Hybrid Model with Word2vector in Information Retrieval Ranking
Shweta Pandey, Iti Mathur, Nisheeth Joshi · Lecture notes on data engineering and communications technologies · 2021
People have realized the importance of finding and archiving information with the computer advents for thousands of years, and storing of large amount of information became possible. It is actually not related to the fetching of the documents, it informs the user on the whereabouts and existence of the documents. In this paper, hybrid model has been used in which the document is classified using the support vector machine (SVM) classifier, and after the condition is applied, if it is satisfied, the extraction of the matched paragraph and the sentence is responsible for the generation of relevant answer. The knowledge base gets updated if condition does not match, and new updated answer will be generated. Finally, the best answer is displayed after ranking by using the PSO optimization. Word2vector is applied for feature extraction. In this paper, comparison of RankSVM, RankPSO and RankHSVM + PSO for the implementation of IR ranking is considered. Here, first SVM is used as a classifier for dividing most relevant and non-relevant results, and afterward PSO is used for the optimization of the result means extraction of the best answer or document. Selection of appropriate parameters is difficult in case of simple SVM, but for the ranking of the answers it gives potential solutions. PSO is used for optimization which has global search capability and is easy to implement and thus to optimize the ranking of document retrieval. We propose the RankHSVM + PSO model to find the fitness function. This technique improves the performance of the system as comparative to other techniques. The result shows that the algorithm applied here improves the value of performance evaluation by 4–5%. TREC 2004 QA DATA dataset is used which contains my datasets. It has a question answering track since 1999. The task was defined in each track. Retrieval of true equivalent test collection for standard retrieval is an open problem. In a retrieval test collection, the unit that is judged the document has a unique identifier.