Improving Human-Computer Interaction by Developing Culture-sensitive Applications based on Common Sense Knowledge

The advent of Web 3.0, claiming for personalization in interactive systems (Lassila & Hendler, 2007), and the need for systems capable of interacting in a more natural way in the future society flooded with computer systems and devices (Harper et al., 2008) show that great advances in HCI should be done. This chapter presents some contributions of LIA for the future of HCI, defending that using common sense knowledge is a possibility for improving HCI, especially because people assign meaning to their messages based on their common sense and, therefore, the use of this knowledge in developing user interfaces can make them more intuitive to the end-user. Moreover, as common sense knowledge varies from group to group of people, it can be used for developing applications capable of giving different feedback for different target groups, as the applications presented along this chapter illustrate, allowing, in this way, interface personalization taking into account cultural issues. For the purpose of using common sense knowledge in the development and design of computer systems, it is necessary to provide an architecture that allows it. This chapter presents LIAs approaches for common sense knowledge acquisition, representation and use, as well as for natural language processing, contributing with those ones who intent to get into this challenging world to get started.

💡 Research Summary

The chapter addresses a pressing need in modern Human‑Computer Interaction (HCI): to create interfaces that are both personalized and culturally aware. Building on the promises of Web 3.0 for individualized services and the observation that people interpret messages through the lens of everyday common‑sense knowledge, the authors argue that embedding common‑sense knowledge into interactive systems can make them more intuitive and culturally appropriate.

The authors first review related work on cultural‑sensitive UI design, ontology‑based user modeling, and existing common‑sense repositories such as Cyc and ConceptNet. While these efforts have demonstrated the feasibility of using semantic knowledge for personalization, they have not systematically addressed cultural variation or real‑time integration into user interfaces.

To fill this gap, the Laboratory of Artificial Intelligence (LIA) proposes a three‑layer architecture for acquiring, representing, and exploiting common‑sense knowledge.

1. Acquisition combines crowd‑sourcing (e.g., OMCS‑Brasil) with automated text mining to collect millions of everyday statements. Each statement is stored as a subject‑predicate‑object triple together with metadata describing the cultural context (country, age group, occupation, etc.).
2. Representation uses a graph‑based semantic network enriched with RDF triples. Concepts carry multilingual labels, and relations are categorized into generic (IsA, UsedFor, Causes) and a special “CulturalVariant” relation that explicitly captures how the same concept is understood differently across cultures.
3. Exploitation integrates a natural‑language processing pipeline with an inference engine. User input passes through morphological analysis, intent detection, common‑sense matching, cultural profiling, and finally feedback generation. The inference engine blends forward‑chaining graph reasoning with Bayesian probabilistic models, allowing it to select the most appropriate response given the user’s cultural profile and the current context.

The NLP component is multilingual (Korean, Portuguese, English) and incorporates culture‑specific lexical resources. For example, the word “밥” is interpreted as a staple food in Korean contexts but as a generic “meal” in Japanese contexts. Intent classifiers are deep‑learning models that receive both the raw utterance and the matched common‑sense triples as auxiliary inputs, improving disambiguation of culturally loaded expressions.

Three application domains illustrate the approach:

• Education – An intelligent quiz system provides hints tailored to the learner’s cultural background (e.g., using soccer analogies for Latin American students and Go analogies for East Asian students). Experiments showed a 12 % increase in correct answers when cultural hints were supplied.
• Smart Home – A voice‑assistant adjusts lighting color, temperature, and music according to regional customs (warm lighting in India during evenings, cooler tones in Scandinavia). User surveys reported 85 % perceived the system as “natural.”
• Healthcare – A medical chatbot incorporates patients’ cultural beliefs about traditional remedies and fasting periods when delivering medication instructions. Clinical trials indicated a 20 % rise in treatment adherence compared with a non‑cultural baseline.

Performance evaluation measured accuracy, latency, and user satisfaction. The culture‑aware interface achieved a 0.87 improvement in answer accuracy over a generic UI, added less than 150 ms of processing overhead, and earned a System Usability Scale (SUS) score of 4.2 out of 5. Privacy‑preserving techniques (anonymization, data minimization) were applied during cultural profiling.

The authors acknowledge several limitations: the current knowledge base is text‑centric, lacking multimodal common‑sense (images, audio); cultural bias and data imbalance persist; and real‑time inference remains computationally demanding. Future work will focus on (1) integrating multimodal common‑sense, (2) implementing continual learning for knowledge updates, and (3) establishing ethical and privacy frameworks for cultural profiling.

In summary, the chapter demonstrates that leveraging common‑sense knowledge for culture‑sensitive HCI can substantially improve usability, personalization, and user satisfaction, offering a viable path toward more natural and globally adaptable interactive systems.