Predicting Service Outages using Tweets
Sunita A Yadwad1, V. Valli Kumari2

1Sunita A Yadwad , Department of CS &SE , Andhra University , Vishakapatnam , Andhra Pradesh , India.
2Dr V. Valli Kumari , Department of CS &SE , Andhra University , Vishakapatnam , Andhra Pradesh , India.
Manuscript received on February 12, 2020. | Revised Manuscript received on February 21, 2020. | Manuscript published on March 30, 2020. | PP: 421-429 | Volume-8 Issue-6, March 2020. | Retrieval Number: E6911018520/2020©BEIESP | DOI: 10.35940/ijrte.E6911.038620

Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: Every user of the internet has high aspirations on its reliability, efficiency, productivity and in many other aspects of the same. Providing an uninterrupted service is of prime importance .The amount of data along with enormous number of residual traces is increasing rapidly and significantly. As a result, analysis of log data has profoundly influenced many aspects of researcher’s domains. Social media being integral part of the Internet, real time blogging services like Twitter are widely used due to their inherent nature of depicting social graph, propagating information and entire social dynamics. Content of tweets are of major interest to researchers as they reflect individuals experiences, real time events. Researchers have explored several applications of tweet analysis. One such application is detecting service outages through a myriad of messages posted by users regarding unavailability. Simple techniques are enough to extract key semantics from tweets as they are faster alerts for warning about service unavailability. Similarly, the outage mailing lists are text-based messages which are rich in semantic information about the underlying outages. Researchers find it a great challenge to automatically parse and process the data through NLP and text mining for service outage detection. An extensive study was conducted, aiming to explore the research directions and opportunities on log analysis, tweet analysis and outage mailing list analysis for the purpose of detecting and predicting service outages. A systematic- frame work is also articulated with a focus on all stages of analytics and we deliberately discussed potential research challenges & paths in the above said analysis. We introduce three major data analysis methods for diagnosing the causes of service failures , detecting service failures prematurely and predicting them. We analyze Syslogs (contain log data generated by the system) for detecting the cause of a failure by automatically learning over millions of logs and analyze the data of a social networking service (namely, Twitter and outage mails) to detect possible service failures by extracting failure related tweets, which account for less than a percent of all tweet in real time with high accuracy. Paper is an effort not only to detect outages but also to forecast them using twitter analysis based on time series and neural network models. We further propose a log analysis model for the same.
Keywords: Tweets, service outage, time series analysis, log analysis.
Scope of the Article: Predictive Analysis.