京都大学 / 0048 / 新制・課程博士 / 博士(地球環境学) / 甲第20594号 / 地環博第165号 / 新制||地環||33(附属図書館) / 京都大学大学院地球環境学舎環境マネジメント専攻 / (主査)教授 星野 敏, 教授 柴田 昌三, 准教授 西前 出 / 学位規則第4条第1項該当 / Doctor of Global Environmental Studies / Kyoto University / DFAM

This thesis presents an algorithm incorporating pattern grammar with random indexing to solve three English synonym benchmarks. A pattern grammar model and a baseline random indexing implementation benchmarked the solution. The results show an significant improvement on the synonym benchmark compared to a baseline random indexing implementation. Most language models today focus on vector space models where the linguistic origins of the information are lost. Even though these algorithms produce good results, it is hard to know where the model learned something. With the help of patterns, we can learn more about how these models work. / Den här uppsatsen presenterar en algoritm som använder sig av mallgrammatik tillsammans med random indexing för att lösa tre synonymtest för engelska. En mallgrammatiksmodell och en referensimplementation av random indexing utvärderades. Resultaten visade en tydlig förbättring på de olika testerna jämfört med referensimplementationen. De flesta språkmodeller idag fokuserar på vektorrepresentationer av språk där det lingvistiska ursprunget hos språket försvinner. Dessa modeller är mycket framgångsrika, men det är svårt att säga något om vad och hur en modell kommit fram till en slutsats. Med hjälp av språkmönster baserade på mallgrammatik kan vi lära oss mer om hur dessa modeller fungerar.

The latest developments of mobile devices include a variety of hardware features that allow for more rich data collection and services. Numerous sensors, Internet connectivity, low energy Bluetooth connectivity to other devices (e.g., smart watches, activity tracker, health data monitoring devices) are just some examples of hardware that helps to provide additional information that can be beneficially used for many application domains. Among others, they could be utilized in mobile learning scenarios (for data collection in science education, field trips), in mobile health scenarios (for health data collection and monitoring the health state of patients, changes in health conditions and/or detection of emergency situations), and in personalized recommender systems. This information captures the current context situation of the user that could help to make mobile applications more personalized and deliver a better user experience. Moreover, the context related information collected by the mobile device and the different applications can be enriched by using additional external information sources (e.g., Web Service APIs), which help to describe the user’s context situation in more details. The main challenge in context modeling is the lack of generalization at the core of the model, as most of the existing context models depend on particular application domains or scenarios. We tackle this challenge by conceptualizing and designing a rich generic context model. In this thesis, we present the state of the art of recent approaches used for context modeling and introduce a rich context model as an approach for modeling context in a domain-independent way. Additionally, we investigate whether context information can enhance existing mobile applications by making them sensible to the user’s current situation. We demonstrate the reusability and flexibility of the rich context model in a several case studies. The main contributions of this thesis are: (1) an overview of recent, existing research in context modeling for different application domains; (2) a theoretical foundation of the proposed approach for modeling context in a domain-independent way; (3) several case studies in different mobile application domains.

This study presents a new approach for finding semantically similar words from corpora using window based context methods. Previous studies mainly concentrate on either finding new combination of distance-weight measurement methods or proposing new context methods. The main difference of this new approach is that this study reprocesses the outputs of the existing methods to update the representation of related word vectors used for measuring semantic distance between words, to improve the results further. Moreover, this novel technique provides a solution to the data sparseness of vectors which is a common problem in methods which uses vector space model. The main advantage of this new approach is that it is applicable to many of the existing word similarity methods using the vector space model. The other and the most important advantage of this approach is that it improves the performance of some of these existing word similarity measuring methods.

This thesis focuses on index and retrieval mechanism of an ontology based medical domain search engine. First, indexing techniques and retrieval methods are reviewed. Then, a special indexing and retrieval mechanism are introduced. This thesis also specifies the functional requirements of these mechanisms. Finally, an evaluation is given by indicating the positive and negative aspects of mechanisms.

