Using semantic similarity for identifying relevant page numbers for indexed term of textual book

Back-of-book index page is one of navigation tools for reader. It helps reader to immediately jump to a page that contains relevant information regarding a specific term. It helps reader to retrieve information about specific topics in mind without having to read the complete book. Indexed terms are usually determined by author based on one’s subjective preference on what indications should be used to decide whether a term should be indexed and what pages are relevant. Therefore, indexing a book inherits subjectivity of author side. The book size is proportional to the indexing effort and consistency. This leads to the fact that page numbers are not always referred to relevant pages. This paper proposes an approach to identify relevancy of a page that contains an indexed term. This approach measures the semantic relation between indexed term with the respective sentence in the page. To measure the semantic relation, the approach utilizes semantic distance algorithm that based on Wordnet thesaurus. We measure the reliability of our system by measuring its degree of agreement with the book indexer using kappa statistics. The experimental result shows that the proposed approach are considered as good as the domain expert, given average kappa value 0.6034.