Question Bank Similarity Searching System (QB3S) Using NLP and Information Retrieval Technique

Abstract

Problem Based e-learning(PBeL) in bangla language is one of the most progressing areas of the use of ICT in education. Question Bank(QB) is the main component of any PBeL system. Searching similarity in the complex structure of QB is a challenging task in the development of PBeL system. We have been developed an efficient Question Bank Similarity Searching System(QB3S) to find similar questions, handle duplicate question and rank search result of a query input based on NLP and Information Retrieval techniques. QB3S has four modules: bangla documents processing, question structure analysis and clustered indexing by B+ tree , word-net construction and Information retrieval module. Lexical analysis, stemming by finite automata rules and stopwords removing have been used for bangla document processing. The most challenging procedures of QB3S were Analyzing the structure of data for clustered indexing in the sorted sequential file of the QB database with a B+ tree data structure and improved TF-IDF algorithm with weighted functionality. A Word-net has been used for handling synonyms. Vector Space Model(VSM) has been designed from the value of TF-IDF weighted matrix. By using cosine similarity product rule, we have been Calculated the similarity value between the query input and all mcq of DB from VSM. QB3S has been evaluated in some experimental dataset to find results by imposing different test cases. The accuracy of searching performance which has found to be satisfactory.