Knowledge-guided Continual Learning for Behavioral Analytics Systems

TL;DR

This paper tackles performance degradation in behavioral analytics models caused by evolving online deviant behaviors by integrating external knowledge into replay-based continual learning. The authors introduce a knowledge-guided CL framework that augments replay data with external knowledge from Wiktionary (via Wiktextract) and semantic modeling, and they augment exemplars in memory using synonym/hyponym/instance relations. Across three hate/deviant-behavior datasets, the approach reduces catastrophic forgetting and improves accuracy and AUC compared to standard replay baselines, with semantic augmentation (KR_sem) often providing the best gains. The work demonstrates the practicality of knowledge-enhanced continual learning for dynamic text classification tasks and discusses future avenues including richer KBs, multilingual settings, and efficiency considerations.

Abstract

User behavior on online platforms is evolving, reflecting real-world changes in how people post, whether it's helpful messages or hate speech. Models that learn to capture this content can experience a decrease in performance over time due to data drift, which can lead to ineffective behavioral analytics systems. However, fine-tuning such a model over time with new data can be detrimental due to catastrophic forgetting. Replay-based approaches in continual learning offer a simple yet efficient method to update such models, minimizing forgetting by maintaining a buffer of important training instances from past learned tasks. However, the main limitation of this approach is the fixed size of the buffer. External knowledge bases can be utilized to overcome this limitation through data augmentation. We propose a novel augmentation-based approach to incorporate external knowledge in the replay-based continual learning framework. We evaluate several strategies with three datasets from prior studies related to deviant behavior classification to assess the integration of external knowledge in continual learning and demonstrate that augmentation helps outperform baseline replay-based approaches.

Figures (6)

• Figure 1: An illustration of a task series from an existing dataset kennedy2020constructing and an overview of the proposed model training with an external knowledge graph.
• Figure 2: The proposed knowledge-guided CL system that incorporates external knowledge in the learning process. The numbers in circles indicate the order in which the data flows, as described in Section \ref{['sec:method']}.
• Figure 3: Overview of external knowledge base preparation and input data augmentation. The data flow relating to the dotted line applies only to semantic augmentation, as discussed in Section \ref{['sec:augmentation']}.
• Figure 4: Accuracy of individual tasks at each task training stage. The x-axis represents the training task stage, and the y-axis represents the accuracy. Each plot in the column and row corresponds to a test task and dataset, respectively. The tasks are numbered in the same order as they appear for training.
• Figure 5: An augmentation example. Illustrates both semantic augmentation (A1) and incorrect augmentation (A2).