Abstract During a disaster, social media can be both a source of help and of danger: Social media has a potential to diffuse rumors, and officials involved in disaster mitigation must react quickly to the spread of rumor on social media. In this paper, we investigate how topic diversity (i.e., homogeneity of opinions in a topic) depends on the truthfulness of a topic (whether it is a rumor or a non-rumor) and how the topic diversity changes in time after a disaster. To do so, we develop a method for quantifying the topic diversity of the tweet data based on text content. The proposed method is based on clustering a tweet graph using Data polishing that automatically determines the number of subtopics. We perform a case study of tweets posted after the East Japan Great Earthquake on March 11, 2011. We find that rumor topics exhibit more homogeneity of opinions in a topic during diffusion than non-rumor topics. Furthermore, we evaluate the performance of our method and demonstrate its improvement on the runtime for data processing over existing methods.

Tree data analysis has many applications in information security. In particular, HTML pages' DOM trees are an important target of analysis because web pages can be vectors for, and targets of, major cyberattacks like phishing. Previous attempts to incorporate tree data analysis into security applications, however, have been hampered by the lack of efficient methods for tree data analysis in machine learning. As such, most security research has focused on data representable as vectors of real numbers, like most machine learning work. Recent work, however, has yielded several efficiency break-throughs in tree analysis. One example is kernel methods, a methodological bridge that fills the gap between discretely-structured data (like trees) and multivariate analysis. Kernel methods enable applying a variety of multivariate analysis techniques such as SVM and PCA to trees. The method we are interested in is the subpath kernel. The subpath kernel offers the following advantages: (1) it is invariant over ordered and unordered trees; (2) it can be computed using an extremely fast linear-time algorithm compared to the quadratic time required to compute values of most tree kernels; (3) its excellent prediction accuracy has been proven through intensive experiments. This paper proposes a subpath kernel-based method for tree-structured security data. To demonstrate the effectiveness of our method, we apply it to the problem of detecting fake e-commerce sites, a sub-problem of phishing detection with a significant real-world financial cost. In an experiment on a real dataset of fake sites provided by a major e-commerce company, our method exhibited accuracy as high as 0.998 when training SVM with as few as 1,000 instances. Its generalization efficiency is also excellent: with only 100 training instances, the accuracy score reaches 0.996. While previous phishing detection methods relied on textual content, URL components, and blacklists, our approach is the first to leverage DOM trees, which makes it both more effective and more robust against adversarial attacks. Unlike URL or content changes, changing a page's DOM structure incurs large costs to criminals.

This paper presents FITS, or Feature-value / Instance Transposition Selection, a method for unsupervised clustering. FITS is a tractable, explicable clustering method, which leverages the unsupervised feature value selection algorithm known as UFVS in the literature. FITS combines repeated rounds of UFVS with alternating steps of matrix transposition to produce a set of homogenous clusters that describe data well. By repeatedly swapping the role of feature and instance and applying the same selection process to them, FITS leverages UFVS's speed and can perform clustering in our experiments in tens milliseconds for datasets of thousands of features and thousands of instances.We performed feature selection-based clustering on two real-world data sets. One is aimed at topic extraction from Twitter data, and the other is aimed at gaining awareness of energy conservation from time-series power consumption data. This study also proposes a novel method based on iterative feature extraction and transposition. The effectiveness of this method is shown in an application of Twitter data analysis. On the other hand, a more straightforward use of feature selection is adopted in the application of time series power consumption data analysis.

In this paper, we proposed a method for determining whether a given restaurant review comment is a repeater's review, or not. We often use restaurant review sites to decide which restaurant to go to. When we read a restaurant review comment, we can know whether the reviewer is a repeater of the restaurant. If a certain restaurant has many repeaters, the restaurant must be great. However, restaurant review sites usually do not provide a "revisit rate". Therefore, we tackle a problem for determining whether a review is a repeater's review, or not. There are many sentences in a review comment that are completely not useful for determining whether the review is a repeater review, such as what was ordered, what was delicious, or how was the price. To confront such difficulties, we have taken the following approach. First, very various features are extracted from review comments so as not to miss the features that represent repeaters' reviews. Next, from the very various features, only the necessary features that really contribute to the classification is selected by a feature selection method. Finally, classification is performed using a classifier. We have implemented the proposed method using super-CWC [12], a state-of-the-art feature selection method, and SVM. The experimental results show that the proposed method is better than other methods.

