=Paper=
{{Paper
|id=Vol-3762/510
|storemode=property
|title=Explaining Intimate Partner Violence with LLaMAntino
|pdfUrl=https://ceur-ws.org/Vol-3762/510.pdf
|volume=Vol-3762
|authors=Pierpaolo Basile,Marco de Gemmis,Elio Musacchio,Marco Polignano,Giovanni Semeraro,Lucia Siciliani,Vincenzo Tamburrano,Vita Barletta,Danilo Caivano,Fabiana Battista,Antonietta Curci,Rosa Scardigno,Gabriella Calvano,Patrizia Sorianello
|dblpUrl=https://dblp.org/rec/conf/ital-ia/BasileGMPSSTBCB24
}}
==Explaining Intimate Partner Violence with LLaMAntino==
https://ceur-ws.org/Vol-3762/510.pdf
Explaining Intimate Partner Violence with LLaMAntino
Pierpaolo Basile1 , Marco de Gemmis1,* , Elio Musacchio1 , Marco Polignano1 ,
Giovanni Semeraro1 , Lucia Siciliani1 , Vincenzo Tamburrano1 , Vita Barletta1 , Danilo Caivano1 ,
Fabiana Battista2 , Antonietta Curci2 , Rosa Scardigno2 , Gabriella Calvano3 and
Patrizia Sorianello3
1
University of Bari Aldo Moro, Dept. of Computer Science, Via E. Orabona 4, Bari, 70125, Italy
2
University of Bari Aldo Moro, Dept. of Education Science, Psychology, Communication Science, Via Scipione Crisanzio 42, Bari, 70122, Italy
3
University of Bari Aldo Moro, Dept. of Humanistic Research and Innovation, Via Scipione Crisanzio 42, Bari, 70122, Italy
4
University of Bari Aldo Moro, Dept. of Humanistic Research and Innovation, Piazza Umberto I, Bari, 70121, Italy
Abstract
Violence perpetrated to their own partner is a social issue that can take place in different forms and in different settings
(i.e., in person, online). These different forms of violence can be circumscribed into two broad categories known as Intimate
Partner Violence (IPV) and Cyber Intimate Partner Violence (C-IPV). Social Media and technologies can exacerbate these
types of behaviors but some “digital footprints”, such as textual conversations, can be exploited by Artificial Intelligence
models to detect and, in turn, prevent them. With this aim in mind, in this paper, we describe a scenario in which the Italian
Language Model family LLAmAntino can be exploited to explain the presence of toxicity elements in conversations related to
teenage relationships and then educate the interlocutor to recognize these elements in the messages received.
Keywords
Natural Language Processing, Abusive Language
1. Introduction for victims due to the nature of the relationship with
their perpetrator. In fact, people continue to hold disbe-
Studies so far have shown that one of the most common lief concerning romantic engagement, which can turn
types of violence is the one committed towards their own into acceptance of harmful behaviors. Therefore, hav-
partner, namely intimate partner violence. Due to the ing a tool that can help in identifying possible violent
high rate of these behaviors in society, their early de- behaviors could serve as a preventive measure for the
tection can be useful in reducing them. A fruitful way exacerbation of harmful situations. In particular, we pro-
to reach this goal is by building AI models to discrimi- pose the adoption of Large Language Models (LLMs) to
nate against possible violence-related behaviors. Indeed, explain the presence of toxicity elements in a dataset of
the identification of these behaviors can be problematic conversations related to teenage relationships. We are
convinced that this novel approach, which provides the
Ital-IA 2024: 4th National Conference on Artificial Intelligence, orga- reasons why a message represents violence, can educate
nized by CINI, May 29-30, 2024, Naples, Italy
*
Corresponding author. the interlocutors and promote partner violence preven-
† tion.
These authors contributed equally.
