Transactions of the Japanese Society for Artificial Intelligence Vol. 31(2016) No. 1 Special Paper “Linked Open Data in Practice”, “Intelligent Dialogue Systems”,“AI Challenge in the Near Future 2014”.
Linked Data is a promising technology for knowledge integration on the web. Many research groups have developed ontologies and terminologies, and recently, they have published a wide variety of Linked Data in the biomedical domain. We have systematized an ontology of abnormal states in the definition of diseases. For effective use of existing biomedical data, one of the difficulties is a conceptual discrepancy rather than a superficial one since data are heterogeneous. This article focuses on knowledge integration with Linked Data in terms of abnormal states. First, we discuss ontological issues of reusing and integrating knowledge of abnormal states in existing biomedical resources. Next, we introduce our ontology of abnormal states. By using our ontology and making explicit the meaning of each concept, we show a solution for the integration. Then, applying a Linked Data technology, we introduce a prototype system to link our ontology as a hub of existing resources across species. In cooperation with disease ontology, we demonstrate finding commonality of causal relationships of abnormal states between diseases across clinical departments. Our approach will bring benefits to fill the gap between basic research and clinical medicine, and contribute to disease knowledge integration of good practice.
As Linked Data consumption has not yet been discussed from the points of information services, we discuss here how to build web applications with function extendibility and service maintainability, integrating Linked Data and rule bases. The function extendibility has been done with less cost just by linking original Linked Data and other Linked Data. The service maintainability has been done just by exchanging initial rules for given task with other rules for another task. Using the framework, we have implemented web application for diet therapy and then discussed how well the diet therapy service goes well with the questionnaire survey of a diabetic.
To address social issues about the sustainability of local societies, inter-organizational collaboration in public sphere is important. Although conventional social networking services (SNSs) have recently been used for public collaboration, the SNSs are not suitable to look for potential collaborators because the conventional SNSs emphasize recency of information and lack a function for sharing information about ``who are trying to address what kind of social issues''. We designed a data model for structuring public issues and goals and built a linked open dataset (LOD) based on the above data model. Moreover, we developed a method for calculating similarities of public goals and implemented a Web service for matching public goals for finding potential collaborators. Our method for similarity calculation incorporates surficial features, semantic features, and contextual features. We conducted an experiment to investigate an optimal balance of parameters of the contextual features, which is suitable for facilitating public collaboration. Furthermore, we held a participatory event for trial use by citizens and observed positive feedbacks from the participants.
Biomimetics contributes to innovative engineering by imitating the models, systems, and elements of nature. Well-known examples of biomimetics include paint and cleaning technologies that imitate the water repellency of the lotus, adhesive tapes that imitate the adhesiveness of gecko feet, and high-speed swimsuits that imitate the low resistance of a shark’s skin. These results integrate studies on the biological mechanisms of organisms with engineering technologies to develop new materials. Facilitating such biomimetics-based innovations requires integrating knowledge, data, requirements, and viewpoints across different domains. Researchers and engineers need to develop a biomimetics database to assist them in achieving this goal. Because ontologies clarify concepts that appear in target domains, we assume that it is important to develop a biomimetics ontology that contributes to improvement of knowledge interoperability between the biology and engineering domains. Furthermore, linked data technologies are very effective for integrating a database with existing biological diversity databases. On the basis of these observations, we developed a biomimetics ontology and keyword exploration tool based on linked data techniques. The tool allows users to find important keywords for retrieving meaningful knowledge from various biomimetics databases. Such a technique could support idea creation by users based on a biomimetics ontology. This paper describes a prototype of our proposed biomimetics ontology and keyword exploration tool.
In this paper, we propose a non-task-oriented dialogue system considering user's preference and human relations. The proposed system has the following two features. First, the system estimates user's emotions from the analysis results of user's utterances and stores the individual preference with specific nouns. It allows the system to determine more suitable topics and answer sentences. Second, the system estimates user's human relations from the analysis of user-to-user acquaintance relationship and similarity of preferences. It makes user more interested in conversation because the user can feel more familiar with the system. The performance of proposed system has been evaluated by subjective experiments. We confirmed the effectiveness of the proposed system, receiving higher score about user's satisfaction than the conventional system.
We propose a method to extract user information in a structured form for personalized dialogue systems. Assuming that user information can be represented as a quadruple <predicate-argument structure, entity, attribute category, topic>, we focus on solving problems in extracting predicate argument structures from question-answer pairs in which arguments and predicates are frequently omitted, and in estimating attribute categories related to user behavior which a method using only content words cannot distinguish. Experimental results show that the proposed method significantly outperformed baseline methods and was able to extract user information with 81.2% precision and 58.1% recall.
