Toshiyuki Takezawa

Construction of a cooking ontology from cooking recipes and patents

A cooking ontology is an indispensable language resource for the language processing of cooking recipes. We have constructed a cooking ontology by means of pattern matching, statistical natural language processing techniques, and manual steps to identify hyponymy, synonymy, attributes, and meronymy.

Automatic Compilation of an Online Travel Portal from Automatically Extracted Travel Blog Entries

For travelers who plan to visit a particular tourist spot, information about it is required. In this paper, we propose a method for extracting and organizing appropriate information from weblogs (blogs). Recently, increased numbers of travelers have been writing of their travel experiences via blogs. We call these travel blog entries, and they contain much useful travel information. For example, some bloggers introduce useful web sites for a tourist spot, while others report on transportation between tourist spots. Here, we extract hyperlinks of web sites for tourist spots from travel blog entries and organize them via automatic classification. We also extract transportation information automatically from travel blog entries. To investigate the effectiveness of our method, we conducted experiments. For the extraction of transportation information, we obtained an 80.3% for Precision. For the classification of hyperlinks, we obtained a high Precision. Finally, we constructed a prototype system, which provides information about (1) transportation between tourist spots and (2) useful web sites for tourist spots.

Extraction and visualization of technical trend information from research papers and patents

To a researcher in a field with high industrial relevance, retrieving and analyzing research papers and patents are important aspects of assessing the scope of the field. Knowledge of the history and effects of the elemental technologies is important for understanding trends. We propose a method for automatically creating a technical trend map from both research papers and patents by focusing on the elemental (underlying) technologies and their effects. We constructed a method that can be used in any research field. To investigate the effectiveness of our method, we conducted an experiment using the data in the NTCIR-8 Workshop Patent Mining Task. The results of our experiment showed recall and precision scores of 0.254 and 0.496, respectively, for the analysis of research papers, and recall and precision scores of 0.455 and 0.507, respectively, for the analysis of patents. Those results indicate that our method for mapping technical trends is both useful and sound.

Automatic creation of a technical trend map from research papers and patents

For a researcher in a field of great industrial relevance, retrieving and analyzing research papers and patents has become an important aspect of assessing the scope of the field. We propose a method for creating a technical trend map automatically from both research papers and patents. For the construction of the technical trend map, we focus on the elemental (underlying) technologies used in a particular field, and their effects. Knowledge of the history and effects of the elemental technologies used in a particular field is essential for grasping the outline of technical trends in the field. Therefore, we have constructed a method that can recognize the application of elemental technologies and their effects in any research field. To investigate the effectiveness of our method, we conducted an experiment using the data in the NTCIR-8 Patent Mining Task. From our experimental results, we obtained Recall and Precision scores of 0.160 and 0.491, respectively, for the analysis of research papers. We also obtained Recall and Precision scores of 0.431 and 0.545, respectively, for the analysis of patents. Finally, we have constructed a system that creates an effective technical trend map for a given field.

Enriching Travel Guidebooks with Travel Blog Entries and Archives of Answered Question

Travellers planning to visit particular tourist spots need information about their destination and they often use travel guidebooks to collect this information. However, guidebooks lack specific information, such as first-hand accounts by users who have visited the specific destination. To compensate for the lack of such information, we focused on travel blog entries and archives of answered questions. In this paper, we propose a method for enriching guidebooks by matching and aligning the information with blog entries and questions answered (QA) archives. This is a three-step method. In Step 1, we classify pages of guidebooks, blog entries, and QA archives into five types of content, such as “watch,” “dine,” etc. In Step 2, we align each blog entry and QA archive with guidebooks by taking these content types into account. In Step 3, we match each blog entry and QA archive with individual pages in guidebooks. We conducted some experiments, and confirmed the effectiveness of our method.

A method using acoustic features to detect inadequate utterances in medical communication

We previously proposed a method that uses grammatical features to detect inadequate utterances of doctors. However, nonverbal information such as that conveyed by gestures, facial expression, and tone of voice are also important. In this paper, we propose a method that uses eight acoustic features to detect three types of mental states (sincerity, confidence, and doubtfulness/acceptance). A Support Vector Machine (SVM) is used to learn these features. Experiments showed that the systems accuracy and recall rates respectively ranged from 0.79-0.91 and 0.80-0.94.

A question-and-answer classification technique for constructing and managing spoken dialog system

To recognize user speech accurately and respond to it appropriately, a spoken dialog system usually uses a question-and-answer database (QADB) which contains many question-and-answer pairs. The systems first select a question example which is the most similar to the recognition result for the input voice from the database. An answer sentence which is then paired with the selected question example is output to the user. Many systems have a large database to enable a more appropriate answer to be output. However, when such a database is used, the waiting time increases because the system needs to find the most appropriate question example from a vast number of question examples. We propose a method of classifying the queries in the QADB. By classifying question examples into some clusters using pLSA, an appropriate question example can be found more quickly than when using the conventional method. We evaluated the validity of our proposed method by changing various parameters.

Classification of Research Papers into a Patent Classification System Using Two Translation Models

Classifying research papers into patent classification systems enables an exhaustive and effective invalidity search, prior art search, and technical trend analysis. However, it is very costly to classify research papers manually. Therefore, we have studied automatic classification of research papers into a patent classification system. To classify research papers into patent classification systems, the differences in terms used in research papers and patents should be taken into account. This is because the terms used in patents are often more abstract or creative than those used in research papers in order to widen the scope of the claims. It is also necessary to do exhaustive searches and analyses that focus on classification of research papers written in various languages. To solve these problems, we propose some classification methods using two machine translation models. When translating English research papers into Japanese, the performance of a translation model for patents is inferior to that for research papers due to the differences in terms used in research papers and patents. However, the model for patents is thought to be useful for our task because translation results by patent translation models tend to contain more patent terms than those for research papers. To confirm the effectiveness of our methods, we conducted some experiments using the data of the Patent Mining Task in the NTCIR-7 Workshop. From the experimental results, we found that our method using translation models for both research papers and patents was more effective than using a single translation model.

Natural language dialog system considering speaker’s emotion for open-ended conversation

To respond appropriately to an utterance, human-like communication system, should consider not only words in the utterance but also the speaker’s emotion. We thus proposed a natural language dialog system that can estimate the user’s emotion from utterances and respond on the basis of the estimated emotion. To estimate a speaker’s emotion (positive, negative, or neutral), 384 acoustic features extracted from an utterance are utilized by a Support Vector Machine (SVM). Artificial Intelligence Markup Language (AIML)-based response generating rules are expanded so that the speaker’s emotion can be considered as a condition of these rules. Two experiments were carried out to compare impressions of a dialog agent that considered emotion (proposed system) with those of an agent that did not (previous system). In the first experiment, 10 subjects evaluated the impressions after watch four conversation videos (no emotion estimation, correct emotion estimation, inadequate emotion estimation, and imperfect emotion ...

Automatic Extraction of Event Information from Newspaper Articles and Web Pages

In this paper, we propose a method for extracting travel-related event information, such as an event name or a schedule from automatically identified newspaper articles, in which particular events are mentioned. We analyze news corpora using our method, extracting venue names from them. We then find web pages that refer to event schedules for these venues. To confirm the effectiveness of our method, we conducted several experiments. From the experimental results, we obtained a precision of 91.5% and a recall of 75.9% for the automatic extraction of event information from news articles, and a precision of 90.8% and a recall of 52.8% for the automatic identification of event-related web pages.