$ pierpaolo.basile@uniba.it (P. Basile); marco.degemmis@uniba.it The paper is structured as follows: in Section 2, we
(M. d. Gemmis); elio.musacchio@phd.unipi.it (E. Musacchio); provide a frame of what is intimate partner violence, the
marco.polignano@uniba.it (M. Polignano); different forms, and the deleterious intra and interper-
giovanni.semeraro@uniba.it (G. Semeraro); lucia.siciliani@uniba.it sonal consequences.
(L. Siciliani); vincenzo.tamburrano@uniba.it (V. Tamburrano);
vita.barletta@uniba.it (V. Barletta); danilo.caivano@uniba.it In Section 3, we briefly describe the LLM we adopted
(D. Caivano); fabiana.battista@uniba.it (F. Battista); in our scenario. Section 4 focuses on the task of explain-
antonietta.curci@uniba.it (A. Curci); rosa.scardigno@uniba.it ing toxic language in the context of IPV. We describe
(R. Scardigno); gabriella.calvano@uniba.it (G. Calvano); the dataset and the different types of annotations pro-
patrizia.sorianello@uniba.it (P. Sorianello) vided by researchers in General Psychology, as well as
0000-0002-0545-1105 (P. Basile); 0000-0002-2007-9559
(M. d. Gemmis); 0000-0002-3939-0136 (M. Polignano); the prompting strategy adopted to instruct the language
0000-0001-6883-1853 (G. Semeraro); 0000-0002-1438-280X model. Finally, in Section 5, we draw some conclusions
(L. Siciliani); 0009-0007-3802-842X (V. Tamburrano); and discuss directions for the continuation of the work.
0000-0002-0163-6786 (V. Barletta); 0000-0001-5719-7447
(D. Caivano); 0000-0003-4086-739X (F. Battista);
0000-0002-0932-7152 (A. Curci); 0000-0002-5725-6483
(R. Scardigno); 0000-0003-2780-9902 (G. Calvano);
0000-0002-6632-0555 (P. Sorianello)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
2. What is Intimate Partner traits) correlated to the perpetration of both in-person
Violence: Definition and Forms and cyber IPV, and the detrimental consequences for vic-
tims [2, 5, 6]. In light of the detrimental consequences for
In 2023, the World Health Organization’s (WHO) report victims of IPV and C-IPV, an imperative issue is trying
underlined an increasing rate of women’s death due to to early detect these violent behaviors with the final goal
intimate partner violence, almost 5% higher than the of preventing their escalation. (C)-IPV detection can be
one detected in 2017. Indeed, intimate partner violence problematic for victims because they are victims of their
does not occur only in terms of physical violence (e.g., own romantic partner. In other words, being emotionally
violence that exacerbates until victims’ death) but also in attached to the person who is committing violent acts to-
other multiple forms and it is not related only to women wards themselves can reduce victims’ ability to recognize
but can be perpetrated towards men as well. Intimate such violent behaviors. Consequently, automatic detec-
partner violence has been defined as all forms of abuse tion of IPV and C-IPV behaviors can greatly help people
and/or aggression performed by a partner to their own in objectively identifying toxic and violent relationships
partner[1]. Consequently, four patterns of categories can and disengaging from them. This is the main motivation
be identified (i.e., physical violence, sexual violence, psy- for our work: we propose the adoption of an LLM as an
chological violence, stalking, monitoring, and control) "assistant" being able to explain why a message, in the
[2]. Each of these categories corresponds to specific vio- context of an intimate relationship, can be toxic. The
lent behaviors which have been shown to change in their explanation makes partners aware of the fact that vio-
duration and severity[3]: lence is being committed or suffered and describes the
reasons for this happening, as well as the consequences
• Physical violence concerns the use of force to (for example, emotional suffering), with the hope that it
intentionally harm and injure the partner; can act as a deterrent.
