VoD System User Loyalty Evolution Analysis Based On Large-scale Data Mining

User loyalty evolution research helps us to understand watching behavior of Internet user, assess system quality, model system load, optimize system algorithm and analyze, predict system performance. Conventional user behavior analysis methods are based on static data without temporal dimension or an index that overall assesses user loyalty; therefore this paper targets to build a model that assess user loyalty, and analyze temporal user behavior process.This paper’s consists of below three problems.1) How to define user loyalty, how it changes over time? This paper utilizes APH to build user loyalty model, and measures temporal user behavior.2) Can user lifespan be predicted with as little data as possible? This paper utilizes machine learning classifier to predict user lifespan with data in a week.3) Can user visiting behavior be modeled and applied in a system level? This paper statistics and models user visiting probabilities, and proposes UV predicting model.The research result shows that user visiting has a inherent mode, and so are loyalty and engagement, people with different visiting patterns behave differently. In addition, user lifespan can be predicted basing on the first week data, and UV of a group of people that newly visit in the same week can be predicted basing on this week’s data. This paper is significant for analyzing user behavior in terms of life cycle and introducing tools such as AHP and machine learning; modeling user visiting behavior and system UV. In addition, this paper defines user lifecycle, user loyalty model and user visiting patterns, which refresh conventional user behavior analyzing methods.