Knowledge-Enhanced Multi-semantic Fusion for Concept Similarity Measurement in Continuous Vector Space

Semantic similarity measurement is one of the significant components in the fields of information retrieval, artificial intelligence, etc. Also, it has been used in Internet of Things (IoT) to resolve the problems of inter-operability and integration of the heterogeneous resources such as ubiquitous services and distributed data, while increasing the accuracy of semantic similarity measurement is still a challenge. In this work, we propose a novel Multiple Semantic Fusion (MSF) model where various semantic properties and relationship information from knowledge base are encoded into the continuous distributed vectors derived from corpus. To capture and understand latent semantics of words, this model retrofits the semantic features in distributed representations and carries out a sense vector per word. On basis of the MSF model, we evaluate the semantic similarity between concepts by calculating the cosine similarity of the corresponding sense vectors. Our evaluation has been validated against different similarity methods in terms of Pearson correlation coefficient between computational results with human judgments. Finally, empirical evaluation results and discussions on various benchmark datasets are provided. Our finding provides a new perspective to understand and evaluated the word semantics.