Light-weight AI and IoT collaboration for surveillance video pre-processing

Language-agnostic deep learning framework for automatic monitoring of population-level mental health from social networks

The majority of visual based surveillance applications and security systems heavily rely on object detection, which serves as a critical module. In the context of crime scene analysis, images and videos play an essential role in capturing visual documentation of a particular scene. By detecting objects associated with a specific crime, police officers are able to reconstruct a scene for subsequent analysis. Nevertheless, the task of identifying objects of interest can be highly arduous for law enforcement agencies, mainly because of the massive amount of data that must be processed. Hence, the main objective of this paper is to propose a DL-based model for detecting tracked objects such as handheld firearms and informing the authority about the threat before the incident happens. We have applied VGG-19, ResNet, and GoogleNet as our deep learning models. The experiment result shows that ResNet50 has achieved the highest average accuracy of 0.92% compared to VGG19 and GoogleNet, which have achieved 0.91% and 0.89%, respectively. Also, YOLOv6 has achieved the highest MAP and inference speed compared to the faster R-CNN. Keywords—Object detection; deep learning; crime scenes; video surveillance; convolutional neural network; YOLOv6

/pdf/deep-learning-models-for-crime-intention-detection-using-1j50cu1g.pdf

Deep Learning Models for Crime Intention Detection Using Object Detection

In recent years, systems have been developed to monitor online content and remove abusive, offensive or hateful content. Comments in online social media have been analyzed to find and stop the spread of negativity using methods such as hate speech detection, identification of offensive language or detection of abusive language. We define hope speech as the type of speech that is able to relax a hostile environment and that helps, gives suggestions and inspires for good to a number of people when they are in times of illness, stress, loneliness or depression. Detecting it automatically, in order to give greater diffusion to positive comments, can have a very significant effect when it comes to fighting against sexual or racial discrimination or when we intend to foster less bellicose environments. In this article we perform a complete study on hope speech, analyzing existing solutions and available resources. In addition, we have generated a quality resource, SpanishHopeEDI, a new Spanish Twitter dataset on LGBT community, and we have conducted some experiments that can serve as a baseline for further research. 

/pdf/hope-speech-detection-in-spanish-c1bwxy4o.pdf

Hope speech detection in Spanish

Abstract Speaking or expressing oneself in an abusive manner is a form of verbal abuse that targets individuals or groups on the basis of their membership in a particular social group, which is differentiated by traits such as culture, gender, sexual orientation, religious affiliation etc. In today's world, the dissemination of evil and depraved content on social media has increased exponentially. Abusive language on the internet has been linked to an increase in violence against minorities around the world, including mass shootings, murders, and ethnic cleansing. People who use social media in places where English is not the main language often use a code-mixed form of text. This makes it harder to find abusive texts, and when combined with the fact that there aren't many resources for languages like Tamil, the task becomes significantly challenging. This work makes use of abusive Tamil language comments released by the workshop “Tamil DravidianLangTech@ACL 2022” and develops adapter-based multilingual transformer models namely Muril, XLMRoBERTa and mBERT to classify the abusive comments. These transformers have been utilized as fine-tuners and adapters. This study shows that in low-resource languages like Tamil, adapter-based strategies work better than fine-tuned models. In addition, we use Optuna, a hyperparameter optimization framework to find the ideal values of the hyper-parameters that lead to better classification. Of all the proposed models, MuRIL (Large) gives 74.7%, which is comparatively better than other models proposed for the same dataset. 

/pdf/on-finetuning-adapter-based-transformer-models-for-2i143gie.pdf

On finetuning Adapter-based Transformer models for classifying Abusive Social Media Tamil Comments

The task of hope speech detection has gained traction in the natural language processing field owing to the need for an increase in positive reinforcement online during the COVID-19 pandemic. Hope speech detection focuses on identifying texts among social media comments that could invoke positive emotions in people. Students and working adults alike posit that they experience a lot of work-induced stress further proving that there exists a need for external inspiration which in this current scenario, is mostly found online. In this paper, we propose a multilingual model, with main emphasis on Dravidian languages, to automatically detect hope speech. We have employed a stacked encoder architecture which makes use of language agnostic cross-lingual word embeddings as the dataset consists of code-mixed YouTube comments. Additionally, we have carried out an empirical analysis and tested our architecture against various traditional, transformer, and transfer learning methods. Furthermore a k-fold paired t test was conducted which corroborates that our model outperforms the other approaches. Our methodology achieved an F1-score of 0.61 and 0.85 for Tamil and Malayalam, respectively. Our methodology is quite competitive to the state-of-the-art methods. The code for our work can be found in our GitHub repository (https://github.com/arunimasundar/Hope-Speech-LT-EDI).

Hope Speech Detection for Dravidian Languages Using Cross-Lingual Embeddings with Stacked Encoder Architecture.

Real-time Surveillance based Crime Detection for Edge Devices.

In recent times, there exists an abundance of research to classify abusive and offensive texts focusing on negative comments but only minimal research using the positive reinforcement approach. The task was aimed at classifying texts into ‘Hope_speech’, ‘Non_hope_speech’, and ‘Not in language’. The datasets were provided by the LT-EDI organisers in English, Tamil, and Malayalam language with texts sourced from YouTube comments. We trained our data using transformer models, specifically mBERT for Tamil and Malayalam and BERT for English, and achieved weighted average F1-scores of 0.46, 0.81, 0.92 for Tamil, Malayalam, and English respectively.

Akshay Ramakrishnan

Papers

Hope Speech Detection for Dravidian Languages Using Cross-Lingual Embeddings with Stacked Encoder Architecture.

Real-time Surveillance based Crime Detection for Edge Devices.

ssn_diBERTsity@LT-EDI-EACL2021:Hope Speech Detection on multilingual YouTube comments via transformer based approach