Article Structure
Abstract
Dialog topic tracking aims at analyzing and maintaining topic transitions in ongoing dialogs.
Introduction
Human communications in real world situations interlace multiple topics which are related to each other in conversational contexts.
Dialog Topic Tracking
Dialog topic tracking can be considered as a classification problem to detect topic transitions.
Wikipedia-based Composite Kernel for Dialog Topic Tracking
The classifier f can be built on the training examples annotated with topic labels using supervised machine learning techniques.
Evaluation
To demonstrate the effectiveness of our proposed kernel method for dialog topic tracking, we performed experiments on the Singapore tour guide dialogs which consists of 35 dialog sessions collected from real human-human mixed initiative conversations related to Singapore between guides
Conclusions
This paper presented a composite kernel approach for dialog topic tracking.
Topics
tree kernel
- Our composite kernel consists of a history sequence and a domain context tree kernels , both of which are composed based on similar textual units in Wikipedia articles to a given dialog context.Page 1, “Introduction”
- Our composite kernel consists of two different kernels: a history sequence kernel and a domain context tree kernel .Page 2, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- 3.2 Domain Context Tree KernelPage 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- Since this constructed tree structure represents semantic, discourse, and structural information extracted from the similar Wikipedia paragraphs to each given instance, we can explore these more enriched features to build the topic tracking model using a subset tree kernel (Collins and Duffy, 2002) which computes the similarity between each pair of trees in the feature space as follows:Page 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- In this work, a composite kernel is defined by combining the individual kernels including history sequence and domain context tree kernels , as well asPage 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- the domain context tree kernel contributed to produce more precise outputs.Page 4, “Evaluation”
See all papers in Proc. ACL 2014 that mention tree kernel.
See all papers in Proc. ACL that mention tree kernel.
Back to top.
feature space
- Since our hypothesis is that the more similar the dialog histories of the two inputs are, the more similar aspects of topic transtions occur for them, we propose a subsequence kernel (Lodhi et al., 2002) to map the data into a new feature space defined based on the similarity of each pair of history sequences as follows:Page 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- The other kernel incorporates more various types of domain knowledge obtained from Wikipedia into the feature space .Page 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
- Since this constructed tree structure represents semantic, discourse, and structural information extracted from the similar Wikipedia paragraphs to each given instance, we can explore these more enriched features to build the topic tracking model using a subset tree kernel (Collins and Duffy, 2002) which computes the similarity between each pair of trees in the feature space as follows:Page 3, “Wikipedia-based Composite Kernel for Dialog Topic Tracking”
See all papers in Proc. ACL 2014 that mention feature space.
See all papers in Proc. ACL that mention feature space.
Back to top.
manually annotated
- All the recorded dialogs with the total length of 21 hours were manually transcribed, then these transcribed dialogs with 19,651 utterances were manually annotated with the following nine topic categories: Opening, Closing, Itinerary, Accommodation, Attraction, Food, Transportation, Shopping, and Other.Page 4, “Evaluation”
- For the linear kernel baseline, we used the following features: n-gram words, previous system actions, and current user acts which were manually annotated .Page 4, “Evaluation”
- All the evaluations were done in fivefold cross validation to the manual annotations with two different metrics: one is accuracy of the predicted topic label for every turn, and the other is precisiorflrecall/F-measure for each event of topic transition occurred either in the answer or the predicted result.Page 4, “Evaluation”
See all papers in Proc. ACL 2014 that mention manually annotated.
See all papers in Proc. ACL that mention manually annotated.
Back to top.