In dialogue systems, dialogue modeling is one of the most important factors contributing to user satisfaction. Especially in example-based dialogue modeling (EBDM), effective methods for dialog example databases and selecting response utterances from examples improve dialogue quality. Conventional EBDM-based systems use example database consisting of pair of user query and system response. However, the best responses for the same user query are different depending on the user's preference. We propose an EBDM framework that predicts user satisfaction to select the best system response for the user from multiple response candidates. We define two methods for user satisfaction prediction; prediction using user query and system response pairs, and prediction using user feedback for the system response. Prediction using query/response pairs allows for evaluation of examples themselves, while prediction using user feedback can be used to adapt the system responses to user feedback. We also propose two response selection methods for example-based dialog, one static and one user adaptive, based on these satisfaction prediction methods. Experimental results showed that the proposed methods can estimate user satisfaction and adapt to user preference, improving user satisfaction score.
In conversational dialogue, a talker sometimes asks questions that relate to the other talker's personality, such as his/her favorites and experiences. This behavior also appears in conversational dialogues with a dialogue system; therefore, the system should be developed so that it responds to this kind of questions. Previous systems realized this function by creating question-answer pairs by hand. However, there is no work that examines the coverage of the created question-answer pairs over real conversations. This study analyzes a huge amount of question-answer pairs created by many question-generators, with one answer-generator for each character. Our analysis shows that 41% of personality questions that appeared in real conversations are covered by the created pairs. We also investigated the types of questions that are frequently asked.
Characterizing dialogue system utterances is important in making human-computer interaction systems more friendly and human-like. A method is described for achieving this by converting functional expressions according to their generation probabilities, which are calculated for specific characters. Experimental results show that the method can add characteristics of the target profiles (i.e., gender, age and closeness with a conversation partner) to dialogue system utterances and in so doing can generate a large variety of linguistic expressions.
This paper presents an open domain monologue generation method for non-task-oriented dialogue systems to be able to speak their opinions and impressions as a speaker. To generate monologues, we acquire suitable utterances that contain a given topic from Twitter. Our method determines whether utterances have cohesion or not using the support vector machine and concatenate them in a row. It scores the utterance sequences from the aspect of their humor, unexpectedness and speciality in the given topic. We acquire the utterance sequences that ranks high as monologues. Results of an experiment demonstrate that our method can generate amusing and semantically appropriate monologues.
Special Paper “AI Challenge in the Near Future 2014”
In this article, we investigate “retweeting in Twitter” or information transfer behavior in social media to figure out some characteristics of our information processing behavior in emergency situation from social psychological perspective. We made an exploratory log analysis of Twitter focusing on the relationship between diffusion of disaster information and user's emotional response on them. Disaster-related tweets which were retweeted over 10 times around the time of the Great East Japan Earthquake were extracted and emotional words in them were categorized and counted. Frequently retweeted tweets tended to include more negative (anxious or angry) or active emotional words than positive or inactive words. As results of multiple and quantile regression analyses, negative (especially anxious) or active emotional words in tweets had a significant effect on the increase of retweeting regardless of a kind of disasters. The results were discussed in terms of the difference with those based on common tweets.
This paper proposes a method for generating pairwise constraints from the sequence of user's grouping operations. Constrained clustering has been studied as one of promising approaches of interactive data mining. However, when it is applied to actual tasks, it is important to reduce user's cost of specifying constraints. As one of the solutions, this paper focuses on a method for automatically generating a set of pairwise constraints from user's grouping operations. Two types of constraint generation methods are proposed: one is based on the history of user's past grouping operations, and another is based on hierarchical cluster structure. As another contribution, this paper also proposes a design pattern for TETDM (Total Environment for Text Data Mining). TETDM has been proposed for developing various text mining systems, and it is also known as suitable platform for developing a prototype system. Although key concept of TETDM is that a user can dynamically combine various mining and visualization modules when analyzing text data, it is difficult to change combination of mining processing dynamically. The proposed design pattern aims to enable dynamic combination of mining modules by introducing the concept of control panel. An interactive document clustering system is developed on TETDM using the proposed design pattern. Effectiveness of the proposed constraint generation methods is evaluated using the system. Comparison of the effectiveness of several approaches for generating pairwise constraints is performed with simulation.