Background: Producing, distributing or discussing child sexual abuse materials (CSAM) is often committed through the dark web in order to remain hidden from search engines and regular users. Additionally, on the dark web, the CSAM creators employ various techniques to avoid detection and conceal their activities. The large volume of CSAM on the dark web presents a global social problem and poses a significant challenge for helplines, hotlines and law enforcement agencies. Objective: Identifying CSAM discussions on the dark web and uncovering associated metadata insights into characteristics, behaviours and motivation of CSAM creators. Participants and Setting: We have conducted an analysis of more than 353,000 posts generated by 35,400 distinct users and written in 118 different languages across eight dark web forums in 2022. Out of these, approximately 221,000 posts were written in English and contributed by around 29,500 unique users. Method: We propose a CSAM detection intelligence system. The system uses a manually labelled dataset to train, evaluate and select an efficient CSAM classification model. Once we identify CSAM creators and victims through CSAM posts on the dark web, we proceed to analyze, visualize and uncover information concerning the behaviors of CSAM creators and victims. Result: The CSAM classifier, based on Support Vector Machine model, exhibited good performance, achieving the highest precision of 92.3\%, accuracy of 87.6\% and recall of 84.2\%. Its prediction time is fast, taking only 0.3 milliseconds to process a single post on our laptop. While, the Naive Bayes combination is the best in term of recall, achieving 89\%, and its prediction time is just 0.1 microseconds per post. Across the eight forums in 2022, our Support Vector Machine model detected around 63,000 English CSAM posts and identified near 10,500 English CSAM creators. The analysis of metadata of CSAM posts revealed meaningful information about CSAM creators and their victims, such as: (1) the ages and nationalities of the victims typically mentioned by CSAM creators, (2) forum topics where the CSAM creators assign their posts, and (3) online platforms preferred by CSAM creators for sharing or uploading CSAM. Conclusion: Our CSAM detection system exhibits high performance in precision, recall, and accuracy in real-time when classifying CSAM and non-CSAM posts. Additionally, it can extract and visualize valuable and unique insights about CSAM creators and victims by employing advanced statistical methods. These insights prove beneficial to our partners, i.e. national hotlines and child agencies.

Child sexual abuse inflicts lifelong devastating consequences for victims and is an ongoing social concern. In most countries, child sexual abuse material (CSAM) distribution is illegal. As a result, there are many research papers in the literature which proposed technologies to detect and investigate CSAM. In this survey, a comprehensive search of the peer-reviewed journal and conference paper databases (including preprints) is conducted to identify high-quality literature. We use the PRISMA methodology to refine our search space to 2,761 papers published by Springer, Elsevier, IEEE and ACM. After iterative reviews of title, abstract and full text for relevance to our topics, 43 papers are included for full review. Our paper provides a comprehensive synthesis of the tasks of the current research and how the papers use techniques and datasets to solve their tasks and evaluate their models. To the best of our knowledge, we are the first to focus exclusively on online CSAM detection and prevention with no geographic boundaries, and the first survey to review papers published after 2018. It can be used by researchers to identify gaps in knowledge and relevant publicly available datasets that may be useful for their research.

Sentiment analysis of public opinion expressed in social networks has been developed into various applications, especially in English. Hybrid approaches are potential models for reducing sentiment errors on increasingly complex training data. This paper aims to test some hybrid deep learning models' reliability in some domains' Vietnamese language. Our research questions are to determine whether it is possible to produce hybrid models that outperform the Vietnamese language. Hybrid deep sentiment-analysis learning models are built and tested on reviews and feedback of the Vietnamese language. The hybrid models outperformed the accuracy of Vietnamese sentiment analysis on Vietnamese datasets. It contributes to the growing body of research on Vietnamese NLP, providing insights and directions for future studies in this area.