A Pheromone-Like Model for Semantic Context Extraction from Collaborative Networks

Abstract

The extraction of semantic contexts is a relevant issue in information retrieval to provide high quality query results. This paper introduces the semantic context underlying a set of given input concepts as defined by the relevant multiple explanation paths connecting the input concepts in a collaborative network. A pheromone-like model based on this approach is introduced for the detection and the extraction of multiple paths of explanation between seed concepts. The exploration of the online collaborative network of explanation uses a heuristic driven random walk, based on semantic proximity measures. Random walks distribute pheromone on the traversed arcs used to evaluate the relevance of concepts in the multiple explanatory paths to be extracted. Experimental results obtained on accepted datasets and contexts extracted from the Wikipedia collaborative network show that the proposed algorithm can extract contexts with high relevance degree, which outperforms other methods. The approach has a general applicability and can be extended to other explanation-based online collaborative networks.