People can readily post their products on e-commerce sites. When users present their product on an e-commerce site, they must create a document describing the item and encouraging its purchase. However, it is not easy for beginner users to create a sentence that describes a product. For this study, the sentence describing the product is called the product description. We propose a method for automatically generating the product description based on a comparison of LSTM and GPT-2. Specifically, we examine the data structure of the product included in the existing product description. Then we use data based on these data structures as input to compare these two methods. Furthermore, we conduct two experiments to measure the benefits of our proposed method based on our proposed evaluation index.

Many types of content exist on SNSs. Sometimes authors' opinions are not properly communicated to the reader. The content might be inflammatory, known as flaming. We infer the importance of extracting passages in which the author's opinion is not communicated correctly when it is presented to the reader. This study particularly examines tweets, a popular message system of the Twitter SNS, and also specifically examines "rhetorical questions." Rhetorical questions are sometimes known as mandarin sentences. People might misunderstand them and might flame the author. We consider it important to extract rhetorical question tweets automatically and present them.

When people use online shopping, they often refer reviews which are written about the products. They can understand the product more deeply by reading the reviews. The review has a star rating that shows what other people think about the product. The star rating is not always appropriate for the evaluation of the product. There are so many reviews and it is difficult to find the review that affects the users' willingness to buy. We call such review "high-value review". Our proposed high-value review does not depend on the number of star ratings. The high-value review is that people find useful information when they read the review and they think it is a good review. In this paper, we investigate the relation between high-value reviews and their sentiment of clause based on four hypotheses. Our analyzing sentiment is three-axis which are positive/negative/neutral. Finally, we extract the characteristics of high-value reviews from the results of our investigate.

Disasters of many types have occurred in recent years, such as strong earthquakes, heavy rain, and typhoons. In such disaster situations, people often use social network services (SNS) and exchange information of all types to help each other. Especially, people exchange information using Twitter during disasters. Such tweet messages include much information that promotes people's behaviors. We designate such tweets as behavioral facilitation tweets. When psychologically unstable in the aftermath of a disaster, behavioral facilitation tweets can strongly affect people, irrespective of a message's authenticity. We regard the extraction of the behavioral facilitation tweets automatically as important. In this paper, we propose a method that extracts behavioral facilitation tweets in disaster situations. Specifically, we propose and compare three methods to extract behavioral facilitation tweets in disaster situations: rule-based, support vector machine (SVM) and long short-term memory (LSTM). Furthermore, we conducted experiments to assess the benefits of our proposed method.

Today, vast amounts of reviews are posted on the internet. Businesses must extract negative opinions of products and services from reviews to improve their products and services. Still, some issues related to automatic extraction of negative sentiment must be addressed if they intend to use the information to improve their products and services. (1) Reviews are usually long texts. Finding only negative opinions for improvement is therefore time-consuming. (2) Many studies proposed about sentiment classification using machine learning. When we use the machine learning technique, we must prepare a lot of training data of the same product and services as test data. It is high cost. As described herein, we propose a clause-level sentiment classification method using Conditional Random Field (CRF) to address the issue (1). Also, we describe experiments of sentiment classification on reviews of multiple domains for the issue (2).

There are many disasters such as big earthquakes, flood disasters, hurricanes, and typhoons. Especially in Japan, there are so many natural disasters. Nowadays, after the disaster, we help each other not only in the real world but also on the social media. However, there are many rumors on the social media. Especially, twitter spreads many rumors after the disaster, because it is easy to spread information. According to our former study, there are many behavioral facilitation tweets in the rumors tweets. We consider it is important to extract automatically such rumor behavioral facilitation tweets. In this paper, as a first step of extracting rumor behavioral facilitation tweets, we propose the method which extracts behavioral facilitation tweets from disaster tweets. Our proposed method is rule-based. We also conducted the small experiment to measure our proposed method is suitable for extracting behavioral facilitation tweets.

