The web is a complex information ecosystem that provides a large variety of content changing over time as a consequence of the combined effects of management policies, user interactions and external events. These highly dynamic scenarios challenge technologies dealing with discovery, management and retrieval of web content. In this paper, we address the problem of modeling and predicting web dynamics in the framework of time series analysis and forecasting. We present a general methodological approach that allows the identification of the patterns describing the behavior of the time series, the formulation of suitable models and the use of these models for predicting the future behavior. Moreover, to improve the forecasts, we propose a method for detecting and modeling the spiky patterns that might be present in a time series. To test our methodological approach, we analyze the temporal patterns of page uploads of the Reuters news agency website over one year. We discover that the upload process is characterized by a diurnal behavior and by a much larger number of uploads during weekdays with respect to weekend days. Moreover, we identify several sudden spikes and a daily periodicity. The overall model of the upload process – obtained as a superposition of the models of its individual components – accurately fits the data, including most of the spikes.
(2019). A methodological approach for time series analysis and forecasting of web dynamics . Retrieved from http://hdl.handle.net/10446/202728
A methodological approach for time series analysis and forecasting of web dynamics
Della Vedova, Marco L.;
2019-01-01
Abstract
The web is a complex information ecosystem that provides a large variety of content changing over time as a consequence of the combined effects of management policies, user interactions and external events. These highly dynamic scenarios challenge technologies dealing with discovery, management and retrieval of web content. In this paper, we address the problem of modeling and predicting web dynamics in the framework of time series analysis and forecasting. We present a general methodological approach that allows the identification of the patterns describing the behavior of the time series, the formulation of suitable models and the use of these models for predicting the future behavior. Moreover, to improve the forecasts, we propose a method for detecting and modeling the spiky patterns that might be present in a time series. To test our methodological approach, we analyze the temporal patterns of page uploads of the Reuters news agency website over one year. We discover that the upload process is characterized by a diurnal behavior and by a much larger number of uploads during weekdays with respect to weekend days. Moreover, we identify several sudden spikes and a daily periodicity. The overall model of the upload process – obtained as a superposition of the models of its individual components – accurately fits the data, including most of the spikes.File | Dimensione del file | Formato | |
---|---|---|---|
2019_trans-comp-collective-intelligence.pdf
Open Access dal 23/06/2021
Descrizione: Under Springer Nature terms of use for archived accepted manuscripts (AMs)
Versione:
postprint - versione referata/accettata senza referaggio
Licenza:
Licenza default Aisberg
Dimensione del file
483.34 kB
Formato
Adobe PDF
|
483.34 kB | Adobe PDF | Visualizza/Apri |
2019_DellaVedova_Transactions.pdf
Solo gestori di archivio
Versione:
publisher's version - versione editoriale
Licenza:
Licenza default Aisberg
Dimensione del file
385.12 kB
Formato
Adobe PDF
|
385.12 kB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
Aisberg ©2008 Servizi bibliotecari, Università degli studi di Bergamo | Terms of use/Condizioni di utilizzo