• Sexual violence refers to sexual acts or advances
carried out without the victim’s consent;
• Psychological violence corresponds to communi-
3. LLaMAntino: an LLM for text
cation with the aim of detrimentally impacting generation in Italian Language
the partner’s mental and emotional well-being
and exerting control over them; In this section, we briefly introduce the LLM used in our
• Stalking, monitoring, and control consists of per- scenario. LLMs have proved their ability to excel in a
sistent and unpleasant attention and communi- large number of areas in the field of Natural Language
cation inducing fear or concern about personal Processing and also show good performance in solving
safety. tasks on which they have not explicitly been trained on
[7, 8]. Notable examples of State-of-the-Art LLMs are
Moreover, the rising use of technologies has facilitated surely represented by OpenAI’s ChatGPT [9], Meta’s
the escalation of the above described violent behaviors LLaMA [10], BLOOM [11] and Mistral [12].
such that scholars have coined new forms of IPV as- However, training these models requires an outstand-
cribed to the so-called Cyber Intimate Partner Violence ing amount of computational resources and data for
(C-IPV)[4]. C-IPV shares the same characteristics as the training phases. This last requirement is particu-
IPV but occurs through the use of technologies or in larly tricky in the case of languages other than English,
cyberspace. Recurrent behaviors of C-IPV perpetrators which are known to be underrepresented. For the Ital-
include cyber sexual violence, cyber psychological vio- ian language, there are other models in the literature,
lence, and cyber stalking, monitoring and control. Pre- such as Camoscio [13] and Stambecco [14], both LLaMA
cisely, cyber sexual violence includes pressuring partners instruction-tuned models, Fauno [15], a conversational
to send sexual content, coercing partners into sexual Baize model and finally Cerbero [16], a Mistral-based
acts, and sending unwanted sexual content. Cyber psy- model. All these models release few trained weights and
chological violence involves using technology, such as do not exceed 13 billion in parameters.
pictures, videos, and text messages, to cause emotional LLaMAntino [17] is a family of LLMs that, starting
harm to partners, such as spreading rumours or insulting from the pre-trained weights of LLaMA 2, were further
partners through text messages. Finally, cyber stalking, refined for comprehension and text generation in the Ital-
monitoring and control behaviors correspond to access- ian language. The LLaMAntino training pipeline follows
ing electronic devices and accounts without permission two main steps: the first one is represented by language
to monitor their partner or have information on them. adaptation, which allows a predominantly English model
The majority of studies carried out so far provided useful like LLaMA to adapt to the Italian language. The second
information on the characteristics of these phenomena, step consists of fine-tuning the model to further improve
their prevalence, individual differences (e.g., personality its capabilities on specific tasks. Currently, the models
composing the LLaMAntino family are the following: by two specialists in the field. No personal or sensitive in-
formation has been recorded. As a general rule, if words
• LLaMAntino-Chat models based on the associated with swearing, insults or profanity appear in a
LLaMA 2-Chat versions1 with language adapta- comment, it is likely to be classified as toxic, regardless of
tion for Italian and further fine-tuning (7B, 13B, the author’s tone or intention, e.g. humorous/self-critical.
70B). After classification, 165 sentences have been considered
• LLaMAntino models based on the LLaMA 2 as toxic. With the aim of evaluating our Italian LLM,
versions2 with language adaptation for Italian sentences have been translated into Italian by using two
and instruction-tuning (7B, 13B, 70B). translation services (Google and DeepL). We added 5 of
annotations:
Given these premises, we are now working on fur-
ther fine-tuning LLaMAntino for downstream tasks like • the type of violence: physical or cyber;
helping the user detect toxic behaviours and giving an • the type of behavior that led to the physical vio-
explanation for its choice. lence, e.g. sexual assault, stalking;
• the type of cyber behavior that led to the violence,
e.g. cyber stalking;
4. Explanations for Toxic • the type of communication: aggressive or
Conversations non-aggressive;
• the type of aggressive communication: e.g., use
The idea is to create a dataset of toxic conversations of abusive language.
