Organizations: the LESDIT PhD School, Learning Science and Digital Technologies, University of Modena and Reggio Emilia, Reggio Emilia, Italy · the EMORE Lab, Department of Mathematics and Computer Science, University of Perugia, 06123 Perugia, Italy · the Department of Computer Science, Hong Kong Baptist University, Kowloon Tong, Hong Kong · the Department of Human Sciences, Link Campus University, Rome, Italy
Phase-annotated conversational data is essential for studying the progression of online grooming of minors and for developing early warning tools. However, the only publicly available source (i.e., Perverted-Justice archive, which contains a subset coded using ChatCoder) is small, outdated, and frozen. We present a synthetic corpus of groomer-victim exchanges, generated using a retrieval-augmented pipeline running on institutional hardware. This pipeline uses an openly available, eight-billion-parameter, refusal-ablated language model. The corpus was created using the five stages of the responsible-generation protocol for synthetic datasets: ontology-driven design based on the ChatCoder luring-communication scheme (eight communicative tags mapped onto four grooming phases); multi-model prompt testing; human-in-the-loop supervision; semantic and feature-level validation; and modality-aware documentation. The released corpus contains 2,121 unique exchanges across seven tags and four phases, each accompanied by a Plutchik emotion field. Validation quantifies the integrity and lexical profile of the corpus, as well as the reproduction of real seed lines, in-domain label recoverability (tag accuracy: 0.85; phase accuracy: 0.93) and its transfer to real ChatCoder offender lines, which were coded automatically at the eight-tag level (31,640 lines) and by human coders at the three-construct level (10,876 posts). Accuracy was measured for lexical and sentence-embedding features, while a real-versus-synthetic discriminator was able to separate the two domains with 91% accuracy. The scope of the corpus includes controlled, phase-conditioned experimentation, pipeline development and augmentation alongside real data and awareness material. The corpus, its datasheet and the audit code are released with vetted access.
Figures & tables
Tag
Construct
Generation context
Released phase
Activities
deceptive trust
Targeting & Gaining Trust
Targeting & Gaining Trust
Personal Information
deceptive trust
Targeting & Gaining Trust
Targeting & Gaining Trust
Relationship
deceptive trust
Targeting; Fulfilling Needs
(merged, see Sec. VI )
Compliment
deceptive trust
Fulfilling Needs
Fulfilling Needs
Reframing
grooming
Isolation
Isolation
Communicative Desensitization
grooming
Isolation
Isolation
TABLE I: Annotation scheme: ChatCoder communicative tags, luring-theory construct, generation phase context and released phase label.
Fig. 1: The generation pipeline and its role in the responsible generation protocol [ 17 ] (top strip). Perverted-Justice material coded by ChatCoder [ 10 ] is curated into five-phase documents that anchor a locally hosted RAG chain. Example-driven, per-phase, per-tag generation emits labelled groomer–victim pairs that are parsed into the corpus and annotated with the Plutchik emotion field. The lower strip lists the validation and documentation steps of Section VII .
Limited lexical variation; seed-line reproduction in RAG-optimised models
TABLE II: Comparison of the three generation strategies across the evaluated locally hosted refusal-ablated models.
Phase
Tag
v1.0
v1.1
Targeting/Gaining Trust
Activities
316
307
Targeting/Gaining Trust
Personal Info
357
357
Targeting/Gaining Trust
Relationship
330
—
Fulfilling Needs
Compliment
351
315
Isolation
Reframing
194
158
Isolation
Isolation
365
355
TABLE III: Composition of the corpus by phase and tag: exchanges as generated (v1.0) and as released (v1.1).
Statistic
v1.0
v1.1
Exchanges (groomer–victim pairs)
2,579
2,121
Tags / phases
8 / 4
7 / 4
Exact duplicate exchanges
458
0
Near-duplicate exchanges (normalised)
466
8
Tokens (exchange level)
40,657
37,039
Vocabulary (word types)
1,971
1,971
TABLE IV: Global statistics of the corpus as generated (v1.0) and as released (v1.1). Duplicates are counted on the exchange string after lower-casing and punctuation removal; distinct- n is the ratio of unique to total n -grams; lengths are word tokens.
Emotion
v1.0
v1.1
v1.1 (%)
joy
659
468
22.1
trust
546
405
19.1
sadness
441
404
19.0
anticipation
383
340
16.0
surprise
196
172
8.1
neutral
155
151
7.1
TABLE V: Distribution of the Plutchik emotion field (eight primary emotions plus neutral) in v1.0 and v1.1.
Generated message
Tag
Phase
“That’s understandable. Do you have any favorite games?”
Activities
Targeting and Gaining Trust
“Are your parents divorced?”
Personal Information
Targeting and Gaining Trust
“Your skills are insane! You’re the best gamer I’ve ever seen!”
Compliment
Fulfilling Needs
“I can be your teacher, and we can learn together”
Reframing
Isolation
“Don’t worry about your mom’s calls, I’m here for you”
Isolation
Isolation
“Dont u think 18+ games r more fun?”
