Design of an Expert System for Early Detection of Domestic Violence Using Keyword Matching, Sentiment Analysis and Forward Chaining

Abstract

Domestic Violence (KDRT) is a critical humanitarian issue where victims often under-report due to fear, dependence, and stigma. Consequently, many victims turn to social media to express distress implicitly using vague language, rendering existing passive reporting systems and manual detection ineffective against unstructured narratives. This research aims to design a Hybrid Expert System architecture that integrates Keyword Matching and Sentiment Analysis with Forward Chaining to objectively detect indications of KDRT in Indonesian text, specifically targeting implicit venting that lacks explicit violence keywords. The study employs a systematic development method involving knowledge acquisition from psychological (cycle of abuse) and legal  domains to construct a robust knowledge base. The technical architecture combines sentiment analysis to gauge emotional intensity with Forward Chaining inference logic. This logic utilizes dynamic frequency parameters to validate findings through case tracing simulations. The results demonstrate that the proposed architecture successfully classifies various violence types, including physical, verbal, economic, and multi-type violence. The simulation confirms the system’s capability to distinguish between common household conflicts and specific abuse patterns by applying zero-tolerance thresholds for acute violence and repetition filters for chronic psychological abuse. Consequently, this system functions as a robust decision support tool, providing measurable risk assessments and appropriate intervention recommendations for early detection.Domestic Violence (KDRT) is a critical humanitarian issue where victims often under-report due to fear, dependence, and stigma. Consequently, many victims turn to social media to express distress implicitly using vague language, rendering existing passive reporting systems and manual detection ineffective against unstructured narratives. This research aims to design a Hybrid Expert System architecture that integrates Keyword Matching and Sentiment Analysis with Forward Chaining to objectively detect indications of KDRT in Indonesian text, specifically targeting implicit venting that lacks explicit violence keywords. The study employs a systematic development method involving knowledge acquisition from psychological (cycle of abuse) and legal  domains to construct a robust knowledge base. The technical architecture combines sentiment analysis to gauge emotional intensity with Forward Chaining inference logic. This logic utilizes dynamic frequency parameters to validate findings through case tracing simulations. The results demonstrate that the proposed architecture successfully classifies various violence types, including physical, verbal, economic, and multi-type violence. The simulation confirms the system’s capability to distinguish between common household conflicts and specific abuse patterns by applying zero-tolerance thresholds for acute violence and repetition filters for chronic psychological abuse. Consequently, this system functions as a robust decision support tool, providing measurable risk assessments and appropriate intervention recommendations for early detection.