Temporal specificity-based text classification for information retrieval

Abstract

Time is an important aspect in temporal information retrieval (TIR), a subfield of information retrieval(IR). Web search engines like Google or Bing are common examples of IR systems. An important constituent of asearch engine is news retrieval, where users present their information needs in the form of temporal queries. Users areusually interested in news documents focusing on a particular time period. Existing search engines rarely fulfill thetemporal information requirements as they ignore the temporal information available in the content of news documents,also known as document focus time. Furthermore, information related to multiple time periods in a news documentmakes the identification of document focus time a challenging task. Therefore, it is necessary to classify news documentsbased on temporal specificity before it is possible to use the temporal information in the retrieval process. In this study,we formulate the temporal specificity problem as a time-based classification task by classifying news documents intothree temporal classes, i.e. high temporal specificity, medium temporal specificity, and low temporal specificity. Forsuch classification, rule-based and temporal specificity score (TSS)-based classification approaches are proposed. In theformer approach, news documents are classified using a defined set of rules that are based on temporal features. Thelater approach classifies news documents based on a TSS score using the temporal features. The results of the proposedtechniques are compared with four machine learning classification algorithms: Bayes net, support vector machine, randomforest, and decision tree. The results show that the proposed rule-based classifier outperforms the four algorithms byachieving 82% accuracy, whereas TSS classification achieves 77% accuracy