Atualmente, pesquisadores e acadêmicos têm beneficiado-se muito com o crescimento acelerado das tecnologias Web, pois os resultados de pesquisa podem ser publicados e acessados eletronicamente tão logo a mesma tenha sido realizada. Esta possibilidade é vantajosa na medida em que minimiza as barreiras de tempo e espaço associadas à publicação tradicional. Neste contexto, surgem as Bibliotecas Digitais como repositórios de dados que, além dos documentos digitais propriamente ditos, ou de apontadores para estes documentos, armazenam os metadados associados. Para permitir que diferentes Bibliotecas Digitais possam interoperar surgiu a Open Archives Initiative (OAI) e, para resolver a questão da padronização dos metadados utilizados pelos repositórios, foi criado o formato Dublin Core (DC). Por outro lado, a enorme quantidade de documentos digitais disponíveis na Web tem causado o fenômeno conhecido como “sobrecarga de informação”. Com o objetivo de suprir esta dificuldade, Sistemas de Recomendação têm sido propostos e desenvolvidos. Estes sistemas visam prover uma interface alternativa para tecnologias de filtragem e recuperação de informações, tendo como foco a predição daqueles itens ou partes da informação que o usuário acharia interessante e útil. Portanto, os Sistemas de Recomendação atuam baseados em personalização da informação sendo que as predições geralmente são realizadas utilizando-se um perfil de cada usuário. A personalização está relacionada com o modo pelo qual a informação e serviços podem ser ajustados às necessidades específicas de um usuário ou comunidade. Esta dissertação descreve um Sistema de Recomendação de artigos científicos, armazenados em bibliotecas digitais. Este sistema é dirigido à comunidade científica da área da Ciência da Computação. Tecnologicamente, o sistema proposto foi desenvolvido sob a perspectiva da Web Semântica, à medida que faz uso de suas tecnologias emergentes tais como: uso de metadados padrão para a descrição de documentos - Dublin Core, uso do padrão XML para a descrição do perfil do usuário - Currículo Lattes, e provedores de serviços e de dados (OAI) envolvidos no processo de geração das recomendações. Este trabalho ainda apresenta e discute alguns resultados de experimentos baseados em avaliações quantitativas e qualitativas de recomendações geradas pelo sistema. / Currently, researchers and academics have been benefited by the expressive growth of web technologies, due to the possibility of publishing and accessing research results as soon as they are achieved. This possibility is advantageous as it minimizes the time and space barriers that traditional publications present. In this context, Digital Libraries emerged as data repositories that, beyond digital documents or links to them, store associated metadata. To allow the interoperability among different Digital Libraries, the Open Archives Initiative (OAI) was defined and, to solve the problem of metadata standardization, the Dublin Core standard (DC) was created. On the other hand, the great amount of available digital documents in the Web has caused the phenomenon known as “information overload”. In order to avoid this difficulty, Recommender Systems have been proposed and developed. These systems intend to provide an alternative interface for information filtering and retrieval technologies, focusing on the prediction of items or information parts that are interesting and useful for the user. Therefore, Recommender Systems act based on information personalization, and the predictions are generally generated using each user’s profile. The personalization is related to the way the information and the provided services can be adjusted to the specific necessities of a user or community. This dissertation describes a Recommender System for scientific articles stored in digital libraries. This system is geared towards the Computer Science scientific community. Technologically, the proposed system was developed under the Semantic Web perspective, as it explores its emergent technologies such as: use of standard metadata for document description - Dublin Core, use of the XML standard for users’ profile description - Lattes Curriculum Vitae, and services and data providers (OAI) involved on the recommendations generation process. In addition, this work presents and discusses some experimental results; the experiments are based on quantitative and qualitative evaluations of recommendations generated by the system.

abstract: For the past three decades, the design of an effective strategy for generating poetry that matches that of a human’s creative capabilities and complexities has been an elusive goal in artificial intelligence (AI) and natural language generation (NLG) research, and among linguistic creativity researchers in particular. This thesis presents a novel approach to fixed verse poetry generation using neural word embeddings. During the course of generation, a two layered poetry classifier is developed. The first layer uses a lexicon based method to classify poems into types based on form and structure, and the second layer uses a supervised classification method to classify poems into subtypes based on content with an accuracy of 92%. The system then uses a two-layer neural network to generate poetry based on word similarities and word movements in a 50-dimensional vector space. The verses generated by the system are evaluated using rhyme, rhythm, syllable counts and stress patterns. These computational features of language are considered for generating haikus, limericks and iambic pentameter verses. The generated poems are evaluated using a Turing test on both experts and non-experts. The user study finds that only 38% computer generated poems were correctly identified by nonexperts while 65% of the computer generated poems were correctly identified by experts. Although the system does not pass the Turing test, the results from the Turing test suggest an improvement of over 17% when compared to previous methods which use Turing tests to evaluate poetry generators. / Dissertation/Thesis / Masters Thesis Computer Science 2016

