParlAI software license is not
 asserted to license these datasets. Attribution is retained without assigning an
 invented or blanket dataset license; applicable source terms remain unchanged.
+
+## Humanual-Book
+
+Source: [Humanual-Book, Shirley Wu and colleagues, HumanLM / snap-stanford](https://huggingface.co/datasets/snap-stanford/humanual-book/tree/8272ed5890ae2f226d563fe5edccd14847da0653),
+derived from Amazon Reviews 2023. The source declares
+[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/); these
+source-specific terms apply to this subset, without assigning a blanket license
+to the collection. Credit: Wu, Choi, Khatua, Wang, He-Yueya, Weerasooriya, Wei,
+Yang, Leskovec, and Zou, *HUMANLM: Simulating Users with State Alignment Beats
+Response Imitation* (2026).
+
+Each row contains a supplied persona as system context, a book title and ordered
+description paragraphs as user context, and the source review title and text as
+the assistant target. These are review/preference examples, not multi-turn chats.
+Ratings, reviewer metadata, and images are not rendered. Supplied personas are
+inferred upstream and preserved unchanged; no new personas are generated.
+
+Changes: unpack book descriptions; convert HTML line/paragraph boundaries to
+newlines; decode entities once and remove recognized formatting tags and
+stylesheet-only HTML comments. Preserve
+wording, spelling, punctuation, and literal angle-bracket text. The earliest
+review is retained in each of two confirmed duplicate pairs in training. Native
+splits are preserved, renaming val to validation: 34,168 train, 492 validation,
+and 1,960 test examples. Original IDs and all cleanup/duplicate decisions are
+retained in local audits.
+
+The supplied personas may overlap with target reviews; this has not been ruled
+out. Reviewers also occur across native splits. These holdouts must not be
+described as verified leakage-free or as unseen-reviewer evaluations. Automated
+consistency checks confirm source transformation, not authorship or semantic
+quality of every individual review.
+
+Sampled source-quality finding: one test example for product `B09M7MR24D` has
+a notebook title but a review about a pillow cover. The source product IDs agree;
+the original pairing is retained and flagged locally rather than guessed away.
+Literal source typos and expressive angle-bracket labels are also preserved.

data/test/Humanual-Book.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46273f738791e2cfb25cf46ff947693e68615eacadacd86f1dc8a7aea1e9ed22
+size 4553222

data/train/Humanual-Book.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:040cbf59fe1303db53ac5665f3c0743dd988ca369885b504669335d709a8ddaa
+size 98720716

data/validation/Humanual-Book.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eef98e8222574c05b52bb6a46831ad5d609b1021208c4d07b39e7ba960ab717
+size 1055436