Language learners encounter a wide range of grammar items across the beginner, intermediate, and advanced levels.To develop grammatical error correction (GEC) models effectively, it is crucial to identify which grammar items are easier or more challenging for models to correct.However, conventional benchmarks based on learner-produced texts are insufficient for conducting detailed evaluations of GEC model performance across a wide range of grammar items due to biases in their distribution.To address this issue, we propose a new evaluation paradigm that assesses GEC models using minimal pairs of ungrammatical and grammatical sentences for each grammar item.As the first benchmark within this paradigm, we introduce the CEFR-based Targeted Syntactic Evaluation Dataset for Grammatical Error Correction (CTSEG), which complements existing English benchmarks by enabling fine-grained analyses previously unattainable with conventional datasets.Using CTSEG, we evaluate three mainstream types of English GEC models: sequence-to-sequence models, sequence tagging models, and prompt-based models.The results indicate that while current models perform well on beginner-level grammar items, their performance deteriorates substantially for intermediate and advanced items.Dataset Sents.Refs.Error Tags CEFR Level FCE (Yannakoudakis et al., 2011) 2,695 1 71 B1-B2 KJ (Nagata et al., 2011) 3,199 1 22 A1-A2?CoNLL-2014 (Ng et al., 2014) 1,312 2 28 C1 AESW (Daudaravicius et al., 2016) 143,804 1 N/A C1-C2 (+Native) JFLEG (Napoles et al., 2017) 747 4 N/A A1-C2?BEA-2019 (Bryant et al., 2019) 4,384 5 25 A1-C2 (+Native) GMEG (Napoles et al., 2019) 2,960 4 N/A B1-B2 (+Native) CWEB (Flachs et al., 2020)

This paper proposes a simple Knowledge Graph Embedding (KGE) framework that considers the entity types. The KGE finds appropriate representations of entities and relations by learning structured information in triples and uses these representations to predict the missing links in a knowledge graph (KG). The entity types are included in many KGs, but most of the existing methods ignored its potential for the link prediction task. The proposed framework, which is called EETCRL (Entity and Entity Type Composition Representation Learning), learns representations for entities and entity types in a simple way. Those combined representations are used for link prediction. The proposed framework is evaluated on three datasets and the results show that while the overall performance was comparable to baseline methods, its performance depends on the entity type. This paper also discusses how the information about entity types affects the link prediction through analysis of the fine-grained results.

This paper investigates the impression of streetscapes users receive from street view movies. In travel recommendation, travelers do not only want to reach the destination, but also want to enjoy streetscapes throughout the journey. While traditional route recommendation systems tend to focus on how to get users to the destination faster, we think routes to the destination should be determined considering user experience on the route. To consider such a user experience, the lack of actual route conditions due to 2D maps is also a problem of the traditional route recommendation systems. Therefore, we are developing a recommendation system that can provide street view movies of the recommended routes. As the preliminary investigation for realizing the system, this paper compares the impression users receive from street view movies and that from 2D map. The results show that even while presenting information of the same route with 2D Map and Street view, users still receive the different impressions.

This paper proposes a visual analytics framework for formulating metrics for evaluating multi-dimensional time-series data. Multidimensional time-series data has been collected and utilized in different domains. We believe evaluation metrics play an important role in utilizing those data, such as decision making and labeling training data used in machine learning. However, it is a difficult task for even domain experts to formulate metrics. To support the process of formulating metrics, the proposed framework represents metrics as a linear combination of data attributes, and provides a means for formulating it through interactive data exploration. A prototype interface that visualizes target data as an animated scatter plot was implemented. Through this interface, several visualized objects can be directly manipulated: a node and a trajectory of an instance, and a convex hull as the group of nodes and trajectories. Linear combinations of attributes are adjusted in accordance with the manipulation of different objects' types by the user. The effectiveness of the proposed framework was demonstrated through two application examples with real-world data.

This paper proposes a method to estimate the posterior distribution of a Boltzmann Machine. Due to high feature extraction ability, a Boltzmann Machine is often used for both of supervised and unsupervised learning. It is expected to be suitable for multimodal data because of its bi-directional connection property. However, it needs a sampling method to estimate the posterior distribution, which becomes a problem during an inference period because of the computation time and instability. Therefore, it is usually converted to feedforward Neural Networks, which means to lose its bi-directional property. To deal with these problems, this paper proposes a method to estimate the posterior distribution of the Boltzmann Machine fast and stably without converting it to feedforward Neural Networks. The key idea of the proposed method is to estimate the posterior distribution using a Simulated Annealing on non-uniform temperature distribution. The advantage of the proposed method against normal sampling method and conventional Simulated Annealing is shown through experiments with artificial dataset and MNIST. Furthermore, this paper also gives the mathematical analysis of Boltzmann Machine's behaviour with regard to temperature distribution.