Atualmente, pesquisadores e acadêmicos têm beneficiado-se muito com o crescimento acelerado das tecnologias Web, pois os resultados de pesquisa podem ser publicados e acessados eletronicamente tão logo a mesma tenha sido realizada. Esta possibilidade é vantajosa na medida em que minimiza as barreiras de tempo e espaço associadas à publicação tradicional. Neste contexto, surgem as Bibliotecas Digitais como repositórios de dados que, além dos documentos digitais propriamente ditos, ou de apontadores para estes documentos, armazenam os metadados associados. Para permitir que diferentes Bibliotecas Digitais possam interoperar surgiu a Open Archives Initiative (OAI) e, para resolver a questão da padronização dos metadados utilizados pelos repositórios, foi criado o formato Dublin Core (DC). Por outro lado, a enorme quantidade de documentos digitais disponíveis na Web tem causado o fenômeno conhecido como “sobrecarga de informação”. Com o objetivo de suprir esta dificuldade, Sistemas de Recomendação têm sido propostos e desenvolvidos. Estes sistemas visam prover uma interface alternativa para tecnologias de filtragem e recuperação de informações, tendo como foco a predição daqueles itens ou partes da informação que o usuário acharia interessante e útil. Portanto, os Sistemas de Recomendação atuam baseados em personalização da informação sendo que as predições geralmente são realizadas utilizando-se um perfil de cada usuário. A personalização está relacionada com o modo pelo qual a informação e serviços podem ser ajustados às necessidades específicas de um usuário ou comunidade. Esta dissertação descreve um Sistema de Recomendação de artigos científicos, armazenados em bibliotecas digitais. Este sistema é dirigido à comunidade científica da área da Ciência da Computação. Tecnologicamente, o sistema proposto foi desenvolvido sob a perspectiva da Web Semântica, à medida que faz uso de suas tecnologias emergentes tais como: uso de metadados padrão para a descrição de documentos - Dublin Core, uso do padrão XML para a descrição do perfil do usuário - Currículo Lattes, e provedores de serviços e de dados (OAI) envolvidos no processo de geração das recomendações. Este trabalho ainda apresenta e discute alguns resultados de experimentos baseados em avaliações quantitativas e qualitativas de recomendações geradas pelo sistema. / Currently, researchers and academics have been benefited by the expressive growth of web technologies, due to the possibility of publishing and accessing research results as soon as they are achieved. This possibility is advantageous as it minimizes the time and space barriers that traditional publications present. In this context, Digital Libraries emerged as data repositories that, beyond digital documents or links to them, store associated metadata. To allow the interoperability among different Digital Libraries, the Open Archives Initiative (OAI) was defined and, to solve the problem of metadata standardization, the Dublin Core standard (DC) was created. On the other hand, the great amount of available digital documents in the Web has caused the phenomenon known as “information overload”. In order to avoid this difficulty, Recommender Systems have been proposed and developed. These systems intend to provide an alternative interface for information filtering and retrieval technologies, focusing on the prediction of items or information parts that are interesting and useful for the user. Therefore, Recommender Systems act based on information personalization, and the predictions are generally generated using each user’s profile. The personalization is related to the way the information and the provided services can be adjusted to the specific necessities of a user or community. This dissertation describes a Recommender System for scientific articles stored in digital libraries. This system is geared towards the Computer Science scientific community. Technologically, the proposed system was developed under the Semantic Web perspective, as it explores its emergent technologies such as: use of standard metadata for document description - Dublin Core, use of the XML standard for users’ profile description - Lattes Curriculum Vitae, and services and data providers (OAI) involved on the recommendations generation process. In addition, this work presents and discusses some experimental results; the experiments are based on quantitative and qualitative evaluations of recommendations generated by the system.

