An Improved Algorithm for Semantic Similarity Based on HowNet

This thesis puts forward an improved algorithm for semantic similarity based on HowNet. This method is built on the sememe tree, and continues to use the thought of calculation of semantic similarity based on the semantic distance, and has made two improvements in this idea. Firstly, the influence of the children's node density under the common parent node on semantic similarity is increased, and the problem of parallel similarity between words with different information densities is corrected. Secondly, the similarity calculation between words of polysemy terms is reconsidered. The problem of the existence of different word concepts between polysemy terms, through a comprehensive calculation of semantic terms, has been corrected. The methods put forward in the thesis can also improve the situation of improper classification of HowNet words and the imperfection of its sememet. After experimental tests, it is a method with high efficiency and practicality.