Recently, there are many resources of scientific research on the Web. Because the gap between the amount of academic information available on the Web and human processing abilities becomes large, several problems have arisen: (1) losing opportunities of research presentation, (2) loosing opportunities of gathering research information, (3) increasing burden of peer review, (4) difficulty in selecting papers to read. In order to solve these problems, quantitative evaluation index of a paper as a selection criterion is needed.

In recent years, time series data have been collected in many fields, and a visual analytics (VA) interface is expected to be useful for utilizing such data. When developing such interfaces for time series data, several problems arising from the property of time series data need to be resolved. For example, the temporal trend of data is usually visualized with animation. However, with this approach, a collision would occur between the movement of the time series data itself and that caused by interaction with users. As a result, visual clutter often occurs on a display. To solve these problems, this paper focuses on trajectories, which can handle temporal and spatial changes uniformly, and proposes a VA interface that enables direct manipulation of trajectories. The usefulness of the proposed interface is demonstrated through experiments.

This paper applies a Tangible User Interface (TUI) for information curation using Near Field Communication (NFC) to an analogy game. The increase in text data is more remarkable in current IT society. Although those are usually accessed with using Graphical User Interface (GUI), users except experienced computer users have difficulty in reading and organizing data with GUI. In particular, information curation such as grouping related data / information and finding relationship among them is difficult. In order to solve this problem, an interface that can access text data intuitively is expected. We are developing a TUI based on NFC, by which a user can move and group text data in a similar manner when handling paper documents. As one of the promising applications of the proposed TUI, this paper focuses on creative thinking support, for which touching externalized thought by hand is expected to be effective. An experiment is conducted, in which test participants did an analogy game with using the proposed TUI. The experimental result shows experience of using the TUI affects the participants’ self-evaluation about idea creation.

This paper proposes an interactive visualization system that supports tasks of monitoring multiple BBS threads regularly. A BBS is useful as means to know the latest information and wide range of opinions about the topics of interest. However, we must spend much time on reading large number of posts in a thread to recognize topics discussed in it. In addition, it is so difficult to read multiple threads simultaneously that we may miss valuable information. The proposed monitoring system resolves these problems by keyword-based visualization. Furthermore, this system realizes flexible monitoring reflecting user's interest by keyword tracking function for confirming the latest topics about specified keywords. This paper analyzes the behaviors of test participants using the prototype system based on eye-tracking data.

This paper proposes a visualization system for getting insight into future research activities from co-authorship networks. A bibliographic network such as a co-authorship network and a citation network is important information for researchers when doing a research survey. In particular, there are many requests on research survey that relate with researchers' future activities, such as identification of remarkable of researchers including growing researchers and supervisors. Although a citation network has received many attentions from researchers, it is not suitable for such surveys because it reflects researchers' past activities. Since collaboration of researchers is essential for researchers' activities, co-authorship network is suitable for predicting future activities. In order to get insights into future research activities by discriminating growing research areas from grown-up areas, the proposed visualization system provides the function for identifying research areas and that for identifying time variation of both network structure and keyword distribution. As a basis for getting insights into future research activities, this paper focuses on the task of discriminating growing researchers from supervisors. The effectiveness of the proposed system is evaluated through the detailed analysis of two participants' analyzing process of InfoVis 2004 Contest dataset.

This paper investigates the characteristics of robots for non-industrial use such as home robots, when those are used as an interface for accessing information. Although the support based on information is one of important capabilities home robots should have, it has not been fully studied what are the merits of providing information via a robot compared with accessing information via PC or a mobile phone. In order to investigate the merit, two experiments are performed in this paper. The experimental result show a robot can effectively attract users when providing (recommending) information. It is also shown information provided by the prototype can promote the communication among users. The obtained result supports the significance of performing information support by home robots, and be used for designing home robots that provide information-based support.

This paper studies the communication support between human and partner robot based on TV program recommendation. It is expected in near future that various robots designed for supporting humans will come into our daily lives. In order for such robots to coexist with humans, those should be recognized as ldquopartnersrdquo for us. Therefore, studying communication between humans and robots is very important. In this paper, we focus on the communication under the situation of watching TV, because TV is bound up with the ordinary, natural rhythms of our daily life. If a robot can estimate our interests / preference by observing our watching behavior and have conversation about the topics we might be interested in, we could recognize the robot as a partner. This paper applies TV program recommendation to communication support between human and partner robot. First, the paper proposes a method for generating user profile for TV program recommendation based on fuzzy inference. The method does not estimate userpsilas interest in a TV program only from its watching time as most of existing methods do, but also from userpsilas utterances by applying sentiment analysis. Experiments are performed with test subjects, and the results show the proposed method can generate a user profile that can reflect userpsilas interests. The paper also introduces a prototype system for studying communication support between human and partner robot under TV watching environment, in which the proposed profile generation method is implemented. The system architecture as well as how it works for TV program recommendation is described.