Atualmente, pesquisadores e acadêmicos têm beneficiado-se muito com o crescimento acelerado das tecnologias Web, pois os resultados de pesquisa podem ser publicados e acessados eletronicamente tão logo a mesma tenha sido realizada. Esta possibilidade é vantajosa na medida em que minimiza as barreiras de tempo e espaço associadas à publicação tradicional. Neste contexto, surgem as Bibliotecas Digitais como repositórios de dados que, além dos documentos digitais propriamente ditos, ou de apontadores para estes documentos, armazenam os metadados associados. Para permitir que diferentes Bibliotecas Digitais possam interoperar surgiu a Open Archives Initiative (OAI) e, para resolver a questão da padronização dos metadados utilizados pelos repositórios, foi criado o formato Dublin Core (DC). Por outro lado, a enorme quantidade de documentos digitais disponíveis na Web tem causado o fenômeno conhecido como “sobrecarga de informação”. Com o objetivo de suprir esta dificuldade, Sistemas de Recomendação têm sido propostos e desenvolvidos. Estes sistemas visam prover uma interface alternativa para tecnologias de filtragem e recuperação de informações, tendo como foco a predição daqueles itens ou partes da informação que o usuário acharia interessante e útil. Portanto, os Sistemas de Recomendação atuam baseados em personalização da informação sendo que as predições geralmente são realizadas utilizando-se um perfil de cada usuário. A personalização está relacionada com o modo pelo qual a informação e serviços podem ser ajustados às necessidades específicas de um usuário ou comunidade. Esta dissertação descreve um Sistema de Recomendação de artigos científicos, armazenados em bibliotecas digitais. Este sistema é dirigido à comunidade científica da área da Ciência da Computação. Tecnologicamente, o sistema proposto foi desenvolvido sob a perspectiva da Web Semântica, à medida que faz uso de suas tecnologias emergentes tais como: uso de metadados padrão para a descrição de documentos - Dublin Core, uso do padrão XML para a descrição do perfil do usuário - Currículo Lattes, e provedores de serviços e de dados (OAI) envolvidos no processo de geração das recomendações. Este trabalho ainda apresenta e discute alguns resultados de experimentos baseados em avaliações quantitativas e qualitativas de recomendações geradas pelo sistema. / Currently, researchers and academics have been benefited by the expressive growth of web technologies, due to the possibility of publishing and accessing research results as soon as they are achieved. This possibility is advantageous as it minimizes the time and space barriers that traditional publications present. In this context, Digital Libraries emerged as data repositories that, beyond digital documents or links to them, store associated metadata. To allow the interoperability among different Digital Libraries, the Open Archives Initiative (OAI) was defined and, to solve the problem of metadata standardization, the Dublin Core standard (DC) was created. On the other hand, the great amount of available digital documents in the Web has caused the phenomenon known as “information overload”. In order to avoid this difficulty, Recommender Systems have been proposed and developed. These systems intend to provide an alternative interface for information filtering and retrieval technologies, focusing on the prediction of items or information parts that are interesting and useful for the user. Therefore, Recommender Systems act based on information personalization, and the predictions are generally generated using each user’s profile. The personalization is related to the way the information and the provided services can be adjusted to the specific necessities of a user or community. This dissertation describes a Recommender System for scientific articles stored in digital libraries. This system is geared towards the Computer Science scientific community. Technologically, the proposed system was developed under the Semantic Web perspective, as it explores its emergent technologies such as: use of standard metadata for document description - Dublin Core, use of the XML standard for users’ profile description - Lattes Curriculum Vitae, and services and data providers (OAI) involved on the recommendations generation process. In addition, this work presents and discusses some experimental results; the experiments are based on quantitative and qualitative evaluations of recommendations generated by the system.

This thesis addresses the new item problem in recommender systems, which pertains to the challenges of providing personalized recommendations for items which have limited user interaction history. The study proposes and evaluates a method for generating personalized recommendations for movies, shows, and series on one of Sweden’s largest streaming platforms. By treating these items as documents of the attributes which characterize them and utilizing item similarity through the k-nearest neighbor algorithm, user preferences for new items are predicted based on users’ past preferences for similar items. Two models for feature representation, namely the Vector Space Model (VSM) and a Latent Dirichlet Allocation (LDA) topic model, are considered and compared. The k-nearest neighbor algorithm is utilized to identify similar items for each type of representation, with cosine distance for VSM and Kullback-Leibler divergence for LDA. Furthermore, three different ways of predicting user preferences based on the preferences for the neighbors are presented and compared. The performances of the models in terms of predicting preferences for new items are evaluated with historical streaming data. The results indicate the potential of leveraging item similarity and previous streaming history to predict preferences of new items. The VSM representation proved more successful; using this representation, 77 percent of actual positive instances were correctly classified as positive. For both types of representations, giving higher weight to preferences for more similar items when predicting preferences yielded higher F2 scores, and optimizing for the F2 score implied that recommendations should be made when there is the slightest indication of preference for the neighboring items. The results indicate that the neighbors identified through the VSM representation were more representative of user preferences for new items, compared to those identified through the LDA representation.