The rapid development of Large Language Models (LLMs) has led to their widespread adoption across various domains, leveraging vast pre-training knowledge and impressive generalization capabilities. However, these models often inherit biased knowledge, resulting in unfair decisions in sensitive applications. It is challenging to remove this biased knowledge without compromising reasoning abilities due to the entangled nature of the learned knowledge within LLMs. To solve this problem, existing approaches have attempted to mitigate the bias using techniques such as fine-tuning with unbiased datasets, model merging, and gradient ascent. While these methods have experimentally proven effective, they can still be sub-optimum in fully disentangling biases from reasoning. To address this gap, we propose Selective Disentanglement Unlearning (SDU), a novel unlearning framework that selectively removes biased knowledge while preserving reasoning capabilities. SDU operates in three stages: identifying biased parameters using a shadow LLM, fine-tuning with unbiased data, and performing selective parameter updates based on weight saliency. Experimental results across multiple LLMs show that SDU improves fairness accuracy by 14.7% and enhances reasoning performance by 62.6% compared to existing baselines.

Most of the existing graph analytics for understanding social behavior focuses on learning from static rather than dynamic graphs using hand-crafted network features or recently emerged graph embeddings learned independently from a downstream predictive task, and solving predictive (e.g., link prediction) rather than forecasting tasks directly. To address these limitations, we propose (1) a novel task -- forecasting user interactions over dynamic social graphs, and (2) a novel deep learning, multi-task, node-aware attention model that focuses on forecasting social interactions, going beyond recently emerged approaches for learning dynamic graph embeddings. Our model relies on graph convolutions and recurrent layers to forecast future social behavior and interaction patterns in dynamic social graphs. We evaluate our model on the ability to forecast the number of retweets and mentions of a specific news source on Twitter (focusing on deceptive and credible news sources) with R^2 of 0.79 for retweets and 0.81 for mentions. An additional evaluation includes model forecasts of user-repository interactions on GitHub and comments to a specific video on YouTube with a mean absolute error close to 2% and R^2 exceeding 0.69. Our results demonstrate that learning from connectivity information over time in combination with node embeddings yields better forecasting results than when we incorporate the state-of-the-art graph embeddings e.g., Node2Vec and DeepWalk into our model. Finally, we perform in-depth analyses to examine factors that influence model performance across tasks and different graph types e.g., the influence of training and forecasting windows as well as graph topological properties.

Folksonomy of movies covers a wide range of heterogeneous information about movies, like the genre, plot structure, visual experiences, soundtracks, metadata, and emotional experiences from watching a movie. Being able to automatically generate or predict tags for movies can help recommendation engines improve retrieval of similar movies, and help viewers know what to expect from a movie in advance. In this work, we explore the problem of creating tags for movies from plot synopses. We propose a novel neural network model that merges information from synopses and emotion flows throughout the plots to predict a set of tags for movies. We compare our system with multiple baselines and found that the addition of emotion flows boosts the performance of the network by learning ≈18% more tags than a traditional machine learning system.

Likability prediction of books has many uses. Readers, writers, as well as the publishing industry, can all benefit from automatic book likability prediction systems. In order to make reliable decisions, these systems need to assimilate information from different aspects of a book in a sensible way. We propose a novel multimodal neural architecture that incorporates genre supervision to assign weights to individual feature types. Our proposed method is capable of dynamically tailoring weights given to feature types based on the characteristics of each book. Our architecture achieves competitive results and even outperforms state-of-the-art for this task.

Suraj Maharjan, John Arevalo, Manuel Montes, Fabio A. González, Thamar Solorio. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.