Neurofuzzy Modeling of Context–Contingent Proximity Relations

Abstract

The notion of proximity is one of the foundational elements in humans' understanding and reasoning of the geographical environments. The perception and cognition of distances plays a significant role in many daily human activities. Yet, few studies have thus far provided context–contingent translation mechanisms between linguistic proximity descriptors (e.g., “near,”“far”) and metric distance measures. One problem with previous fuzzy logic proximity modeling studies is that they presume the form of the fuzzy membership functions of proximity relations. Another problem is that previous studies have fundamental weaknesses in considering context factors in proximity models. We argue that statistical approaches are ill suited to proximity modeling because of the inherently fuzzy nature of the relations between linguistic and metric distance measures. In this study, we propose a neurofuzzy system approach to solve this problem. The approach allows for the dynamic construction of context–contingent proximity models based on sample data. An empirical case study with human subject survey data is carried out to test the validity of the approach and to compare it with the previous statistical approach. Interpretation and prediction accuracy of the empirical study are discussed.

An Intelligent Spatial Proximity System Using Neurofuzzy Classifiers and Contextual Information

“…Therefore, implementing this work in a generic GIS is quite challenging. (Yao and Thill, 2007) proposed an innovative solution to reason about proximity by using contextual information to handle the cognitive aspect and neurofuzzy techniques to handle the qualitative aspect of spatial proximity. Although Yao and Thill's approach may be suitable for qualitative spatial reasoning, it suffers from several drawbacks.…”

“…Reasoning with spatial proximity is a research area which has been addressed by the qualitative spatial reasoning community, adopting different perspectives such as geography, cognitive science, linguistics and others (Yao and Thill, 2007). A large number of prior works used fuzzy logic and qualitative techniques to deal with spatial proximity because it has inherent fuzziness (Robinson, 1990).…”

“…Yao and Thill used ANFIS (Adaptive NeuroFuzzy Inference System) to implement and validate their approach. ANFIS is a TakagiSugeno fuzzy inference system (Yao and Thill, 2007). It uses a back propagation algorithm to train the fuzzy neural network.…”

“…ISPRS Technical Commission II Symposium, 6 -8 October 2014, Toronto, Canada (Yao and Thill, 2007) The contextual factors used in Yao and Thill's approach (objective and subjective) take the form of crisp inputs. They are fuzzyfied using membership functions (which can be predetermined by an expert).…”

Interpreting the Fuzzy Semantics of Natural-Language Spatial Relation Terms with the Fuzzy Random Forest Algorithm

“…Models for fuzzifying the 9-intersection model [22,23] and RCC [24,25] have been also proposed. However, for handling NLSR terms only a handful of fuzzy models were developed, and these models can handle only a limited number of NLSR terms, such as proximity [26,27], along, surround, overlap and disjoint [28], and north and south [29]. As such, they are not robust in differentiating large number of NLSR terms.…”