In this research, we propose intention understanding method utilizing multi-task transfer learning. Our method improves intention understanding accuracy using data of different kind of domain as source domain. As source domain's training data, we use Japanese-English translation data (translation task) and Japanese Wikipedia data (sentence prediction task). As target domain's training data, we use transcribed utterance data of voice control of equipment. In this data, each utterance has one intention label. As an experimental result, we found that proposed method provides a performance improvement over previous transfer learning method in the case of small training data (the number of data for each intention label are 1, 3, 5, 10 and 30).

In recent years, interactive systems and dialogue generation in natural language processing have attracted attention. Due to the spread of the chat bot to the call center, an accurate human interactive response is required. On the other hand, qualitative interactions in sociology's ethnomethodology and discourse analysis / conversation analysis are beneficial. Therefore, once again, using the Japanese language learner conversation data corpus provided by the National Institute for Japanese Language, we examine the effect and aim at applying to the tendency of dialogue breakdown and dialogue generation.

This paper proposes a method of recommending sightseeing spots without discussion for a tourist group. As members in a group usually have different interests in sightseeing spots, it tends to take a lot of time to decide a sightseeing plan which satisfies all members' preference. Furthermore, it is difficult for those who are not good at expressing their opinions to take part in the discussion. With the proposed method, each member in a group inputs his/her interests and conditions about sightseeing spots, and then evaluates the recommended spot list one by one. This paper also proposes to determine the order of evaluating the list on the basis of the questionnaire using the MBTI taxonomy, which have been proposed in the field of psychology. Effectiveness of the proposed method is shown by user experiment.

In Analects, there is a saying "visiting old, learn new". This means to investigate old things, in order to obtain new knowledge and insights.The process of data analysis could be interpreted as an act to analyze the data of the past, discover new knowledge and insights mathematically and make good use of them toward better future. Until end of the 20th century data was very valuable and less reliable, but now the accumulation of data became remarkable due to the explosive spread of the Internet society in recent years. However, proper use of data has not yet been established. The reason for this is that since the data is accumulated according to the operation of each business, there is no standardized analytical method because the accumulation state of data varies. Therefore, it is necessary to edit and integrate data by processing on computer for analytical purpose. Furthermore, it is necessary to examine whether the edited data is appropriate for analysis. To do so, it is convenient to have a tool that analyzes data while visually showing and checking data by editing and examining data. This paper proposes a graphical analysis integration environment "PADOC" to facilitate data editing and data review