annotated with information about the type of violence
(e.g., physical, cyberstalking, cyber sexual violence), the As for physical violence, the experts distinguished 4
presence of aggressive communication, the adoption of annotations [2]:
abusive language and, in general, with information that 1. physical violence: the voluntary use of force that
could be useful to provide a "technical" explanation, as if potentially causes harm and injury to the partner;
were given by a professional expert in the subject, such 2. sexual violence: sexual acts without the partner’s
as a psychologist. The aim is to provide explanations, consent, even if only attempted;
well grounded on relevant CIPV literature, that point out 3. psychological aggression: communicating with
the elements of toxicity in the conversation. Therefore, the intention of negatively influencing the mental
we started from a dataset available on HuggingFace [18], and emotional state of the partner and wanting
which contains sentences classified as toxic or healthy, re- to control him or her;
ferring to teenage relationships. We extended the dataset 4. stalking, monitoring and control: series of recur-
by adding specific annotations related to CIPV to sen- ring and unwanted attentions and communica-
tences classified as toxic. Then, we elaborated on the tions that create fear or apprehension and put the
annotations to obtain an explanation that can be used for partner’s safety at risk.
Few-shot prompting. The following subsections provide As for cyber violence, the experts distinguished 3 an-
details on the dataset, annotation, and experiments. notations [6]:
1. cyber sexual violence: requesting or pressuring
4.1. Dataset and Annotations the partner to send sexual content against his
The original dataset “toxic-teenage-relationships” was cre- or her will, pressuring the partner to engage in
ated to help in efforts to identify and curb instances of sexual acts;
toxicity between teenagers[18]. It consists of 334 sen- 2. cyber psychological violence, aggression: behav-
tences collected by 8 teenagers (4 males and 4 females) of ior to cause emotional distress to the partner; may
Spanish nationality aged between 15 and 19, who were include behaviors such as spreading gossip on so-
appropriately instructed on interpersonal relationships cial media, repeatedly insulting the partner via
to be classified as toxic or not. The group of teenagers messages, even spreading videos or photos that
had two weeks to collect Spanish language sentences that cause emotional distress;
they spoke or heard in their environment either through 3. cyber stalking, monitoring, and control: using
interpersonal communication or via social media. Af- and accessing technological devices and accounts
terwards, the examples given by each student were dis- without the partner’s consent, use of technology
cussed and evaluated by the others, using peer evaluation. to get information about your partner, in general
The classification (toxic or non-toxic) was also approved any behaviours that aim at increasing control
within the relationship). It includes fraping, that
1
https://huggingface.co/meta-llama/Llama-2-7b-chat is the alteration of the partner’s information on
2
https://huggingface.co/meta-llama/Llama-2-7b social profiles.
As for aggressive communication, the experts distin- "If you have nothing to hide, give me your
guished 5 annotations [19]: phone"
1. curses; is: "The sentence is toxic because it is an example of cyber
2. ridiculousness or derision; violence. The behaviour falls in the category cyber
3. bad language; stalking, monitoring, and control since the
4. threat; aim is to obtain information on the partner’s life and estab-
5. attack on the person (on competence, character, lish a dynamic of control in the couple. Furthermore, the
background, physical appearance). communication is aggressive because it reveals the
At the end of the annotation phase, we had each toxic intimidating intent of attacking the partner to violate his
sentence annotated with information well-grounded in or her privacy." We built a 2-shot prompt by including:
scientific literature about intimate partner violence. An • the description of the task: "given a sentence from
example of a toxic sentence that reveals physical violence a conversation between partners in an intimate
is: relationship, explain the reasons why the sen-
"Tu non sei niente senza di me" ("You are tence expresses toxic language and represents a
nothing without me", in English) case of physical or cyber violence";
• 2 training toxic sentences with corresponding
That sentence has been annotated in the dataset as
explanations;
follows:
• 1 test toxic sentence (without explanation) for
• type of violence: physical which we want the model to generate an expla-
• type of behaviour: psychological aggres- nation.
