Japanese | English
Overview
In order to improve product and service quality management, early problem detection is the key issue. In practice, however, it is not easy to monitor unforeseenable events. The alerting technology is studied and developed to make a solution for such purpose by utilizing large unstructured data such as log data stored in call centers.
The alerting system analyzes both structured and unstructured(text) data to extract information with reference to related original documents which can be the key to detect problems in timely manner.
Problem Monitoring by Alerting
Correlation Analysis
Once we prepare the category trees of elements that can be a feature of a defect such as product categories, component categories and problem categories, the alerting system monitors all the patterns of the problems expressed by those categories. Each category is mapped to a document set by flexible rules of natural language processing.
Defect Expression by Categories and Detection of Defects
Time Series Analysis
In the time series analysis, the alerting system ranks top several tens of word frequency time series from the several thousands of series by likelihood of increase in the future. The large set of time series contains those with variety of scales, and each of them includes stationary noise. The alerting system is specialized to call center log data to consider the scale dependency and to distinguish noise and anomaly in high accuracy.
Example of time series which was expected to increase
Example of time series whose changes in frequency was ignored