Identification of Relevance and Support for Consumer Health Information

Abstract

With a rapid growth of queries posted for the search on internet that were related to medical information raised the need of acquiring the right and relevant information related to those queries. The information systems available at present are providing the documents that matches the user query, but to check whether they really matches the queries is becoming a difficult task for the layman. This is because the consumer does not have any knowledge related to Medical records and its nomenclature. Consumer Health Information Search (CHIS) is a track organized to cater the need of medical information search. As a part of this track two tasks were designed. (1) Given a query and a document containing a set of sentences, the task is to identify whether the sentence selected is relevant/irrelevant to the query posted. (2) To identify whether the sentence selected from the document is supporting the query or opposing the query or is in neutral state. The solution of Task_1 is achieved by selecting the similarity scores as a feature. The mean of this similarity scores were computed to identify the relevant nature of the sentence to the query. Task 2 is viewed as a multiple classification problem and is solved by making use of C-Support Vector Machine Classifier. The model was tested on data set provided by the CHIS track organizers. The results obtained by our model that was designed for task_1 were not to the satisfactory level when compared to the results of other track participants. For Task 2 C-support vector machine model is applied. In that Tf-idf score is used as a feature for the model. Results obtained for task_2 obtained the highest accuracy scores when compared with the other models submitted by different participants of the track.