Neural models for information retrieval : towards asymmetry sensitive approaches based on attention models
Texte intégral
Documents relatifs
Manual lexical expansion (step 3) At the previ- ous step, the K closest terms provided by the word embedding model were considered as seed term synonyms by default.. At this step,
Although the current evaluation practice works well with the publicly available data collections, it cannot support the evaluation over proprietary data, which can not be easily
The second approach uses the multinomial Naïve Bayes classifier in order to detect sexual predators.. The first approach performed better than the second one with low percentages
Lexical features are extracted from all the legal documents and the simi- larity between each current case document and all the prior case documents are determined using
The classification task and retrieval task are developed based on the distributional representation of the text by utilizing term - document matrix and non-negative
and Carlson, R.W., "Investigation of Causes of Delayed Expansion of Concrete in Buck Hydroelectric Plant," American Concrete Institute Journal, Proc.. Canadian
The second component estimates, given a query, a score for each document and for each modality (textual and visual). Finally, the last component combines linearly the score obtained
Our approach, based on Fisher Linear Discriminant Analysis, aims to learn these weights for multimedia documents composed of text and images.. Text and images are both represented