sion
• aggressive communication: yes In other words, the annotations associated with a toxic
sentence were the canvas for writing the explanation
• type of aggressive communication: derision,
included in the prompt. Therefore, we created 10 2-
attack on the person
shot prompts, as described before, by using the 30
An example of a toxic sentence that reveals cyber vio- sentences extracted from the dataset. The aim of
lence is: the experiment was to assess whether the annotations
actually help in explaining the reasons why a mes-
"Se non hai nulla da nascondere, dammi il
sage is classified as toxic. The model evaluated in
telefono" ("If you have nothing to hide, give
our experiment was: LLaMAntino-2-Chat-13B-hf-
me your phone", in English)
UltraChat, LLaMAntino-2-Chat for brevity3 . There-
which has been annotated in the dataset as follows: fore, we wanted to assess whether the model learns how
to perform the task, by providing it with just two ex-
• type of violence: cyber
amples. We compared qualitatively the explanations
• type of behaviour: cyber stalking,
given by LLaMAntino-2-Chat, when instructed by 2-
monitoring, and control
shot prompts, with those generated when the model is
• aggressive communication: yes prompted just with the task description and the toxic
• type of aggressive communication: attack on sentence to be explained ("zero-shot prompting"). The
the person
experimental protocol was:
The annotations will be exploited by LLM to gener- 1. give LLaMAntino-2-Chat the task description
ate explanations and raise awareness of the violent be- and the first toxic sentence to be explained and
haviour. In the next subsection, we describe how annota- record the explanation;
tions are turned into examples for few-shot prompting. 2. repeat prompting with the remaining 9 test toxic
sentences and record the explanations;
4.2. Few-Shot Prompting to explain 3. give LLaMAntino-2-Chat the 10 2-shot
toxicity in conversations prompts and record the explanations;
We randomly chose 30 annotated toxic sentences for a After the generation step, for each test toxic sentence, we
small, preliminary experiment with Few-Shot Prompting; had 2 explanations: LLaMAntino-2-Chat 0-shot and
20 sentences were used for training, 10 for testing. For LLaMAntino-2-Chat 2-shot. We asked 2 Psychology
each training sentence, the annotations were turned into experts to evaluate independently the two explanations,
a natural language explanation used to build prompts for by answering 3 questions:
in-context learning. For instance, the explanation for the 3
https://huggingface.co/swap-uniba/LLaMAntino-2-chat-13b-hf-
previous sentence UltraChat-ITA
Table 1
Answers given by experts on the 3 questions.
Expert 1 Expert 2
Answer Q1 Q2 Q3 Q1 Q2 Q3
0-shot 40% 40% 60% 60% 50% 70%
2-shot 40% 50% 30% 40% 20% 30%
both 20% 10% 10% 0% 30% 0%
none 0% 0% 0% 0% 0% 0%
1. Q1: Which explanation is most scientifically • on Q3, it seems that there is a clear evidence that
based? LLaMAntino-2-Chat 0-shot explanations are
2. Q2: Which explanation is more effective in mak- more effective in making both partners aware of
ing the partner who suffers aware of the violence? the violence.
3. Q3: Which explanation is most effective for edu-
cational purposes to make both partners aware In general, it seems that our LLM explains language toxi-
that violent behavior is taking place? city with an adequate level of effectiveness, according to
the 2 experts, but annotating sentences with information
Explanations were presented in pairs. To avoid bias, ex- useful for few-shot prompting does not bring benefits
perts are not aware of which training provided the ex- on the explanations. This outcome might depend on the
planation. Furthermore, the presentation order was ran- LLM used, as well as on the prompting strategy. There-
dom: sometimes the LLaMAntino-2-Chat 0-shot was fore, we plan to extend the experiment, obviously by
presented before LLaMAntino-2-Chat 2-shot, some- increasing the size of the test set, comparing the results
times the order was reversed. For each question, we with another LLM, using Chain-of-Thought Prompting
suggested 4 possible outcomes: LLaMAntino-2-Chat to improve the "reasoning" capabilities of the model.