In this paper, given the subject-verb pairs, we propose a computational model to express the difference of the meaning of a verb when the subject has changed. We propose a method to generate metaphorical expressions consist of subject-verb pairs from the model. "Airship swims" is one of the example. It is the expression about the event that an airship flies in the sky gracefully. There are a few reasons why the expression is accepted for people. "airship flie" and the motion of a sea creature, for example "whale swims" represent both "the normal move in a space" and there is a similarity. Given the input ("airship," "fly"), we propose a method to detect a verb "swims" to generate metaphorical expressions considering these similarity. At first, we test which vectorization method is the best as the vectorization of a subject-verb pair. We calculate a transformation matrix to conserve between the meaning of (non-human subject, verb) pairs and the meaning of ("man", verb) pairs. We calcurate the transformation matrix between them using the stable meaning verbs as the anchors. In this paper, we test an hypothesis that we can use these transformation matrices to find an appropriate verb considering the difference of the meaning occured from the subjects. We gather 67 cases of target figurative expressions from Web. We evaluated the proposed method by defining the information retrieval problem of verbs.

This paper is showing a time series topic life cycle extraction from millions of Tweets using our original community detection technique in bipartite networks. We suppose that the authors role that means who belong to what topics is important to extract quality topics from social media data. We already proposed the topic extraction method that considers the relationship between the authors and the words as bipartite networks and explores the authors role by forming clusters as topics. As the next step, this paper applies our method to the time series topic life cycle detection. We extract topics in different time slots and analyze the time series of topic transition using the coherence measure that expresses the semantic accuracy of topics. The paper demonstrates that our method can detect the topic life cycle such as the growth, the conflicts and so on over time from millions of Tweets.

Social media offers a wealth of insight into how significant topics such as the Great East Japan Earthquake, the Arab Spring, and the Boston Bombing affect individuals. The scale of available data, however, can be intimidating: during the Great East Japan Earthquake, over 8 million tweets were sent each day from Japan alone. Conventional word vector-based topic-detection techniques for social media that use Latent Semantic Analysis, Latent Dirichlet Allocation, or graph community detection often cannot extract appropriate topics from such a large volume of data with accuracy due to their space and time complexity. To alleviate this problem, we propose an effective topic extraction from millions of tweets based on community detection in bipartite networks. Our method is based on the bipartite community detection technique developed by Okamoto, one of the authors of this paper. The paper demonstrates our method effectiveness on social media analysis and identifies topics from millions of tweets after the Great East Japan Earthquake. To show our method's effectiveness, we compute the coherence measure that can evaluate the semantic accuracy and the running time, and compare the method with LDA that is the major topic model.

Feature selection is a useful tool for identifying which features, or attributes, of a dataset cause or explain phenomena, and improving the efficiency and accuracy of learning algorithms for discovering such phenomena. Consequently, feature selection has been studied intensively in machine learning research. However, advanced feature selection algorithms that can avoid redundant selection of features and can detect interacting features require heavy computation in general and hence are seldom used for big data analysis. To eliminate this limitation, we tried to improve the run-time performance of two of the most advanced feature selection algorithms known in the literature. We have developed two accurate and extremely fast algorithms, namely Super CWC and Super LCC. In experiments with multiple real datasets which are actually studied in big data research, we have demonstrated that our algorithms improve the performance of their original algorithms remarkably. For example, for two datasets, one with 15,568 instances and 15,741 features and another with 200,569 instances and 99,672 features, Super-CWC performed feature selection in 1.4 seconds and in 405 seconds, respectively. This is a remarkable improvement, because it is estimated that the original algorithms would need several hours to a few ten days to perform feature selection on the same datasets.

Haussler's convolution kernel provides a successful framework for engineering new positive semidefinite kernels, and has been applied to a wide range of data types and applications. In the framework, each data object represents a finite set of finer grained components. Then, Haussler's convolution kernel takes a pair of data objects as input, and returns the sum of the return values of the predetermined primitive positive semidefinite kernel calculated for all the possible pairs of the components of the input data objects. On the other hand, the mapping kernel that we introduce in this paper is a natural generalization of Haussler's convolution kernel, in that the input to the primitive kernel moves over a predetermined subset rather than the entire cross product. Although we have plural instances of the mapping kernel in the literature, their positive semidefiniteness was investigated in case-by-case manners, and worse yet, was sometimes incorrectly concluded. In fact, there exists a simple and easily checkable necessary and sufficient condition, which is generic in the sense that it enables us to investigate the positive semidefiniteness of an arbitrary instance of the mapping kernel. This is the first paper that presents and proves the validity of the condition. In addition, we introduce two important instances of the mapping kernel, which we refer to as the size-of-index-structure-distribution kernel and the editcost-distribution kernel. Both of them are naturally derived from well known (dis)similarity measurements in the literature (e.g. the maximum agreement tree, the edit distance), and are reasonably expected to improve the performance of the existing measures by evaluating their distributional features rather than their peak (maximum/minimum) features.