H-LSTM Framework for Temporal Information Retrieval in Code-Mixed Social Media Text

摘要

The contents of social media consists of mixed script information and from these type of sources one can extract temporal information. The temporal contents are difficult to be processed by any machine. The challenge here is to retrieve temporal domain information used by the users for expressing their interests on any issue. The work outlines the comparative study of different approaches of temporal data retrieval in transliterated domain. A novel handcrafted rule based technique is applied which accepts input in mixed script format and on the basis of illustrated rules, the system is suppose to extract temporal expressions from the sentence. The evaluation measures use rule-based approach validations along with the evaluations based on statistical measures. To further validate the results a voting technique is applied that selects the most valid and suitable temporal tag for that word. The experimental results suggests that the voting method performs applied here gives the best result in context to accurate temporal tagging.