Yohei Ikawa

Location inference using microblog messages

In order to sense and analyze disaster information from social media, microblogs as sources of social data have recently attracted attention. In this paper, we attempt to discover geolocation information from microblog messages to assess disasters. Since microblog services are more timely compared to other social media, understanding the geolocation information of each microblog message is useful for quickly responding to a sudden disasters. Some microblog services provide a function for adding geolocation information to messages from mobile device equipped with GPS detectors. However, few users use this function, so most messages do not have geolocation information. Therefore, we attempt to discover the location where a message was generated by using its textual content. The proposed method learns associations between a location and its relevant keywords from past messages, and guesses where a new message came from.

Online User Location Inference Exploiting Spatiotemporal Correlations in Social Streams

The location profiles of social media users are valuable for various applications, such as marketing and real-world analysis. As most users do not disclose their home locations, the problem of inferring home locations has been well studied in recent years. In fact, most existing methods perform batch inference using static (i.e., pre-stored) social media contents. However, social media contents are generated and delivered in real-time as social streams. In this situation, it is important to continuously update current inference results based on the newly arriving contents to improve the results over time. Moreover, it is effective for location inference to use the spatiotemporal correlation between contents and locations. The main idea of this paper is that we can infer the locations of users who simultaneously post about a local event (e.g., earthquakes). Hence, in this paper, we propose an online location inference method over social streams that exploits the spatiotemporal correlation, achieving 1) continuous updates with low computational and storage costs, and 2) better inference accuracy than that of existing methods. The experimental results using a Twitter dataset show that our method reduces the inference error to less than 68% of existing methods. The results also show that the proposed method can update inference results in constant time regardless of the amount of accumulated contents.

Location-based insights from the social web

Citizens, news reporters, relief organizations, and governments are increasingly relying on the Social Web to report on and respond to disasters as they occur. The capability to rapidly react to important events, which can be identified from high-volume streams even when the sources are unknown, still requires precise localization of the events and verification of the reports. In this paper, we propose a framework for classifying location elements and a method for their extraction from Social Web data. We describe the framework in the context of existing Social Web systems used for disaster management. We present a new location-inferencing architecture and evaluate its performance with a data set from a real-world disaster.

Language Translation Tools Drive Productivity Improvements for Global Delivery of Services

Efficient utilization of global resources is crucial for globally integrated enterprises to improve their business performance. Rising costs of language translation by professional translators are inhibiting the growth of global delivery of services from and to countries and geographies where most documents need to be translated. The quality levels of current machine translation engines are inadequate for such business processes, since translation errors may cause significant business problems. Human-quality translations are needed, and we need tools for enhancing these human translation efforts. We have developed an architecture and some tools to help people create translations more efficiently. The framework provides tools for interactively creating translations by aggregating the information necessary for translating the documents, for checking the quality of the source documents to be translated, and for effectively gathering translation-related resources from existing documents. We conducted several experiments to evaluate the performance of the tools, and confirmed that the framework and tools are valuable in assisting human translation work.

A New Document Masking Approach for Removing Confidential Information

In order to protect confidential information such as personal and organizational information written as text, document masking techniques are becoming important. Such document masking methods extract humans, places, and organization names automatically and remove them, so they make documents harmless and allow sharing them safely within an organization, and contribute to improving productivity. However, existing automatic document masking techniques are not reliable enough since they may fail to mask out-of-vocabulary proper nouns. In this paper we propose a novel technique for document masking, the Unmasking Method, in which all of the words are hidden initially and a human specifies the non-confidential words to be unmasked. The proposed method is a high-safety document masking method since it unmasks only words that a human has manually recognized as safe. Our experimental results show its safety and effectiveness.

Detecting invalid dictionary entries for biomedical text mining

In text mining, to calculate precise keyword frequency distributions in a particular document collection, we need to map different keywords that denote the same entity to a canonical form. In the life science domain, we can construct a large dictionary that contains the canonical forms and their variants based on the information from external resources and use this dictionary for the term aggregation. However, in this automatically generated dictionary, there are many invalid entries that have negative effects on the calculations of keyword frequencies. In this paper, we propose and test methods to detect invalid entries in the dictionary.