(2nd-April-2020)
Table 3 we can observe that the categorization part of the knowledge recording module can achieve 84.6% accuracy. Some of the important points to consider were the following: Input voice should be in good quality so as to properly perform voice recognition and ultimately to categorize it. The background noise should be at minimum level. In addition to YAGO2s we also utilized domain specific knowledgebase for improving categorization accuracy.
Next we tested the Semantic Similarity part of the Knowledge Retrieval module. We tried to estimate the similarity score between query text and speech text.
The results of the evaluation for semantic similarity are shown in Table 4. From Table 4 we can observe that the semantic similarity part of the Knowledge Retrieval module has an accuracy of about 46.2%. We need to revise the semantic similarity algorithm for improving the accuracy. This can be one of the important future tasks. Following can be some of the important points to consider:
Comments