Text-ADBench: Text Anomaly Detection Benchmark Based on LLM Embeddings

Text anomaly detection (AD) is a critical task in natural language processing (NLP), with applications spanning fraud detection, misinformation identification, spam detection, and content moderation, and so on. Despite significant advances in large language models (LLMs) and AD algorithms, the absence of standardized and comprehensive benchmarks for evaluating the existing AD methods on text data limits rigorous comparison and development of innovative approaches. This work performs a comprehensive empirical study and introduces a benchmark for text AD, leveraging embeddings from diverse pretrained language models across a wide array of text datasets. Our work systematically evaluates the effectiveness of embedding-based text AD by incorporating: 1) early language models (global vectors for word representation (GloVe), BERT); 2) multiple LLMs [LLaMA-2, LLaMA-3, Mistral, and OpenAI embedding models (small, ada, large)]; 3) multidomain text datasets (news, social media, and scientific publications); and 4) comprehensive evaluation metrics (AUROC, AUPRC). Our experiments reveal a critical empirical insight: embedding quality significantly governs AD efficacy, and deep-learning-based approaches demonstrate no performance advantage over conventional shallow algorithms (e.g., K-nearest neighbor (KNN), OCSVM) when leveraging LLM-derived embeddings. In addition, we observe strongly low-rank characteristics in cross-model performance matrices, which enables an efficient strategy for rapid model evaluation (or embedding evaluation) and selection in practical applications. Furthermore, by open-sourcing our benchmark toolkit that includes all embeddings from different models and code, this work provides a foundation for future research in robust and scalable text AD systems.