The user-generated recipe sites contain much useful information such as tricks and traps information. When we know such useful information, we can extend the cooking menu and have a wide range of knowledge for cooking. We call such useful information as "knack for cooking", and we propose the methods of extraction knack for cooking from user-generated recipe sites. In this paper, our proposed knack for cooking is a sentence that users do not have to know the information when they cook, but if they know it the dish becomes more delicious. We propose three types of methods which are using rule-based, SVM, and hybrid(rule-based+SVM). We also conducted our experiments to measure which methods are the suitable for extracting knack for cooking.

Social media services have become popular. Especially, Twitter is accumulating and distributing vast amounts of information for its numerous users. One feature of Twitter is that a user follows other users, who can obtain information that is tweeted from followees. However, it is difficult for a user to find promising followees because there are so many Twitter users. Therefore, numerous studies have tackled the recommendation of followees for Twitter users. Many methods recommend followees based on topics extracted from their tweets, but many people tweet on the same topic, but with very different emotions about the topic. These people are not beneficial as candidates for new followees. The system should recommend new followees who tweet the same topic while expressing similar emotions about the topics in which the user is interested. Actually, it is easy for users to find which people tweet the same or similar topics, but it is difficult to find people who have similar emotions about the same topic. Therefore, users cannot follow people who have similar emotions about the same topic. As described in this paper, we call a person who tweet same topic and similar emotions about the topic a "latent followee". We propose the new followee recommendation system that recommends latent followees based on similar topics and similar emotions. Our proposed system first extracts same topics using clustering. Next the system extracts emotion of the same topic tweets using SVM. Then the system presents to the user people who tweet the same topic and similar emotions for the topic as latent followees. We also conducted an experiment and confirmed the availability of our proposed system.

On Twitter, vast numbers of tweets have been written about news articles. These tweets include not only opinions and sentiments, but also comments related to the news articles. However, tweets that include comments about news article are believed by people even if their credibility is not clear. In this way, these tweets are sometimes spread by others. Therefore, we consider the importance of raising an alarm about tweets for which the credibility is not clear. As described in this paper, as a first step of extracting tweets with unclear credibility, we propose a method to extract tweets that include commentary about news articles. In this paper, we designate the tweets as "commentary tweets". Our proposed method consists of a rule-based component and a machine learning component. We also conducted our experiments to measure the suitability of our proposed method for extracting commentary tweets.

Microblogging services such as Twitter and Facebook become popular in recent years. In these services, many users post short messages which correspond to many topics such as daily activities, opinions, and new events. Therefore, users need a system to summarize messages if the users receive tons of messages. If the following users tweet about important things which the user does not know, these tweets should be noticed. However, which tweets should be noticed is one important problem. Users should need which topics are on their timeline. However, if the summarization method does not consider topics of tweets, the summarized tweets do not contain rarely tweeted topics. To solve this problem, we propose a method for automatically extracting missing tweets based on topic granularity and missing time of the users. In this study, we map the missing tweets to the Wikipedia category tree by considering topic structure granularity; then we present the topic structures of missing tweets using our proposed visualization interface. In our experiments, we confirmed the effectiveness of our proposed hierarchal topic structure.

Nowadays, travel-related information of many kinds can be found on the Internet. People plan their travel and obtain information about sightseeing spots from the Internet before they travel. When obtaining information related to sightseeing spots, they receive basic information from official pages easily. However, other useful information exists on user-generated travel sites. User-generated travel sites abound on the Internet, offering great amounts of diverse information related to travel and destinations. This study addresses travel information of four types related to user-generated travel sites: basic, useful-buzz, useful-unexpected, and garbage information. Useful-unexpected information benefits users, but extracting it from user-generated content is difficult because it includes so much useful-buzz information and garbage information. We designate useful-unexpected important information as "Welcome-news". As described herein, we propose a means of extracting Welcome-news from user-generated travel contents. Our proposed Welcome-news is "useful information" and "unexpected information" related to travel. We first extract useful information based on Welcome news keywords, which are general keywords and unique keywords. General keywords often appear in Welcome-news. We regard general keywords by our user experiment. Unique keywords depend on the sightseeing spot. We regard unique keywords based on SVM. Next we extract unexpected information based on clustering. Our unexpected information includes topics of unexpected information that are important topics for sightseeing spots and contents that are often not stated. Subsequently, we extract important topics based on topic-based clustering. Then we extract unexpected information contents from the cluster based on its distance from the cluster center. We conducted experiments of three types to extract correct answers, to assess the feasibility of using Welcome-news keywords, and to assess the feasibility of extracting Welcome-news.

