CAUSAL RELATIONSHIP IDENTIFICATION SYSTEM ON INDONESIAN TEXT PROCESSING USING SEMANTIC TEMPLATE APPROACH

• Main Author: Bagas Bhaskoro, Susetyo
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40514
• Institution: Institut Teknologi Bandung

The objective of natural language processing is to transform complex information into brief information, comprehensible and able to be used as an analysis reference for certain domain, i.e. health domain. Information that frequently utilized on health domain is an information about causality or cause-effect which is a knowledge that is built on particular facts that have <cause> meaning and is concluded as general facts that have <effect> meaning, and between those two meanings have a relationship that explains each other. If causal relationship is being formulated, it turns into function , where is <cause> meaning, and is <effect> meaning, so it will be . The issue is not all of <cause> and <effect> meanings are explained in simple sentence ot single sentences, but there is also possibility for <cause> and <effect> meanings to be explained in complex sentence or multiple sentences that the position itself is at the beginning of sentence or at the end of sentence. Furthermore, <cause> and <effect> meanings are not always be abstracted explicitly but also implicitly. Research's aim is to generate semantic template method, which is method to identify <cause> and <effect> meanings on simple sentence and complex sentence. Current developed of semantic template method is focusing on template using for identify causal relationship that has no dependency on certain topic, but variety of textual topic. Proposal contained in semantic template is considering <feature selection>, <word position>, <word meaning>, <frequency of word occurence>, <root dynamic> and use of <template rule>. This research using dataset from collection of causal relationship sentences found in some publications and online articles as many as 1.017 sentences for training and testing. Identification system performance testing is divided into two evaluation cateogries, extrinsic and intrinsic testing. Performance evaluation results of semantic template method on single sentences are 0,874; 0,747 and 0,803, meanwhile results on multiple sentences are 0,877; 0,815 and 0,845 for each recall, precision and f-measure, meanwhile error rate is 0.280. Evaluation result for information accuracy quality is 0.719, meanwhile for precision is 0.781 and for completeness is 0.953. Generated semantic template method that has been successfully implemented on health domain is public health surveillance system. During implementation, there are several proposals have been generated, i.e. natural language transformation into medical element annotation (LPpAJSFPePnGOD) to identify causal relationship pattern, paragraph annotation pattern to classify sentence position on medical article paragraph and build a semantic relationship pattern for semantic annotation.