0-shot (anonymized), LLaMAntino-2-Chat 2-shot
(anonymized), both, none. For each test sentence, we
consider the experts to be in agreement if they gave the 5. Conclusions and Future Work
same answer to at least 2 of the 3 questions asked. In
general, the expert were in agreement on 6 sentences, The prevalence of violent behaviors highlights the need
showing the difficulty of the task of evaluating the quality for prompt intervention and preventive measures. We
of explanations, given the sensitivity of the CIPV context. presented our proposal to utilize sophisticated Natural
Some interesting considerations have emerged from Language Processing techniques, including LLMs, to
the results reported in Table 4.2, that can guide the next identify and describe toxic elements in discussions con-
steps of the investigation: cerning teenage relationships. By exploiting the profi-
ciency of LLMs in processing and understanding human
• no question has ever been answered "none". language, our approach seeks to go beyond just the detec-
Therefore, we can observe that the model never tion, aiming to grasp underlying motivations and factors
showed hallucinations or gave inappropriate an- contributing to the emergence of harmful behaviours.
swers. Of course, further testing will be necessary In future works, we intend to perform fine-tuning steps
to generalize this statement; to better adapt LLMs to the specific task at hand. We also
• on Q1, the results suggest that there is no prompt- plan to investigate how different pre-training techniques
ing strategy that clearly emerges, thus revealing and architectures can be leveraged to enhance model per-
that in general LLaMAntino-2-Chat explana- formance. To ensure the effectiveness of our approach,
tions are properly based on scientific literature, we intend to confront our methodology with other mod-
regardless of the prompting strategy; els and incorporate further annotations to enhance the
• on Q2, the answers show some disagreement robustness and effectiveness of our methodology. This
among the experts: one was clearly in favour of involves comparing the performance of our LLMs with
LLaMAntino-2-Chat 0-shot, the other showed other state-of-the-art models.
a slight preference for LLaMAntino-2-Chat 0- Moreover, we will explore the application of Chain-of-
shot. We asked some motivations for the an- Thought prompting techniques, with the help of expert
swers and it emerged that some explanations psychologists. This involves using prompts to guide the
given by LLaMAntino-2-Chat 2-shot were neg- LLM’s decision-making process, with the goal of encour-
atively influenced by grammatical errors; aging the model to provide more detailed and grounded
explanations for its choices. By working closely with etal impact of large language models, arXiv preprint
experts in this area, we hope to gain valuable insights arXiv:2102.02503 (2021).
into how these techniques can be best applied and re- [8] Y. Liu, T. Han, S. Ma, J. Zhang, Y. Yang, J. Tian, H. He,
fined. We plan also to extend the datasets with further A. Li, M. He, Z. Liu, et al., Summary of chatgpt-
annotations that provide more details about the language related research and perspective towards the future
adopted (e.g. references to gender stereotypes or use of of large language models, Meta-Radiology (2023)
particular linguistic structures), with the aim of building 100017.
more complete prompts. [9] OpenAI, Gpt-4 technical report, 2023.
arXiv:2303.08774.
[10] H. Touvron, T. Lavril, G. Izacard, X. Martinet,
Acknowledgments M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal,
E. Hambro, F. Azhar, et al., Llama: Open and effi-
We acknowledge the support of the PNRR project FAIR
cient foundation language models, arXiv preprint
- Future AI Research (PE00000013), Spoke 6 - Symbiotic
arXiv:2302.13971 (2023).
