People’s Online Reading Behavior Research Base On The Human Dynamics

Revealing the features of human behaviors is a long-standing challenge and attractsmuch attention from various social sectors. Scientists focus on exciting findings, whilefor merchants it’s helpful to increase the success with recommending wares; it couldeven supply useful information for the department of public security. In recent years,human actions are well recorded by smart phones which have been widely used. Andluckily we gained some reading records from a mobile Internet-based platform. Thoughstatistical analysis we find that many users’ tastes change rapidly and that there is anobvious book-type flow when people change their tastes from one type to another.Actually, human behaviors are much related to recommendations. And along with theperfect condition that this platform can be used for some comparative experimentsbetween different recommendations, we optimize a recommend algorithm and examineits effectiveness in practice comparing with a traditional one. We simultaneouslyconsider the action of users, the types and popularities of books, etc., and tactfullycombine them with a mature method based on the topology of bipartite graph(users-books). Our method is verified to be much better than the traditional one. Besides,another contribution is that we apply the output of human dynamics analysis torecommender systems, which might contribute to the progress of recommender systemswith the support of our experiments.We complete two tasks in this thesis:(1) Analyze the experimental data through human dynamics analysis, including thedegree distribution of users and books, levels of users’ activities, variations of users’tastes, check if book-type-flow exists. According to the results of analysis, we establishthe user behavior model to provide the data support for the recommendation algorithm.(2) We optimize an existing recommendation algorithm and examine it in practice.By comparing the first-day feedback, increment of Page View, long-term retention rateetc., we conclude that our method is much more excellent.