In this project, we use Natural language processing (NLP) techniques to quantify the influence within a conversation between two users. We use existing NLP algorithms as well as developing new ones in order to estimate the influence solely based on textual data.
We use social media data for the project purpose. We focus on Twitter due to the high activity volume in the platform and the availability of the data. We focus on spoken Arabic language. We do no limit our research to a specific domain (e.g., politics, sports) within the Arabic corpus.
Human-Directed Sentiment Analysis
As part of The Mine project, we introduced a new sentiment analysis task for detecting the sentiment that is expressed by a user toward another user in a discussion thread. The paper, describing this new task has been published on NSURL, 2022
- Date Sources: Twitter threads.
- Data Annotation: human annotation of sentiment between two users on a Twitter conversation.
- Modeling: BERT models to predict the human-directed sentiment.