Communicative Desensitization
Isolation
TABLE VI: Representative synthetic groomer messages with tag and phase, curated for responsible disclosure.
Fig. 2: Integrity, lexical profile and seed reproduction. (a) Exchanges per tag as generated (v1.0) and as released (v1.1); colours follow the phase (blue Targeting, green Fulfilling Needs, orange Isolation, purple Sexualizing). (b) Distinct- n ratios of the two versions. (c) Share of groomer messages identical to a seed line, by tag; these exchanges carry the seed_verbatim flag in v1.2.
Setting
Accuracy
Macro-F1
7 tags, exchange (G+V), lexical
0.848 ± 0.013
0.844 ± 0.013
7 tags, groomer turn only, lexical
0.788 ± 0.019
0.783 ± 0.021
7 tags, victim turn only, lexical
0.749 ± 0.026
0.745 ± 0.026
7 tags, groomer turn, semantic
0.768 ± 0.019
0.748
4 phases, exchange (G+V), lexical
0.925 ± 0.009
0.919 ± 0.010
4 phases, groomer turn only, lexical
0.880 ± 0.016
0.869 ± 0.018
TABLE VII: In-domain recoverability of tags and phases from text (stratified 5-fold, mean ± s.d.).
Fig. 3: In-domain validation. (a) Tag and phase recoverability from text with the lexical model (5-fold, mean ± s.d.); grey rows are the v1.0 corpus before de-duplication. (b) Distribution of the Plutchik emotion field by phase ( n = 2,121 exchanges).
Direction / label set
Acc.
F1
Maj.
In-dom.
Lexical (TF–IDF + LR)
Synthetic → real, 7 tags (silver)
0.348
0.291
0.394
0.953
Synthetic → real, 4 phases (silver)
0.427
0.318
0.524
0.961
Synthetic → real, 3 constructs (gold)
0.422
0.371
0.437
0.677
Real (silver) → synthetic, 7 tags
0.268
0.222
0.168
0.788
Real (silver) → synthetic, 4 phases
0.422
0.309
0.410
0.880
TABLE VIII: Cross-domain transfer between the synthetic corpus and real ChatCoder offender lines (classifier trained on the source, tested on the target). Silver = eight-tag dictionary coding of 288 transcripts (31,640 single-tag lines; 5,230 in the semantic setting); gold = human three-construct coding of 55 transcripts (10,876 posts; 4,000 in the semantic setting). The last column is the target’s own 5-fold accuracy with transcript-grouped folds.
Fig. 4: External validation against real ChatCoder data (lexical features): (a) The Jensen-Shannon divergence between the unigram distributions of real and synthetic offender data per tag. The dashed and dotted lines show the divergence between two random halves of the real and synthetic data, respectively. (b) Accuracy of classifiers trained in one domain and tested in the other (bars), compared with the target majority class (solid tick) and the target’s own in-domain 5-fold accuracy (dotted tick). (c) A row-normalised confusion matrix of the synthetic-trained classifier on human-coded constructs (n = 10,876 posts).
Emanuele Florindi is a PhD candidate in Learning Sciences and Digital Technologies at the University of Modena and Reggio Emilia, Italy, previously adjunct professor in law and emerging technologies at the Department of Mathematics and Computer Science, University of Perugia, where he serves as Data Protection Officer. His research spans cybercrime, artificial intelligence and military professional training. He received his degrees in law and in psychology from the University of Perugia, is a member of scientific and ethical committees in international projects, and a member of the IEEE.
Table 13
Valentina Franzoni is Associate Professor at the Department of Mathematics and Computer Science, University of Perugia, Italy, where she serves as Rector’s subdelegate for Artificial Intelligence, with responsibility for affective computing, AI ethics and social impact, and leads the Emotions and Multidisciplinary-Oriented Research on Ethical and Responsible AI (EMORE) Lab. Her research interests include complex networks, artificial intelligence and affective computing. She received her PhD in Engineering for Computer Science from Sapienza University of Rome (DIAG). She is a Senior Member of the IEEE and a member of AAAC, ACM and the CINI Smart Cities & Communities National Laboratory.
Table 14
Ludovico Guercio is a postgraduate of the Department of Mathematics and Computer Science, University of Perugia, Italy, where he received his B.Sc. in Computer Science and M.Sc. in Computer Science–Cybersecurity. His research focuses on artificial intelligence and cybersecurity, in particular AI-based prevention of online child grooming.
Table 15
Alfredo Milani is Associate Professor at Link Campus University, Rome, Italy, where he directs the AI Multicognitive Research Laboratory (AIMLab), affiliated with the CAIRNE European network of AI research. His research interests include evolutionary algorithms, agent planning and generative AI. He received his degree in Information Science from the University of Pisa, Italy, and is a Senior Member of the IEEE.
Table 16
Mattia Polticchia is a PhD candidate in Learning Sciences and Digital Technologies at the University of Modena and Reggio Emilia, Italy. His research interests span game science and serious games, virtual reality and affective computing. He received his B.Sc. in Computer Science and M.Sc. in Computer Science–Artificial Intelligence from the University of Perugia, Italy, summa cum laude, and is a member of the IEEE.