Semantic Similarities in Voice Information Retrieval System for Documents

Abstract

Recent advances in text to speech and vice versa have opened a new dimension to the manner via which information is sought on the go. It is becoming a common place for smart device users to search by voice. This development is driving a new order in Information Retrieval circle with researcher trying to improve on the design of information retrieval systems operated by voice. Techniques like the spoken text detected, word transcription, phonetic query expansion in voice information retrieval system for document tend not to retrieve relevant documents to users speech data due to the problems of polysemy, query drift, low precision and low recall values engendered from clustered document. Therefore, this research extends the conceptual query model by using the new concept (term) generated by the Universal Fuzzy Concept Network Language with Latent Semantic Analysis technique to build semantic similarity between concepts in a document before retrieval. Also, the research find the degree of relational relevance concepts in potential retrievable document. Here, the research achieved a higher degree of cohesion, lower degree of entropy between concepts and better semantic similarity of concepts in potential retrievable documents to the user query. When compared to keyword spotting and query expansion technique, a better precision and recall values was also achieved by using the proposed method.