Nowadays, many user-generated recipe sites are accessible on the internet. On user-generated recipe sites, however, are various spam recipe pages that describe closely similar recipes requiring special cooking equipment, with no preparation explanations. These spam recipes are not useful for users. In fact, they impede user's recipe searches. In this paper, we target closely similar recipes as a first step in extracting spam recipes. If user search results could be classified to identify closely similar recipes, user's recipe searches would be easier and more productive. Clustering tools of many kinds are proposed, but it is difficult to cluster closely similar recipes using only existing clustering tools because recipe sites have a unique page structure comprising a title, ingredients, directions (preparation instructions), and comments. The importance of words from each part differs. We propose a clustering method for user-generated recipe sites based on page structure and important words. Next, we conducted an experiment to measure the benefits of our proposed method. The result of experiment presents the benefits of our proposed method which classify the closely similar recipes.

Today there is much information and knowledge on the internet, and many studies have examined the extraction of many kinds of knowledge from the internet. In addition, numerous studies have examined entertainment robots that communicate with people, but it is difficult for robots to communicate smoothly with people. We specifically examine communication between robots based on dialogue. Here, we create a dialogue-based scenario for the robots to undertake automatically, but it is difficult because the dialogue requires knowledge of many kinds. We consider the use of the knowledge from the web and create scenarios automatically. As described herein, we propose a system that generates dialogue scenarios automatically from web news articles in real time. We used the Manzai metaphor, which is Japanese traditional humorous comedy in our system. Our generated Manzai scenario consists of snappy patter and a misunderstanding dialogue based on the gap of our structure of funny points. We create communication robots to amuse people with our generated humorous robot dialogue scenarios.

Twitter users who browse tweets can follow other users in whom they are interested. They can obtain interesting information from other users' tweets on their timeline. If they follow many users, then they can expect numerous tweets on their timeline. However, if users do not browse their timeline for some time, they can lose interesting and important information. Therefore, a system that automatically presents a summary of lost information can be extremely beneficial. As described herein, we propose a method of extracting lost information automatically based on a user's browsing time interval and the topic structure of a followee's tweets. First, we classify a followee's tweets that contain the user's missing information, and assign topics to the groups. Next, we generate a topic graph based on the semantic structure from Wikipedia. We decide whether the tweet groups are missed using the followee's topic graph based on the browsing time interval. Finally, we extract missing information and present it to the user.

Twitter has become a popular social media service, accumulating and distributing vast amounts of information for its numerous users. One feature of Twitter is that it enables a user to follow other users, who can obtain the information her/his followees tweeted. However, it is difficult for the user to find a promising followee because there are so many Twitter users. Therefore, numerous studies have investigated the issue of recommendation of followees for Twitter users. Many methods recommend followees based on topics extracted from their tweets. It seems beneficial to recommend followees who not only have similar interests but also similar sentiments to those of the user. We propose a system that recommends a followee based on topics and their sentiments about topics. In this paper, as a first step of our study on followee recommendation, new followee candidates are limited to people who are being followed by at least one of the user's followees but who the user is not following. In this paper, we refer to the target persons as "ff-users." (1) For each ff-user, our proposed system uses clustering to extract common topics between the user and ff-user, and then it extracts the sentiments of their tweets about each of the common topics. (2) As a result, those ff-users who have a larger number of similar sentiments related to a larger number of common topics are recommended as new followees. (3) The system visualizes user sentiments and those of candidate followees using a radar chart. We also conducted an experiment, and confirmed the validity of our proposed system.

People post their impressions of foods on Twitter in real time after eating. On many user-generated recipe sites on the Internet, ordinary cooks such as homemakers can post their original recipes. They make full use of the recipe titles that incorporate tasty words such as authentic, homey, and spicy because they want to their recipe to become popular. Web sites of companies and restaurants use tasty words such as 'healthy' and 'old-fashioned' to sell their products. In this way, tasty words of many kinds have come to be used on the Internet. We consider that tasty words differ among those used on Internet media which are Twitter, user-generated recipe sites, and ordinary web sites. As described herein, we designate such tasty words as 'Sizzle Words' and then compare these three media using the Sizzle Words.

