A Case Study of Closed-Domain Response Suggestion with Limited Training Data

Published in Database and Expert Systems Applications, 2018

Recommended citation: Lukas Galke, Gunnar Gerstenkorn, Ansgar Scherp, "A Case Study of Closed-Domain Response Suggestion with Limited Training Data." Database and Expert Systems Applications, 2018. https://doi.org/10.1007/978-3-319-99133-7_18

Access paper here

Code available here

Workshop paper

Abstract: We analyze the problem of response suggestion in a closed domain along a real-world scenario of a digital library. We present a text-processing pipeline to generate question-answer pairs from chat transcripts. On this limited amount of training data, we compare retrieval-based, conditioned-generation, and dedicated representation learning approaches for response suggestion. Our results show that retrieval-based methods that strive to find similar, known contexts are preferable over parametric approaches from the conditioned-generation family, when the training data is limited. We, however, identify a specific representation learning approach that is competitive to the retrieval-based approaches despite the training data limitation.

@inproceedings{DBLP:conf/dexaw/GalkeGS18,
  author    = {Lukas Galke and
               Gunnar Gerstenkorn and
               Ansgar Scherp},
  title     = {A Case Study of Closed-Domain Response Suggestion with Limited Training Data},
  booktitle = { {DEXA} Workshops},
  series    = {Communications in Computer and Information Science},
  volume    = {903},
  pages     = {218--229},
  publisher = {Springer},
  year      = {2018}
}