A Clustering Study for the Optimization of Emotional Information Retrieval Systems: DBSCAN vs K-means

Information retrieval systems enable the organization, storage, retrieval and distribution of information. Their main purpose is to retrieve information and return results that match the most the user's needs. The field of Information Retrieval (IR) has seen in recent years the appearance of a new axis which is Personalized Information Retrieval (PIR), in particular emotional PIR. This type of information search systems aims to extract the user's emotions and to include it in the information search process, in order to return results according to the user's sentiments and emotional state expressed through his queries. These emotion-based systems have shown interesting results in terms of recall and precision, however the search time for relevant information increased significantly compared to non-emotion-based systems, due to the extra treatments necessary for emotion extraction from the input queries, as well as the emotional profile construction operations. We propose in this work an optimization approach for emotion-based information retrieval systems, based on data mining techniques, more precisely clustering. We aim to reduce the execution time by grouping the documents of the system in clusters, thus transiting from multiple query/document comparing operations to fewer query/cluster operations while keeping quality of results similar to the original non-clustered system.