Title :
Bootstrap sequential projection multi kernel Locality Sensitive Hashing
Author :
Mehta, Harsham ; Garg, Deepak
Author_Institution :
Dept. of Comput. Sci. & Eng., Thapar Univ., Patiala, India
Abstract :
In Recommender system we have similarity search as a key part for making efficient recommendations. Similarity search have always been a tough task in a high dimensional space. Locality Sensitive Hashing which is most suitable for extracting data in a high dimensional data (Multimedia data). The Idea of locality sensitive hashing is that it decreases the high dimensional data to low dimensions using distance functions and then store this data using hash functions which ensures that distant data is placed much further. This technique has been extended to kernelized Locality sensitive hashing (KLSH). One limitation of regular LSH is they require vector representation of data explicitly. This limitation is addressed by kernel functions. Kernel functions are capable of capturing similarity between data points. KLSH is a breakthrough in content based systems. This method takes a kernel function, a high dimensional database for data inputs and size of hash functions to be built. These kernel functions that are being used may give different degree of result precision. Hence we try to combine these kernels with a bootstrap approach to give an optimal result precision. In this paper we present the related work that has been done in locality sensitive hashing and at the end we propose algorithms for data preprocessing and query evaluation.
Keywords :
computer bootstrapping; data handling; data structures; recommender systems; storage management; vectors; KLSH; bootstrap sequential projection multikernel locality sensitive hashing; content based systems; data preprocessing; data storage; distance functions; hash functions; high dimensional data; kernel functions; kernelized locality sensitive hashing; multimedia data; query evaluation; recommender system; similarity search; vector data representation; Approximation algorithms; Approximation methods; Boosting; Complexity theory; Indexing; Kernel; Vectors;
Conference_Titel :
Advances in Computing, Communications and Informatics (ICACCI, 2014 International Conference on
Conference_Location :
New Delhi
Print_ISBN :
978-1-4799-3078-4
DOI :
10.1109/ICACCI.2014.6968294