Microblogging systems such as Twitter and Facebook have become popular. People can easily post their sentiments to the Internet in real time using such microblogging systems. Twitter is a text-based communication tool. Users cannot use non-verbal communication tools such as gestures and eye contact on Twitter. Users sometimes use emoticons as an alternative non-verbal communication tool to tweet delicate sentiments. In this paper, we propose a method of determining sentiment of a tweet based on the emoticon role. Specifically, we propose the following: (1) compilation of a sentiment lexicon and an emoticon lexicon; (2) emoticon roles can be classified into four types showing "Emphasis", "Assuagement", "Conversion", and "Addition", with roles determined based on a relation between sentiments of sentences and emoticons; and (3) the relation can be formalized using regression analysis in all roles excepting for "Addition".

Microblogging systems such as Twitter have become popular. They are especially useful and helpful for users in disaster situations. Microblogs have facilitated the spread of information of all kinds, even rumors. Rumors block adequate information sharing and cause severe problems. Several studies have analyzed rumors, but it remains unclear how rumors are spread on microblogging systems. As described in this paper, we present a case study of how rumors are spread on Twitter in a recent disaster situation, that of the Great East Japan earthquake in March 11 2011, based on comparison to a normal situation. We also specifically examine the correction of rumors because automatic extraction of rumors is difficult, but extracting rumor-correction is easier than extracting the rumors themselves. We (1) classify tweets in disaster situations, (2) analyze tweets in disaster situations based on user's impression, and (3) compare the spread of rumor tweets in a disaster situation to that in a normal situation.

Wikipedia has multilingual articles, the information of which differs, even for articles on the same topic. As described in this paper, we propose a system to extract and present lack of information of one language on Wikipedia by comparing two languages on the Wikipedia. When we compare Wikipedia articles of two languages, the granularity of information between them differs. Therefore, we propose a method of extracting multiple comparison articles using a Wikipedia link graph. The system extracts lack of information that is included in articles in Wikipedia by comparing one base article with other articles that are found using the link graph.

Using social media, users post and exchange information related to personal behavior, experimentation, and personal sentiment. Sometimes this information is not written in ordinary web pages. This information is important for users who are people of the community and for people outside of the community. Nevertheless, it is difficult to extract important information from social media because such services include so much information. Moreover, the information quality differs. We designate such important and unique information related to social media as "tip information". As described in this paper, we propose a method to extract credible and important tip information from SNSs as a first step in extracting tip information from social media. Then we propose a means to extract tip information from SNS, and propose a means of ranking the tip information.

With the rapid advance of the Internet, everybody has become able to obtain information from it easily. However. there are no systems which are available to extract and present information suitable to a user's sentiment. We propose a system that searches for information based on a user's sentiment. As described in this this paper, we propose a words-of-wisdom search system as a first step of the research. Specifically, we first propose a multi-dimensional sentiment vector based on Nakamura's proposed 10 categories of sentiments. Next, based on our experiment, we calculate the value of sentiment words included in words-of-wisdom. Subsequently we calculate the sentiment value of words-of-wisdom using a value of sentiment words. We developed a prototype system and conducted experiments.

Social Network Services(SNSs), which are maintained by a community of people, are among the popular Web 2.0 tools. Multiple users freely post their comments to an SNS thread. It is difficult to understand the gist of the comments because the dialog in an SNS thread is complicated. In this paper, we propose a system that presents the gist of information at a glance and basic information about an SNS thread by using Wikipedia. We focus on the table of contents (TOC) of the relevant articles on Wikipedia. Our system compares the comments in a thread with the information in the TOC and identifies contents that are similar. We consider the similar contents in the TOC as the gist of the thread and paragraphs in Wikipedia similar to the comments in the thread as comprising basic information about the thread. Thus, a user can obtain the gist of an SNS thread by viewing a table with similar contents.

