The Influence of Thresholding Strategy on Multi-label Topic Identification

Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F15%3A43927059" target="_blank" >RIV/49777513:23520/15:43927059 - isvavai.cz</a>
Výsledek na webu
<a href="http://hdl.handle.net/11025/21317" target="_blank" >http://hdl.handle.net/11025/21317</a>
DOI - Digital Object Identifier
—

Jazyk výsledku
angličtina
Název v původním jazyce
The Influence of Thresholding Strategy on Multi-label Topic Identification
Popis výsledku v původním jazyce
In this paper, we are demonstrating the influence of described thresholding strategies on a multi-label topic identification of Czech news articles. Results suggested, that we can improve (in terms of F1 score) the performance of multli-label topic identification on our data by 4.58% relatively when using label-wise thresholding and roughly the same when using sample-wise thresholding. An interesting result is that in the case of sample-wise thresholding, the performance can be further improved just by sorting scores on the input of the trained regressor.
Název v anglickém jazyce
The Influence of Thresholding Strategy on Multi-label Topic Identification
Popis výsledku anglicky
In this paper, we are demonstrating the influence of described thresholding strategies on a multi-label topic identification of Czech news articles. Results suggested, that we can improve (in terms of F1 score) the performance of multli-label topic identification on our data by 4.58% relatively when using label-wise thresholding and roughly the same when using sample-wise thresholding. An interesting result is that in the case of sample-wise thresholding, the performance can be further improved just by sorting scores on the input of the trained regressor.

Rok uplatnění
2015
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Podobné výsledky(10)