Please enable JavaScript.
Coggle requires JavaScript to display documents.
Interpreting Similarity of Texts Based on
Automated Chunking, Chunk…
Interpreting Similarity of Texts Based on
Automated Chunking, Chunk Alignment and Semantic Relation Prediction
Banjade et al. 2016
-
Methods (supervised)
Chunk alignment system
built upon a previous system called NeRoSim
(Banjade et al., 2015)
-
-
-
-
Results
System chunks
In image and headlines data, our system obtained the
best results.
the F1 scores for alignments are high but the scores for predicting the alignment types are relatively lower.
Gold chunks
alignment scores are higher compared to the baseline system and are very close to the best results
from all submissions in those categories
the alignment type score in each case is relatively lower than the alignment-only score and itultimately impacted the F1 score calculated for type and score together (i.e. T+S)
Room for improvement
It indicates that the systems overall performance will be improved greatly if improvements can be made in predicting the alignment types.
the scores for student-answers are
lower than headlines and image texts and it requires
further analysis to fully understand why this is the
case. One of the reasons might be that we did not use
this dataset while developing the system and no prior
information about such dataset was modeled