SNSs and blogs, both of which are maintained by a community of people, have become popular in Web 2.0. We call these content as "Community-type content." This community is associated with the content, and those who use or contribute to community-type content are considered as members of the community. Occasionally, the members of a community do not understand the theme of the content from multiple viewpoints, hence, the amount of information is often insufficient. It is convenient to present the user missed information. In this way, when Web 2.0 became popular, the content on the Internet and type of users are changed. We believe that there is a need for next-generation search engines in Web 2.0. We require a search engine that can search for information users are unaware of; we call such information as "content holes." In this paper, we propose a method for searching content holes in community-type content. We attempt to extract and represent content holes from discussions on SNSs and blogs. Conventional Web search technique is generally based on similarities. On the other hand, our content-hole search is a different search. In this paper, we classify and represent a number of images for different searching methods; we define content holes and as the first step toward realizing our aim, we propose a content-hole search system using Wikipedia.

In community-type content such as blogs and SNSs, we call the user's unawareness of information as a "content hole" and the search for this information as a "content hole search." A content hole search differs from similarity searching and has a variety of types. In this paper, we propose different types of content holes and define each type. We also propose an analysis of dialogue related to community-type content and introduce content hole search by using Wikipedia as an example.

We propose a system that automatically transforms web content into TV-like video content for ubiquitous environments. We call this system the ubiquitous/universal passive viewer (u-PaV). The u-PaV consists mainly of audio and visual components. The audio component uses synthesized speech to read out titles and lines extracted from the target web content. Simultaneously, the visual component of the u- PaV presents the titles and lines to a user display through a ticker. Keywords and images extracted from the web content are animated on the display. A suitable background color is determined based on the overall impression of the content. The u-PaV synchronizes the ticker, animation, and speech. We introduce the u-PaV and explain how the keywords are extracted and how the impression value of the web content is determined. A test with 50 users showed that the u-PaV is easier to use and understand than browsing web content alone.

We propose a new way of browsing bilingual Web sites through concurrent browsing with automatic similar-content synchronization and difference-detection facilities. Our prototype browser system is called the bilingual comparative Web browser (B-CWB) and it concurrently presents bilingual web pages in a way that enables their content of the Web pages to be automatically synchronized. The B-CWB allows users to browse two Web news sites concurrently and compare the similar news articles written in different languages (English and Japanese). The major characteristics of the B-CWB are its content synchronization and difference detection: content synchronization means that user operation (scrolling or clicking) on one Web page automatically invokes not necessarily same operations on the other Web page to preserve similarity of content between the two Web pages. For example, scrolling a Web page may involve passage-level similarity retrieval on the other Web page. Clicking a Web page (and obtaining a new Web page) invokes page-level similarity retrieval within the other Web site pages through the use of a English-Japanese dictionary. Difference detection means that the B-CWB analyzes two similar Web pages shown concurrently to discover the several "differences" between them. This facility is important in comparing two news articles that report the same affairs.

In this paper, we propose a new type of Web browser, called the Comparative Web Browser(CWB), which concurrently presents multiple Web pages in a way that enables the content of the Web pages to be automatically synchronized. The ability to view multiple Web pages at one time is useful when we wish to make a comparison on the Web, such as when we compare similar products or news articles from different newspapers. The CWB is characterized by (1) automatic content-based retrieval of passages from another Web page based on a passage of the Web page the user is reading, and (2) automatic transformation of a user's behavior (scrolling, clicking, or moving backward or forward) on a Web page into a series of behaviors on the other Web pages. The CWB tries to concurrently present "similar" passages from different Web pages, and for this purpose our CWB automatically navigates Web pages that contain passages similar to those of the initial Web page. Furthermore, we propose an enhancement to the CWB, which enables it to use linkage information to find related documents based on link structure.

In the present paper, we propose a new way of organizing Web search results and of viewing those results passively in the mobile environment which has limited display and limited interaction. Specifically, the system makes Carousel Components, that are composed of image and voice, from Web search result. Each time of user's interaction, the system automatically computes sets of similar, different, more-detailed, and more-abstracted pages, respectively and reorganizes them as carousels. We call this system WebCarousel.