| 生田 和也 | 1651006: M, 2回目発表 | 知能コミュニケーション 中村 哲, 松本 裕治, 吉野 幸一郎, 鈴木 優 |
|
title: Control of statistical language generation based on contents vector
abstract: Information presentation in natural language has a potential to improve the usability of information guidance system, thus, sentence generation module is an important. Existing sentence generation methods, the contents of generated sentences are controlled by templates, however, it is hard to generate a variety of sentences by templates. It contrast, is reported that neural language models has a potential to generate natural and various sentences. In this research, we aim to build a language generation system that can generate a sentence according to a given contents vector, 1-hot vector of contents to be included in the generated sentence. We built a model of generation and will improve the model to represent a variety of information sources. language of the presentation: Japanese | ||
| 河野 誠也 | 1651038: M, 2回目発表 | 知能コミュニケーション 中村 哲, 松本 裕治, 吉野 幸一郎, 鈴木 優 |
|
title: Modeling Reference Interview toward Information Retrieval Dialogue System
abstract: Requests of users for information retrieval systems are often ambiguous and this property makes it difficult to provide the exact information that is related to the real demand of the user in the information seeking process. It is known that clarifications of the user's demand with some system actions such as "confirmation" or "asking about background information" help to find really demanded information. The reference interview, a chat-style information seeking dialogue in reference service at libraries, is an example of information seeking dialogue with these clarification manners. Conducting this reference interview in advance, the accuracy of information provision in the reference service is improved. In this study, we focus on the reference interview for modeling the librarian behaviors to realize a system that can provide information through interactions, even if the intention of the user at the first utterance is ambiguous. In this presentation, I will introduce the background, some progress and research plan. language of the presentation: Japanese | ||
| 村瀬 行俊 | 1651108: M, 2回目発表 | 知能コミュニケーション 中村 哲, 松本 裕治, 吉野 幸一郎 |
|
title: Feature Inference Based on Label Propagation on Wikidata Graph for Dialog State Tracker
abstract: Dialog Sate Tracking (DST) is one of the most important subtasks for task-oriented dialog systems. The module of DST predicts machine readable state of dialog from a current user utterance for the dialog manager. Last few years, Dialog State Tracking Challenge (DSTC) has been held as an annual shared task for dialog state tracking. In DSTC4, the focus of the challenge moves to tracking of state frames in English human-human conversation, which is more complicated than human-machine conversation. DSTC5 provides Chinese dialogs and translations to build a cross-language dialog state tracker. We proposed a method that creates inferred knowledge graph feature to improve the scores by using cross-lingual general knowledge graph such as Wikidata. We applied the method to DSTC4 & 5 corpora and improved the scores based on the baseline method. In the future work, we will implement the state-of-the-art method and combine our method to achieve the best score. language of the presentation: Japanese | ||
| 田中 宏昌 | 1651069: M, 2回目発表 | 知能コミュニケーション 中村 哲, 金谷 重彦, 鈴木 優, 加藤 晃（バイオ）, 吉野 幸一郎 |
|
title: Generating High Translation Efficient mRNA by Machine Learning
abstract: Improving protein translation efficiency of messenger RNA (mRNA) is important for the development of new medicine. The aim of this study is to make the mRNA sequence that improves the protein translation efficiency of rice. However the experiment for making mRNA is expensive and so we have to develop the method which allow us to make high translation efficient mRNA without experiment. Hence in this presentation we propose the method which uanble us to make high translation efficiency mRNA on computer using machine learning. language of the presentation: Japanese | ||