Constructing a Linguistic Expression Corpus on the Values and Behaviors of Time Series Information Toward Compilation of Trend Information

Abstract

Statistical information such as gasoline prices and approval ratings of political parties, that has time as one of its domains and numerical values as its range is called time series information. You can obtain the trend of a topic you are interested in by extracting time series information from documents on the topic and compiling it. You, in turn, use the trend for the understanding of and information access from those documents. In order to support mainly to design and implement such a information compilation system, we have constructed a corpus, by extracting linguistic expressions describing the values and the behaviors of time series information from documents, in this case newspaper articles, and coding those contents into a formal framework. This paper explains the coding schema used for this corpus construction, and shows its appropriateness. This paper also includes findings obtained through the corpus construction, such as some classification of syntactic patterns and the vocabulary for describing the behaviors of time series information.