Article: Dynamic outlier detection for lexical richness in English writing Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.83 pp.82 - 104 Abstract: In the scenario of long-term English writing assessment, learners 'vocabulary output characteristics are susceptible to the continuous change of time span and context, which leads to the distortion of vocabulary richness indicators under static assessment. We design a dynamic context-aware outlier detection mechanism based on graph autoencoder, which aims to eliminate the interference of static word frequency preference and restore the learner's real dynamic vocabulary development trajectory. We use the sliding window-based depth map clustering technique because it can associate isolated words into a structured semantic network, so as to accurately identify those outliers that appear abrupt or completely incoherent depending on the context. Discarding a general qualitative analysis, our extensive experiments lead to specific numerical verification. Compared with the latest dynamic isolation forest benchmark model, the proposed mechanism improves the comprehensive evaluation index score to 92.4%, which proves that it has high stability and processing efficiency. Inderscience Publishers - linking academia, business and industry through research

Title: Dynamic outlier detection for lexical richness in English writing

Authors: Jia Yao

Addresses: School of Foreign Languages, Minnan University of Science and Technology, Shishi, 362700, China

Abstract: In the scenario of long-term English writing assessment, learners 'vocabulary output characteristics are susceptible to the continuous change of time span and context, which leads to the distortion of vocabulary richness indicators under static assessment. We design a dynamic context-aware outlier detection mechanism based on graph autoencoder, which aims to eliminate the interference of static word frequency preference and restore the learner's real dynamic vocabulary development trajectory. We use the sliding window-based depth map clustering technique because it can associate isolated words into a structured semantic network, so as to accurately identify those outliers that appear abrupt or completely incoherent depending on the context. Discarding a general qualitative analysis, our extensive experiments lead to specific numerical verification. Compared with the latest dynamic isolation forest benchmark model, the proposed mechanism improves the comprehensive evaluation index score to 92.4%, which proves that it has high stability and processing efficiency.

Keywords: dynamic outlier detection; vocabulary richness; graph autoencoder; GAE; English writing assessment; spatiotemporal sequence modelling.

DOI: 10.1504/IJICT.2026.155120

International Journal of Information and Communication Technology, 2026 Vol.27 No.83, pp.82 - 104

Received: 28 Apr 2026
Accepted: 03 Jun 2026
Published online: 27 Jul 2026 *

Title: Dynamic outlier detection for lexical richness in English writing

Keep up-to-date