INFORMATION RETRIEVAL USING NEURAL NETWORKS VINEETH REDDY ANUGU - - PowerPoint PPT Presentation
INFORMATION RETRIEVAL USING NEURAL NETWORKS VINEETH REDDY ANUGU CMSC 676 INFORMATION RETRIEVAL INTRODUCTION How can we use neural networks for Information Retrieval? Neural Networks is a field of machine learning How can we train a
VINEETH REDDY ANUGU CMSC 676 INFORMATION RETRIEVAL
ØHow can we use neural networks for Information Retrieval? ØNeural Networks is a field of machine learning ØHow can we train a machine learning model and use that for specific application of information retrieval? ØThis process is highly data demanding
ØNeural networks learn from raw data ØSupervised and Unsupervised learning ØCNNs are set of filters that extract patterns in data ØUsed for image recognition, can be used on textual data ØCNNs are feed-forward networks
ØUsed for ad-hoc retrieval by relevance matching ØConvolutional Neural Network is used in this model ØDRMM is an interaction-based model ØInteraction-based models looks for interaction between queries and terms ØRequires Matching Histogram Mapping as input ØInvolves a term gating network
ØUsed to learn embeddings of given data ØEmbeddings are vector representations of data ØUseful in recommendation systems, embeddings are drawn from user data ØTwo models
Ø Skip-gram Model Ø Continuous bag-of-words Model(CBOW)
ØTwo datasets, Robust04 and ClueWeb-09-Cat-B, are used. ØComparing DRMM with well established IR Models
ØComparing the Word2Vec model with the FastText model ØFastText obtains n-grams instead of embeddings ØUseful for rare and out-of-vocabulary word embeddings ØUses differ based on the task