This paper examines users behavior in decision making through exploratory search on a keyword map. The keyword map is a kind of interactive information visualization system, which displays relationships between objects (keywords). The system has been studied for analyzing data. The keyword map has a multiple relevance controller for analyzing relationship between keywords, which is represented by multiple attributes, from various viewpoints. A user can adjust weights of attributes according to his/her interest, using the controller. We expect that the keyword map can support compensatory/non-compensatory decision strategies by the multiple relevance controller. We perform an experiment for confirming user's behavior during exploratory search on the keyword map. Experimental result shows a user operates flexibly the multiple relevance controller according to his/her decision strategies.

The keyword map, which is a kind of information visualization systems, has been studied for visualizing keyword space. This system shows keywords and relationships between keywords by a graph drawing method, and is equipped with interactive features. However, existing keyword map has not considered positively the reflection of user's intention in keyword space. In this paper, the keyword map equipped with arrangement supporting functions is proposed for emphasizing user's intention. The relation between user's intentions and their usages of arrangement supporting functions is examined through experiments

A bulletin board system (BBS) equipped with KeyGraph is proposed for supporting online chance discovery process. While conventional meeting for chance discovery with KeyGraph requires participants to meet in a meeting room, the paper aims to expand the discussion space over the Internet. In order to support online chance discovery process, the BBS has functions for assisting a user in writing scenarios with visual annotation on the KeyGraph. The system also has a function for retrieving similar scenarios. The experimental result with test subjects shows the effectiveness of the BBS for online chance discovery process.

A comparison method for Web pages in terms of visual similarity is proposed. Conventional Web information retrieval/gathering systems, such as search engines, extract keywords from HTML source files, based on which the similarity between pages is calculated. The extracted keywords are considered as semantic features representing the contents of Web pages. On the other hand, visual feature of Web pages is as important as semantic feature, because HTML is designed for visualizing a Web page in understandable manner for humans. The proposed method compares the layouts of Web pages based on image processing and graph matching. The experimental results show that the accuracy of layout analysis is 91.6% in average, and the visual similarity calculated by the proposed method is closer to the visual judgment by test subjects than color-based comparison method.

We examine the effect of the fuzzy aggregation operators on the image retrieval performance, by empirically comparing 67 operators, applied to the problem of computing the image similarity, given a collection of feature similarities of the image regions. While majority of the existing image similarity models express the image similarity as an aggregation of feature similarities, no study presents a systematic comparison of the different operators. We compare the 67 operators by: (1) incorporating each operator into a hierarchical, region-based similarity model, which expresses the image similarity as an aggregation of region similarities, and each region similarity as an aggregation of the corresponding feature similarities; and (2) evaluating the obtained model(s) on five test databases, containing 64,339 general-purpose images, in 749 semantic categories. Results show that the retrieval performance strongly depends on the operator(s) incorporated in the similarity model - the difference in the average retrieval precision between the best and the worst performing of the 67 operators is up to 50%.

The relevance feedback based on a keyword map is proposed so that a Web interface can be more interactive. There exists vast amount of information in the Web, from which users usually gather information without definite information needs. Therefore, it is difficult for users to organize and understand what they have gathered from the Web. From this viewpoint, we have proposed the concept of RBA-based interaction, in which analysis operation aims to assist users in understanding the context of their Web interaction. However, the prototype interface currently focuses on the information flow from the interface to users. In this paper, the pair of keywords that reflects a user's interest is extracted from the keyword map modified by the user. Experimental results show relevant documents can be retrieved with the extracted keyword pair.

Local similarity pattern (LSP) is proposed as a new method for computing image similarity. Similarity of a pair of images is expressed in terms of similarities of the corresponding image regions, obtained by uniform partitioning of the image area. Different from the conventional methods, each region-wise similarity is computed using a different combination of image features (color, shape, and texture). In addition, a method for optimizing LSP, based on genetic algorithm, is proposed, and incorporated in the relevance feedback process, allowing the user to automatically specify LSP-based queries. LSP is evaluated on four test databases totalling over 2,000 images. Compared with six conventional methods, and SIMPLIcity, an advanced image retrieval system, LSP brings between 15% and 24% increase in the average retrieval precision. LSP, allowing comparison of different image regions using different similarity criteria, is more suited for modeling human perception of image similarity than the conventional methods.

WWW information visualization system based on the plastic clustering is introduced, which assists the user to obtain the background knowledge of the unknown topics through a series of information retrieval. The keyword-based clustering is employed for the plastic clustering, which stocks the generated clusters and reuses them for the subsequent retrieval. The keyword set extraction based on the immune network model is also described with the analysis of its network dynamics.