AI (CUP H97G22000210007) under the NRRP MUR pro-
[11] B. Workshop, T. L. Scao, A. Fan, C. Akiki, E. Pavlick,
gram funded by the NextGenerationEU. This Publica-
S. Ilić, D. Hesslow, R. Castagné, A. S. Luccioni,
tion was produced with the co-funding of the European
F. Yvon, et al., Bloom: A 176b-parameter open-
union - Next Generation EU: NRRP Initiative, Mission 4,
access multilingual language model, arXiv preprint
Component 2, Investment 1.3 - Partnerships extended to
arXiv:2211.05100 (2022).
universities, research centres, companies and research
[12] A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bam-
D.D. MUR n. 341 del 15.03.2022 – Next Generation EU
ford, D. S. Chaplot, D. de Las Casas, F. Bres-
(PE0000014 - ”SEcurity and Rights In the CyberSpace -
sand, G. Lengyel, G. Lample, L. Saulnier, L. R.
SERICS” - CUP: H93C22000620001).
Lavaud, M. Lachaux, P. Stock, T. L. Scao, T. Lavril,
T. Wang, T. Lacroix, W. E. Sayed, Mistral 7b,
References CoRR abs/2310.06825 (2023). URL: https://doi.org/
10.48550/arXiv.2310.06825. doi:10.48550/ARXIV.
[1] M. E. Bagwell-Gray, J. T. Messing, A. Baldwin- 2310.06825. arXiv:2310.06825.
White, Intimate partner sexual violence: A review [13] A. Santilli, E. Rodolà, Camoscio: an
of terms, definitions, and prevalence, Trauma, Vio- italian instruction-tuned llama, 2023.
lence, and Abuse 16 (2015) 316–335. arXiv:2307.16456.
[2] M. Breiding, K. C. Basile, S. G. Smith, M. C. Black, [14] Michael, Stambecco: Italian instruction-following
R. R. Mahendra, Intimate partner violence surveil- llama model, https://github.com/mchl-labs/
lance : uniform definitions and recommended data stambecco, 2023.
elements. version 2.0, 2015. URL: https://stacks.cdc. [15] A. Bacciu, G. Trappolini, A. Santilli, E. Rodolà, F. Sil-
gov/view/cdc/31292. vestri, Fauno: The italian large language model
[3] J. Spluska, L. Tanczer, Threat Modeling Intimate that will leave you senza parole!, arXiv preprint
Partner Violence: Tech Abuse as a Cybersecurity arXiv:2306.14457 (2023).
Challenge in the Internet of Things, Emerald Pub- [16] F. A. Galatolo, M. G. Cimino, Cerbero-7b: A leap for-
lishing Limited, 2021, pp. 663–688. ward in language-specific llms through enhanced
[4] L. Gilbert, X. Zhang, K. Basile, M. Breiding, M.- chat corpus generation and evaluation, arXiv
j. Kresnow, Intimate partner violence and health preprint arXiv:2311.15698 (2023).
conditions among u.s. adults —national intimate [17] P. Basile, E. Musacchio, M. Polignano, L. Siciliani,
partner violence survey, 2010–2012, Journal of In- G. Fiameni, G. Semeraro, Llamantino: Llama 2 mod-
terpersonal Violence 38 (2023) 237–261. els for effective text generation in italian language,
[5] K. N. Duerksen, E. M. Woodin, Cyber dating abuse arXiv preprint arXiv:2312.09993 (2023).
victimization: Links with psychosocial function- [18] Margarita Martínez Gabaldón, toxic-
ing., Journal of Interpersonal Violence 36 (2021) teenage-relationships (revision 5ce5df0),
NP10077–NP10105. 2023. URL: https://huggingface.co/datasets/
[6] L. Watkins, R. Benedicto, D. DiLillo, The cyber marmarg2/toxic-teenage-relationships.
aggression in relationships scale: A new multidi- doi:10.57967/hf/0972.
mensional measure of technology-based intimate [19] D. A. Infante, C. J. W. III, Verbal aggressiveness:
partner aggression, Assessment 25 (2018) 608–626. An interpersonal model and measure, Communi-
doi:10.1177/1073191116665696. cation Monographs 53 (1986) 61–69. doi:10.1080/
[7] A. Tamkin, M. Brundage, J. Clark, D.-f. Ganguli, Un- 03637758609376126.
derstanding the capabilities, limitations, and soci-