hope speech detection | Pranjal Aggarwal

Even in the toughest of times, humans manage to find hope. No better way to illustrate this than with a montage of some of the most poignant moments from the last few years, when people from across the world came together and helped each other amidst the coronavirus pandemic.

Social media has a huge impact on society. Everyone is free to state their opinions and they could be both positive and negative. Constructive criticism is always good but the way things are turning in today’s world, social media has laid its roots deep into our society. It has come up with its own pros and cons. Thus, in order to minimize the cons, we have tried to come up with an NLP based model for hope speech detection (Aggarwal et al., 2022). Removing those harsh comments will definitely help in uplifting the society and strengthening our beliefs.

The model is comprised of three components: the pre-processing unit, the feature extraction and data balancing unit, and the classification unit, as illustrated in the figure.

In the pre-processing unit, the dataset undergoes cleanup, involving the removal of unwanted text portions, expansion of abbreviations, and lemmatization for converting words to their root forms. The resulting output is then directed to the feature extraction unit, where the TF-IDF vectorization technique is employed to discern the significance of specific words or phrases within the dataset. Consequently, sentences in the dataset are transformed into vectors with distinct values. These vectors are subsequently utilized to balance the dataset through methods such as SMOTE and ADASYN, involving the addition of synthetic samples.

The outcomes of the above processes are fed into the final unit, which includes various traditional machine learning models like Naïve Bayes, Logistic Regression, and SVM. These models were implemented using the scikit-learn library in Python. We conducted experiments with several traditional models and also incorporated the BERT architecture into our analysis.

references

2022

Hope Speech Detection on Social Media Platforms

Pranjal Aggarwal, Pasupuleti Chandana, Jagrut Nemade, and 3 more authors

In Cybercrime in Social Media, 2022

Abs DOI Bib

Ever since personal computers became widely available in the consumer market, the amount of harmful content on the internet has significantly expanded. In simple terms, harmful content is anything online which causes a person distress or harm. It may include hate speech, violent content, threats, non-hope speech, etc. The online content must be positive, uplifting and supportive. Over the past few years, many studies have focused on solving this problem through hate speech detection, but very few focused on identifying hope speech. This paper discusses various machine learning approaches to identify a sentence as Hope Speech, Non-Hope Speech, or a Neutral sentence. The dataset used in the study contains English YouTube comments and is released as a part of the shared task "EACL-2021: Hope Speech Detection for Equality, Diversity, and Inclusion". Initially, the dataset obtained from the shared task had three classes: Hope Speech, non-Hope speech, and not in English; however, upon deeper inspection, we discovered that dataset relabeling is required. A group of undergraduates was hired to help perform the entire dataset’s relabeling task. We experimented with conventional machine learning models (such as Naïve Bayes, logistic regression and support vector machine) and pre-trained models (such as BERT) on relabeled data. According to the experimental results, the relabeled data has achieved a better accuracy for Hope speech identification than the original data set.
@incollection{aggarwalhope, author = {Aggarwal, Pranjal and Chandana, Pasupuleti and Nemade, Jagrut and Sharma, Shubham and Saumya, Sunil and Biradar, Shankar}, booktitle = {Cybercrime in Social Media}, title = {Hope Speech Detection on Social Media Platforms}, pages = {67--84}, publisher = {Chapman and Hall/CRC}, doi = {10.1201/9781003304180-4}, year = {2022}, acceptance = {Accepted} }