English online learning behavior pattern mining and adaptive path inference based on learning analytics

English online learning platforms accumulate large amounts of learners’ behavioral data. Mining typical learning patterns from these data and constructing personalized learning paths has become an important issue in the field of educational information. Based on the behavior logs of 1000 learners in an English learning platform of a university for eight weeks, this study extracts behavior characteristics from four dimensions: time, interaction, socialization and cognition, and identifies four learning modes, namely, efficient and balanced learning mode, low input mode, high interaction mode and night-owl pattern, with an silhouette coefficient of 0.52. A deep knowledge tracking model integrating behavioral pattern priors is constructed, with hidden layer dimension of 128, mastery threshold of 0.65, and path utility function integrating learning efficiency, knowledge coverage and estimated time burden. In this paper, the model Precision@5 is 0.583, NDCG@5 is 0.612, the learning gain of the experimental group is 10.7% points, the control group is 4.2% points, the t-test t value is 5.23, and the p value is less than 0.001. This study examines the association between behavioral pattern information and improved recommendation accuracy, and explores the potential of pattern-informed recommendations to support learning outcomes.