Journal of Data Mining and Knowledge Engineering ISSN: 3139-5341 (Online)

Abstract

Social media generates massive amount of multilingual real time data. On Many occasion, Social media have proved that it can be fastest media to break event compare to any other traditional news media e.g. News channel, Internet news portal. In this work, the identification and the normalization of temporal information have been done on tweets. Each tweet can represent or describe event as opposed to previous work which rely of bursty keyword detection. In the Identification of temporal entity, three approach: Conditional Random Field, Support Vector Machine and Rule base approach used. In the Normalization of temporal entity, Rule base approach used. The rule base approach proceeds texts as inputs and from various rules, it identify temporal information. The conditional approach is trained using tagged data and then using generated model tagging on test dataset done. Morphological, Syntactic and gazetteers features are defined in Conditional random field. The use of chunking in Conditional Random Filed is not effect in identification of temporal entity. In support vector machine approach, training is completed on tagged dataset and then using generated model file temporal entity tagged. The result on Hindi temporal named entity identification terms of Precision 0.88, Recall 0.81, F1 - measure 0.85 for rule base approach; Precision 0.93, Recall 0.74, F1 - measure 0.83 for conditional random field; Precision 0.73, Recall 0.69, F1 - measure 0.71 for support vector machine and its normalization in terms of Precision 0.85, Recall 0.80 and F1 measure 0.83. Rule base approach gives good result compare to other two approach for temporal entity identification. Temporal expression identification can be helpful for generation tweet calendar, Question/Answering over social media and event summarization.

Keywords: Hindi temporal, temporal expression identification